The Confluence of Deep Technologies and Advanced Mathematics for Causal Discovery in Complex Adaptive Systems

Co-created by the Catalyzer Think Tank divergent thinking and Gemini Deep Research tool.

 

1. Executive Summary: The New Frontier of Causal Discovery in Complex Adaptive Systems

The quest to understand and model Complex Adaptive Systems (CAS) stands as one of the most significant scientific challenges of our time. These systems, characterized by emergent behavior, non-linearity, feedback loops, and adaptation across multiple scales, permeate diverse domains from molecular biology and climate science to socio-economic structures and engineered networks. Traditional analytical methods often fall short in unraveling the intricate causal webs within CAS. This report details a paradigm shift, driven by the confluence of advanced Artificial Intelligence (AI) and Machine Learning (ML) methodologies with sophisticated mathematical concepts, aimed at discovering fundamental causal mechanisms—often termed “objective first fundamentals.” The ambition is to move beyond mere correlation to a robust, interpretable, and generalizable understanding of why these systems behave as they do, by leveraging the “omnipresence of signals” across matter, molecular, and observable scales.

Key technologies at the forefront of this endeavor include Large Language Models (LLMs), Transformers, Geometric Deep Learning (encompassing Hyperbolic Graph Neural Networks – HGNNs and concepts like signal geometric diffusion), Neuro-Symbolic AI, Physics-Informed Neural Networks (PINNs), and dedicated Causal AI frameworks. These are increasingly being augmented by advanced mathematical ideas from topology (homology) and non-Euclidean geometry (hyperbolic 3-manifolds, isometry), seeking to capture deeper structural and relational properties of complex data.

Current research spans a spectrum from theoretical explorations to applied tool development. While LLMs are being applied for hypothesis generation 1 and Causal AI toolkits are finding practical use 3, the integration of advanced geometric and topological concepts for causal discovery, and the overarching pursuit of “objective first fundamentals,” largely reside in theoretical or basic research phases. The global research landscape is dynamic, with significant investments and contributions from corporate laboratories like Google, Microsoft, and Baidu, which are pushing the applied frontiers. Government laboratories and national initiatives, particularly in the USA, China, and Europe, are fostering foundational work and strategic roadmaps.5 Universities worldwide remain the primary drivers of theoretical breakthroughs in these deep tech domains.

A parallel and potentially synergistic deep tech frontier is quantum computing and quantum sensing. These fields, involving many of the same institutional players, promise novel computational paradigms and unprecedented data acquisition capabilities that could, in the long term, revolutionize scientific discovery and the modeling of CAS.8

The central challenge remains the formidable task of inferring robust, interpretable, and generalizable causal relationships from often observational, high-dimensional, and noisy data generated by CAS. The opportunity, however, is immense: to forge a new scientific methodology where deep technologies, grounded in mathematical rigor, enable the discovery of the fundamental principles governing the complex world around us. This report posits that the true transformative power will emerge not from any single technology, but from their synergistic integration, creating a holistic approach to unraveling causality in complex adaptive systems. The current landscape reveals specialized research tracks, and the next significant leap will likely involve the convergence of these diverse methodologies to address the multifaceted nature of causality in multi-scale, dynamically evolving systems.

2. Foundational Concepts: Causality, Complexity, and First Fundamentals

The endeavor to uncover causal relationships that form objective first fundamentals within complex adaptive systems necessitates a clear understanding of these core concepts. This section lays the groundwork by defining causality in the context of CAS, exploring the pursuit of “objective first fundamentals” through AI, and considering the nature of signals across various scales that serve as the basis for discovery.

2.1. Defining Causal Relationships in Complex Adaptive Systems (CAS)

Complex Adaptive Systems (CAS) are ubiquitous, spanning natural, social, and artificial domains. They are characterized by several key properties that make them distinct from simple, linear systems and render causal discovery particularly challenging. These properties include:

• Emergence: Macro-level patterns and behaviors arise from the interactions of micro-level components, which are not easily predicted from the properties of the individual components alone.11
• Non-linearity: Cause-and-effect relationships are often not proportional; small changes in one part of the system can lead to large, unpredictable changes elsewhere.13
• Feedback Loops: The outputs of a system can influence its inputs, creating complex cycles of interaction that can be both stabilizing (negative feedback) and destabilizing or amplifying (positive feedback).13
• Adaptation: Agents or components within a CAS can learn and change their behavior based on experience and interactions with their environment and other agents.16
• Hierarchy: CAS often exhibit multiple levels of organization, with nested structures and interactions occurring across different scales.11

These characteristics mean that traditional statistical methods, which are often adept at identifying correlations, struggle to disentangle true cause-and-effect relationships from spurious associations, especially when dealing with observational data.1 The presence of unobserved confounders, complex interaction effects, and dynamic, evolving relationships necessitates more sophisticated approaches.21 A critical distinction in causal inference is between observational and interventional data. While randomized controlled trials (RCTs) are the gold standard for establishing causality by directly manipulating variables (interventions), they are often infeasible, unethical, or impractical in many CAS settings, such as large-scale ecological or socio-economic systems.23 Consequently, much of the data available from CAS is observational. A significant challenge, therefore, is the “observational bottleneck”: how to infer causal effects reliably from data where direct experimentation is limited. AI and ML methods offer a promising avenue by learning models from observational data that can then be used to simulate the potential outcomes of hypothetical interventions, thereby providing insights into causal mechanisms without direct manipulation of the system.25 This capability to create “virtual laboratories” or “digital twins” is a key contribution of deep tech to causal discovery in CAS.

2.2. The Pursuit of “Objective First Fundamentals” through AI

The term “first fundamentals” or “first principles” in this context refers to the underlying laws, generative mechanisms, or irreducible causal relationships that govern the behavior and evolution of a CAS. This is a departure from purely phenomenological models that describe what happens, towards models that explain why it happens, based on core truths.

The pursuit of “objectivity” in this endeavor is laden with philosophical and practical challenges. Data-driven AI models are themselves susceptible to biases present in the data they are trained on, or biases introduced by the model architecture and learning algorithms.27 However, AI can also be a tool to mitigate certain human cognitive biases by systematically analyzing vast datasets and complex patterns that are beyond human capacity. The development of “Responsible AI” frameworks, emphasizing fairness, transparency, accountability, and robustness, is crucial in this regard.29 Causal AI, by its nature, aims to improve reliability and explainability.20

A critical consideration is whether AI can move beyond pattern recognition and prediction to offer something akin to mathematical proof or rigorous derivation of these first fundamentals. Early indications suggest potential in this direction, particularly with neuro-symbolic systems that can execute logical programs 33 or AI models designed for mathematical reasoning, such as Google DeepMind’s AlphaGeometry, which combines neural language models with symbolic deduction engines.34 The development of formal languages like StaCL (Statistical Causality Language) also points towards a future where causal claims can be expressed and verified with greater rigor.35

In the context of CAS, “first fundamentals” may not always be pre-existing, universal physical laws in the traditional sense. Instead, they might be the emergent set of core causal rules and interaction topologies that, once identified and modeled (e.g., in a causal graph), can accurately reproduce and predict the system’s complex, adaptive, and emergent behaviors. The “objectivity” of such fundamentals would derive from the model’s predictive power across diverse conditions, its generalizability, and its parsimony (aligning with Occam’s Razor). The search for first fundamentals in CAS using AI should therefore focus not just on fitting data but on identifying the simplest, most generalizable causal graph that explains the system’s behavior.

2.3. The “Omnipresence of Signals”: Data Sources for Causal Discovery Across Scales

The “omnipresence of signals” underscores the rich tapestry of data available for AI-driven causal discovery, spanning multiple scales of organization:

• Matter Scale: This encompasses signals from fundamental physical phenomena. Quantum sensors, for instance, are being developed to measure gravitational fields, magnetic anomalies, and atomic-level interactions with unprecedented precision.8 This includes technologies like atom interferometry for inertial sensing and gravimetry 37, quantum magnetometers 39, and techniques exploiting quantum interference and entanglement.36 Such sensors could provide novel data streams for understanding Earth systems, material properties, and even fundamental physics.40 Microsoft’s Metasurface research also points to new ways of sensing and manipulating electromagnetic signals.42
• Molecular Scale: This scale is rich with data from genomics, proteomics, metabolomics, and chemical process monitoring. AI is heavily utilized in biomedical research for analyzing this data to understand disease mechanisms, identify biomarkers, and accelerate drug discovery.43 Google’s AlphaFold, for example, has revolutionized protein structure prediction 45, and tools like GNoME are discovering new materials by predicting molecular stability.46 Platforms like NVIDIA’s BioNeMo 47 and Baidu’s PaddleHelix 48 are specifically designed for AI-driven molecular research, including drug-target affinity and vaccine design. The van Dijk Lab at Yale is pioneering “Cell2Sentence” to apply LLMs to transcriptomics.50
• Observable Scale (Human & Tool-Generated): This broad category includes data from a vast array of human activities and environmental monitoring systems. Examples include sensor networks (e.g., Google’s ionosphere mapping with smartphones 52), IoT data, economic indicators, social network interactions, climate data from satellites and ground stations 53, astronomical observations, and experimental data from diverse scientific instruments. AI models are used for flood forecasting 55, wildfire tracking 53, analyzing ecological time series 56, and understanding acoustic phenomena.57

These signals manifest in various data modalities, including time series, images, text (scientific literature, reports), graph-structured data (networks of interactions), and increasingly, multimodal datasets that combine these types.58 The challenge and opportunity lie in developing AI methods that can effectively process and integrate these diverse signals to build comprehensive causal models.

A significant hurdle is the “Signal Integration Challenge.” While CAS are inherently multi-scale and multimodal, current deep tech approaches often specialize. True understanding of “objective first fundamentals” will likely require bridging causal links across these scales and modalities—for example, connecting quantum-level material properties to macroscopic system behavior or molecular interactions to population-level health outcomes. This necessitates the development of “meta-deep-tech” frameworks capable of ingesting, aligning, and co-analyzing signals from radically different sources to construct unified causal models of CAS, moving beyond current multimodal AI which typically integrates closer modalities like vision and language.60

3. Deep Tech Arsenal for Unveiling Causal Structures

A diverse array of sophisticated AI and mathematical methodologies is being developed and applied to the challenge of causal discovery in CAS. These tools, ranging from advanced neural network architectures to novel geometric and symbolic approaches, form the “deep tech arsenal” aimed at moving beyond correlational analysis to a more fundamental understanding of cause and effect.

3.1. Large Language Models & Transformers: Beyond Correlation to Causation?

Large Language Models (LLMs) and the underlying Transformer architecture have demonstrated remarkable capabilities in processing and generating human language, and their potential is now being explored for scientific discovery, including causal inference.

3.1.1. Multi-Head Attention: Mechanism and Role in Contextual Understanding for Causal Clues

The multi-head attention mechanism is a core component of the Transformer architecture.59 It allows the model to jointly attend to information from different representation subspaces at different positions. Essentially, instead of calculating a single attention function, multi-head attention runs the attention mechanism multiple times in parallel (each “head” with its own learned linear projections of queries, keys, and values) and then concatenates the results, projecting them again to produce the final output.61 This enables the model to capture various aspects of the input sequence simultaneously. For instance, one head might focus on syntactic relationships, while another captures semantic similarities or long-range dependencies within the text.59

In the context of causal discovery, this deep contextual understanding is crucial. When LLMs process scientific literature or experimental reports, multi-head attention can help identify nuanced statements that suggest causal links, distinguish between direct and indirect influences, or weigh the importance of different experimental conditions described in the text. For time-series data, it can model complex temporal dependencies that might be indicative of causal precedence.58 The interpretation of attention weights as a form of message passing on a fully connected graph, where tokens are nodes and attention scores are weighted edges, provides a conceptual link to graph-based causal models.59

3.1.2. LLMs in Hypothesis Generation, Knowledge Integration, and Causal Graph Refinement

LLMs are increasingly being explored not as standalone causal discoverers but as powerful assistants in the causal inference pipeline:

• Hypothesis Generation: Trained on vast corpora of scientific text and data, LLMs can act as “causal hypothesis engines”.1 They can identify patterns, anomalies, or gaps in existing knowledge that suggest novel causal relationships worthy of investigation. Google’s AI Co-Scientist, built on Gemini, is designed to assist researchers in generating such novel, evidence-backed hypotheses.2 The Auto-Bench benchmark explicitly tests LLMs’ ability to formulate hypotheses in the form of adjacency matrices for causal graphs.62
• Knowledge Integration: LLMs can bridge the gap between statistical causal discovery methods and domain expertise. They can process outputs from statistical algorithms (e.g., a preliminary causal graph) and evaluate the plausibility of suggested relationships based on their embedded knowledge from scientific literature.1 This allows for the refinement of causal structures by incorporating vast amounts of domain-specific information that might be difficult for a human expert to recall or for a statistical model to access directly.
• Causal Structure Learning from Text: There is active research in using LLMs to directly extract causal statements or even complete causal graph structures from unstructured text, such as scientific papers or medical reports.1 This involves training LLMs to recognize linguistic cues indicative of causality.

While promising, LLMs are currently viewed as “imperfect expert systems”.1 Their ability to perform de novo causal discovery in complex systems is limited, and performance can degrade significantly as the complexity of the causal graph increases.62 Their strength lies in augmenting human researchers and other AI tools by rapidly processing information, suggesting avenues for exploration, and integrating diverse knowledge sources. The future likely involves hybrid systems where LLMs collaborate with symbolic reasoning engines and formal causal inference algorithms. This research is primarily in basic and applied stages.

3.1.3. Transformers for Time Series and Sequential Data in Causal Discovery

The Transformer architecture’s proficiency in capturing long-range dependencies and contextual information makes it a strong candidate for analyzing sequential data, particularly time series, to uncover causal relationships where temporal precedence is a key indicator.

• Modeling Temporal Dependencies: Models like CausalFormer utilize causality-aware transformers with mechanisms like multi-kernel causal convolution and multi-variate causal attention to learn causal representations from time series data, explicitly respecting temporal priority (causes precede effects).58 This allows for the discovery of both lagged and instantaneous causality.
• Discovering Governing Equations: An emerging area is the application of Transformers to learn the underlying dynamical laws or governing differential equations directly from observed system behavior.66 For example, the DISCO model uses a transformer hypernetwork to generate parameters for a smaller neural-ODE-like solver, enabling prediction of future states in physical systems.66 MixFunn proposes a novel neural network architecture, potentially combined with transformer principles for sequence processing, to solve differential equations with enhanced precision and interpretability in a physics-informed setting.67

Research in this sub-area is largely basic, with some early applied explorations showing promise for understanding complex physical and biological systems from their temporal signal outputs.

3.2. Geometric Deep Learning & Advanced Mathematical Frameworks

Geometric Deep Learning (GDL) seeks to develop neural networks that can operate on non-Euclidean data such as graphs and manifolds, leveraging principles from geometry and topology. This is particularly relevant for CAS, whose states and interactions often have inherent geometric or topological structures.

3.2.1. Hyperbolic Graph Neural Networks (HGNNs): Modeling Hierarchies and Complex Interactions in CAS

• Theoretical Basis: Euclidean space is often suboptimal for representing data with underlying hierarchical or scale-free structures, common in CAS like social networks, biological pathways, or knowledge graphs. Hyperbolic geometry, with its property of negative curvature and exponential volume growth, allows for low-distortion embeddings of tree-like and hierarchical structures into low-dimensional spaces.68 This means that the intrinsic hierarchical relationships in the data can be more faithfully preserved.
• Applications: HGNNs are applied to learn representations of complex networks, identify influential nodes, model interactions in systems with inherent hierarchies (e.g., taxonomies, organizational structures), and potentially understand information flow or influence propagation within CAS.68
• Causal Implications: The geometry of hyperbolic embeddings might naturally encode causal precedence or influence strength in hierarchical CAS. For instance, nodes higher in a hierarchy (closer to the origin in some hyperbolic models) could represent more fundamental causes, while distances along geodesics could reflect causal path lengths or strengths. However, explicitly linking hyperbolic geometry to causal semantics is an active research area.
• Current Research Stage: This field is largely in the theoretical and basic research phase, with some domain-specific applications emerging in areas like recommendation systems and environmental claim detection.68 The mathematical complexity of hyperbolic geometry presents challenges for broader adoption.68

3.2.2. Signal Geometric Diffusion: Tracing Causal Pathways in Dynamic Systems

• Theoretical Basis: This approach, exemplified by the SHREC algorithm, combines concepts from skew-product dynamical systems theory and topological data analysis.56 The core idea is that if multiple observed time series are driven by a common, unobserved causal signal, then simultaneous “recurrence” events (periods where the series behave similarly to their past states) in the observed series are indicative of recurrences in the shared driver.
• Methodology (SHREC): It is a physics-based unsupervised learning algorithm that reconstructs these unobserved causal drivers. It works by first reconstructing the attractor of each time series using time-delay embedding. Then, it builds a weighted recurrence network for each response, not with a fixed threshold, but using a method based on topological data analysis that captures local attractor density (fuzzy simplicial complex). These individual recurrence graphs are aggregated into a “consensus graph” representing shared patterns. Finally, the driver’s dynamics are reconstructed by traversing this consensus graph, using community detection for discrete drivers or graph Laplacian eigenvectors for continuous drivers.56
• Applications: SHREC has been demonstrated to infer shared causal drivers in diverse experimental datasets, including ecological time series (plankton abundances driven by temperature), genomic data (gene expression regulated by a transcription factor), fluid dynamics (tracer particles in a chaotic flow), and physiology (fetal ECG influenced by maternal respiration).56
• Relation to CAS: This method is directly applicable to understanding hidden influences and information transfer mechanisms in complex, dynamically evolving systems, even when measurements are noisy or non-linear.
• Current Research Stage: Basic research with promising initial applications, outperforming classical signal processing and some deep learning techniques on specific benchmarks.56

3.2.3. Isometry, Homology, and Topological Data Analysis (TDA): Preserving Fundamental Structures for Causal Inference

• Isometry in ML: The concept of isometry in machine learning refers to learning representations that preserve essential structural relationships or “distances” from the original data space (or a latent ground-truth space) into the learned embedding space.21 For causal discovery, an “isometric” representation of causal factors would mean that the relationships (e.g., strength of influence, conditional independencies) between true causal variables are faithfully reflected in the geometry of the learned latent space. This can lead to more robust and generalizable causal feature identification. The ReX method, for example, uses ML models and Shapley values to identify significant causal relationships, aiming to bridge predictive modeling with causal inference.21 Causal representation learning (CRL) often seeks identifiability guarantees, ensuring the learned latents correspond to true causal variables, which is related to preserving the essential causal structure.72
• Homology & TDA in ML: Homology is a concept from algebraic topology that describes the global structure of shapes by counting their “holes” of different dimensions (0D: connected components, 1D: loops, 2D: voids, etc.). Persistent homology, a key tool in TDA, tracks these topological features across multiple scales of a dataset (e.g., by varying a proximity threshold to build a sequence of simplicial complexes, called a filtration).74 This allows for the identification of robust, multi-scale topological features that are less sensitive to noise or minor perturbations. In the context of CAS and causal discovery, TDA can help identify significant patterns, clusters, and cyclic dependencies in high-dimensional signal data.56 For example, persistent loops might indicate feedback mechanisms in a CAS, or persistent connected components could identify distinct regimes of system behavior. The SHREC algorithm utilizes TDA concepts in constructing its recurrence graphs.56
• Causal Implications: Preserving topological invariants (via TDA/homology) or isometric relationships (in CRL) can lead to causal discovery methods that are more robust to data noise and better at generalizing. If a causal structure has an inherent topology (e.g., cyclic dependencies forming a loop), TDA can help detect it. If causal factors have specific relational structures, isometry-preserving embeddings aim to capture them.
• Current Research Stage: TDA is finding increasing applications in various scientific domains. The use of homology for direct causal inference is still largely theoretical and basic research, though its ability to characterize system structure is highly relevant. Isometry in CRL is an active research area.

The “objectivity” sought in understanding CAS could be partially achieved by employing geometric and topological priors that align with the system’s intrinsic structure. If a CAS possesses a true hierarchical causal structure, an HGNN that accurately embeds it with low distortion is capturing a more objective feature than a generic model. Similarly, persistent topological features identified by TDA are likely more fundamental than transient patterns. The challenge lies in selecting the appropriate geometric or topological lens for a given CAS.

3.2.4. Hyperbolic 3-Manifolds: Advanced Geometries for Fundamental Modeling

• Theoretical Basis: Hyperbolic 3-manifolds are specific types of three-dimensional spaces with constant negative curvature. Their complex geometry allows for the representation of highly intricate relational structures and symmetries.79 Thurston’s geometrization conjecture (now a theorem) shows that hyperbolic geometry is fundamental to the structure of many 3-manifolds. Research in machine learning is beginning to explore if these advanced geometric structures can provide more powerful ways to model data with exceptionally complex dependencies or symmetries, potentially beyond what simpler hyperbolic models or graph structures can capture.81
• Potential for Causal Discovery/CAS: This is a highly speculative area. If the causal interactions within a CAS or the structure of its state space exhibit properties that are naturally described by hyperbolic 3-manifolds (e.g., certain types of knot-theoretic relationships or very specific symmetry groups), then embedding data into such manifolds could, in theory, reveal fundamental aspects of the system’s organization or dynamics. For example, the topology of causal loops or feedback networks might find a more natural representation in these spaces.
• Current Research Stage: This is deeply theoretical. While there is mathematical research on hyperbolic 3-manifolds themselves 79 and some early work on using ML to study properties of 3-manifold triangulations 82, direct applications in machine learning for causal discovery or CAS modeling are very nascent. The primary challenge is developing methods to meaningfully map empirical data from signals onto these complex mathematical objects and then perform inference. The group-theoretic framework for ML in hyperbolic spaces proposed in 81, focusing on hyperbolic balls (which are related to models of hyperbolic space), is a step towards more principled mathematical approaches for hyperbolic ML generally.81

3.3. Hybrid and Specialized AI Approaches

Beyond general-purpose architectures and geometric methods, specialized and hybrid AI approaches are being developed to tackle the unique challenges of causal discovery and modeling first principles.

3.3.1. Neuro-Symbolic AI: Integrating Learning with Reasoning for First Principles

• Concept: Neuro-Symbolic AI aims to combine the strengths of connectionist approaches (neural networks), which excel at learning patterns from large, noisy, and unstructured data, with symbolic AI, which is strong in explicit knowledge representation, logical reasoning, and interpretability.83 This integration is seen as crucial for building AI systems that can not only learn but also reason about what they have learned, explain their conclusions, and incorporate existing domain knowledge or fundamental rules.89
• Relevance to First Principles: This paradigm directly supports the discovery and application of “first fundamentals.” Symbolic components can represent known laws, logical rules, or constraints, while neural components can learn from data to instantiate, refine, or discover new aspects of these principles. Google’s Neural Symbolic Machines, for instance, learn to generate Lisp programs (logical forms) that are executed by a symbolic “computer” over a knowledge base.33 Similarly, their work on adaptable neural memory over symbolic knowledge allows for interpretable modification of factual knowledge without full retraining.93 Google DeepMind’s AlphaGeometry solves Olympiad-level geometry problems by combining a neural language model (for suggesting constructions) with a symbolic deduction engine (for formal proof).34
• Applications: Key applications include knowledge graph reasoning 94, where neural methods learn embeddings of entities and relations while symbolic logic defines inference rules; explainable AI (XAI), where symbolic representations can provide justifications for neural network decisions; and building more robust and generalizable AI for complex decision-making in CAS, such as in healthcare or robotics.84
• Current Research Stage: This is an active area of basic and applied research. While the conceptual appeal is strong, developing truly seamless and scalable neuro-symbolic architectures remains a significant challenge. Gaps exist in explainability, trustworthiness, and meta-cognition within these systems.83

The development of hybrid AI systems that can manage the trade-off between the expressive power of deep learning and the interpretability of symbolic or physics-based components is a central theme. For discovering “objective first fundamentals,” the causal reasoning process itself must be transparent. Simply connecting a neural network to a symbolic system might not suffice; the causal claims made by the model need to be scrutable. This points towards a need for neuro-symbolic architectures where the neural components inherently learn interpretable causal representations or where causal attribution methods can clearly trace the reasoning path.

3.3.2. Physics-Informed AI (PINNs): Embedding Physical Laws for Discovering Governed Causality

• Concept: Physics-Informed Neural Networks (PINNs) integrate known physical laws, typically expressed as partial differential equations (PDEs) or ordinary differential equations (ODEs), directly into the training process of neural networks.67 This is usually achieved by adding a component to the loss function that penalizes deviations of the neural network’s output from these physical laws, evaluated at collocation points across the domain. The neural network thus learns to satisfy both the data and the governing physical constraints.67
• Relevance to First Principles: PINNs directly embody the concept of “first principles” by ensuring that the learned solutions are consistent with established scientific laws. They can be used not only to solve PDEs where boundary/initial conditions are known but also for inverse problems, such as discovering unknown parameters within physical equations or even identifying the form of the governing equations themselves from data (system identification).99 This makes them powerful tools for scientific discovery.
• Applications: PINNs are increasingly used for simulating complex physical systems where data may be sparse or noisy, as the physics provides a strong regularization. Applications span fluid dynamics, solid mechanics, heat transfer, materials science, climate modeling, quantum mechanics, and biomedical engineering.55 For example, Microsoft is using physics-informed AI in fusion energy research for plasma and materials simulation 101, and Google has used physics-aware downsampling for scalable flood modeling.55 Quantum PINNs (QPINNs) are an emerging extension that integrates quantum computing with ML to solve or discover differential equations.104
• Current Research Stage: PINNs are an active area of both basic and applied research, with a rapidly growing number of applications in scientific and engineering domains. Challenges remain in training stability, handling complex geometries, and scaling to very large systems.105

3.3.3. Dedicated Causal AI Frameworks and Toolkits

• Purpose: To provide researchers and practitioners with structured software libraries and ecosystems designed for end-to-end causal inference. These frameworks aim to operationalize theoretical concepts from causal inference literature. Key examples include Microsoft’s Causal AI Suite (featuring DoWhy, EconML, Causica, ShowWhy) 3, the PyWhy ecosystem (which includes DoWhy and EconML) 4, and QuantumBlack’s CausalNex.4 Google also contributes significantly to causal inference methods, often focusing on combining experimental and observational data.23
• Key Features: These toolkits typically guide users through the four steps of causal analysis:

1. Model: Explicitly defining causal assumptions, often using causal graphs (e.g., Directed Acyclic Graphs – DAGs) to represent hypothesized relationships and potential confounders.3
2. Identify: Determining if the causal effect of interest is identifiable from the available data and the stated assumptions, using principles like do-calculus or back-door/front-door criteria.3
3. Estimate: Applying statistical or machine learning methods to estimate the magnitude of the causal effect (e.g., Average Treatment Effect – ATE, Conditional Average Treatment Effect – CATE).3 This often involves techniques like matching, weighting, meta-learners, or deep learning models for nuisance function estimation.
4. Refute: Testing the robustness of the causal conclusions to violations of assumptions (e.g., placebo treatments, random common causes, data subset validation).3 Many frameworks also incorporate methods for causal discovery (learning the graph structure from data) [3 (Causica), 21 (ReX)], and for counterfactual reasoning.25

• Current Research Stage: This is an area of very active development and applied research. Toolkits are becoming more mature and user-friendly, leading to increasing adoption in industry (e.g., finance, healthcare, marketing 4) and academia for scientific investigation.43 However, challenges remain, particularly in scaling to very high-dimensional problems, handling complex data types (text, images), ensuring the validity of assumptions in practice, and making these tools accessible to non-experts.28

The table below provides a comparative overview of these deep tech methodologies.

Table 1: Deep Tech Methodologies for Causal Discovery in Complex Adaptive Systems

 

Methodology	Core Theoretical Principle for Causality/CAS	Key Research Stage	Prominent Research Institutions/Groups (Examples)	Potential for “First Fundamentals” Discovery
LLMs for Causal Discovery	Leverage vast textual knowledge for hypothesis generation, knowledge integration, and refining statistical causal models. Contextual understanding via attention mechanisms. 1	Basic & Applied	Google (AI Co-Scientist 2), Stanford, Various academic groups using LLMs for science 62	Moderate: Strong for hypothesis generation and literature synthesis; limited for de novo causal law discovery without integration with other methods.
Transformers for Temporal Causal Discovery / System ID	Capture long-range temporal dependencies and sequential patterns to infer causal precedence or learn system dynamics (governing equations). 58	Basic & Early Applied	Tsinghua University (CausalFormer 58), Groups working on AI for PDEs/ODEs 66	Moderate to High: Potential to learn dynamical laws from time-series data, approaching fundamental system descriptions.
Hyperbolic Graph Neural Networks (HGNNs)	Utilize hyperbolic geometry’s capacity to embed hierarchical and scale-free structures common in CAS with low distortion. 68	Theoretical & Basic	Stanford, McGill, Various GDL research groups.	Moderate: If CAS has inherent hyperbolic structure, HGNNs can reveal fundamental organizational principles. Causal interpretation is an active research area.
Signal Geometric Diffusion (e.g., SHREC)	Infers unobserved shared causal drivers from simultaneous recurrence patterns in multiple time series, using skew-product dynamics and TDA. 56	Basic & Early Applied	UCSD, Salk Institute.	High: Aims to reconstruct fundamental driving signals/mechanisms from observed system responses, even in noisy/non-linear CAS.
Isometry/Homology/TDA in ML	Preserve essential structural/topological features of data (isometry for relations, homology for shape/connectivity) for robust feature learning and pattern discovery in CAS. 72	Theoretical & Basic (TDA more applied)	Various academic groups in TDA and Causal Representation Learning.	Moderate to High: Identifying robust, invariant features can lead to discovery of fundamental system properties or constraints.
Hyperbolic 3-Manifolds in ML	Explores highly complex geometric structures for representing data with intricate relational properties or fundamental symmetries. 81	Deeply Theoretical	Academic groups in mathematics and theoretical physics, with nascent ML exploration.	Low to Moderate (Currently): Highly speculative; potential if CAS properties map to such geometries.
Neuro-Symbolic AI	Integrates neural learning with symbolic reasoning and knowledge representation. 34	Basic & Applied	Google (NSM, AlphaGeometry 33), Microsoft (Neuro-Symbolic VQA 115), Baidu (ERNIE with KGs 89), Academic groups.	High: Directly aims to combine data-driven learning with explicit, interpretable rules and reasoning, suitable for deriving or verifying fundamental principles.
Physics-Informed Neural Networks (PINNs)	Embeds known physical laws (PDEs/ODEs) as soft constraints in neural network training. 67	Basic & Applied	Broad adoption in science/engineering. Microsoft (Fusion 101), Google (Flood modeling 55), NASA (QNN modeling 8).	High: Directly incorporates first principles (physical laws) and can be used for system identification (discovering parameters or terms in laws).
Dedicated Causal AI Toolkits (DoWhy, Causica, etc.)	Provide end-to-end frameworks for causal modeling, identification, estimation, and refutation based on established causal inference theory. 3	Applied & Development	Microsoft (Causal AI Suite 3), Google (Causal inference methods 23), Open-source community (PyWhy 4).	Moderate: Facilitate application of causal principles but rely on user-specified assumptions or statistical discovery methods; “fundamentals” depend on the quality of these inputs.

4. Global Research Landscape: Epicenters of Deep Tech Causal Discovery

The pursuit of causal understanding in CAS using deep tech is a global endeavor, with distinct contributions and strategic emphases emerging from corporate labs, government initiatives, and academic institutions across various nations.

4.1. Leading Corporate Laboratories

Several multinational technology corporations are at the vanguard of developing and applying these advanced AI methodologies.

• Google (incorporating Google Research and Google DeepMind): Google has demonstrated a broad and deep commitment to AI for scientific discovery and fundamental ML research.

• Causal AI: Their research spans foundational methodologies, such as combining randomized trials with observational studies 23, and practical applications in diverse fields like healthcare and online advertising, including the development of tools for counterfactual analysis.25 The “AI Co-Scientist” project, leveraging the Gemini models, aims to automate hypothesis generation and research planning.2 DeepMind’s work on “Robust agents learn causal world models” explores how agents might inherently learn causal structures for better generalization.117
• Neuro-Symbolic AI: Notable projects include Neural Symbolic Machines for semantic parsing 33, research into adaptable and interpretable neural memory operating over symbolic knowledge bases 93, and the AlphaGeometry system that solves complex geometry problems by integrating a neural language model with a symbolic deduction engine.34
• Physics-Informed AI & Complex Systems: Google has applied physics-aware deep learning to scalable flood modeling 55 and supports research into physics-informed ML for applications like wildfire pollutant transport forecasting.102 Their SYMBIOSIS platform aims to make systems thinking and complex adaptive systems modeling more accessible for societal challenges, using AI to translate complex system diagrams into natural language and vice-versa.13
• AI for Scientific Discovery (General): Beyond specific causal methods, Google’s AlphaFold has revolutionized protein structure prediction 45, and GNoME has accelerated materials discovery.46 They are also active in AI for weather and climate modeling and earth observation.45
• Quantum Technologies: The Google Quantum AI lab (a joint initiative with NASA and USRA 119) is a pioneer, known for the Sycamore processor and the 2019 quantum supremacy experiment.120 More recently, the Willow chip represents further hardware advancement.45 They are also exploring quantum sensing, for example, in collaboration with UConn on using qubits as gravity sensors 40, and advocate for industry-academia alliances to tackle scaling challenges.122

• Microsoft Research: Microsoft is heavily invested in creating platforms and tools to empower scientific discovery through AI and quantum computing.

• Causal AI: Their Causal AI Suite, including DoWhy, EconML, Causica, and ShowWhy, aims to democratize causal inference for decision-making and spur “use-inspired basic research”.3 They also explore the intersection of causal reasoning and LLMs.123
• Neuro-Symbolic AI: Projects include Neuro-Symbolic Computation for inference reasoning in NLP and multimodal contexts 124, and research into Neuro-Symbolic Visual Question Answering.115
• Physics-Informed AI & Complex Systems: Microsoft’s AI for Science initiative supports research across biology, chemistry, and physics.125 A significant focus is on AI for fusion energy research, employing physics-informed AI for plasma and materials simulation.101 They are also investigating AI agents and multi-agent systems for complex problem-solving and modeling CAS dynamics.126 Their work touches on CAS theory for service platform generativity 129 and climate health impacts.15
• Quantum Technologies: Microsoft is pursuing a unique path to scalable quantum computing with its topological qubit approach, exemplified by the Majorana 1 chip.9 Azure Quantum provides a platform for accessing quantum hardware and software, including Azure Quantum Elements, designed to accelerate chemistry and materials science R&D by combining AI, HPC, and quantum tools.132 While their primary quantum focus is on computation, the broader strategy acknowledges sensing as a key quantum technology area.9 Their historic Station Q at UC Santa Barbara has been a hub for topological quantum research.135

• Baidu Research: The Chinese technology giant has significant AI capabilities and has explored quantum computing.

• Causal AI: Publications indicate work on mitigating biases in recommendation systems using causal approaches.136 Their PaddlePaddle deep learning platform has been used for simulating information propagation dynamics with physical neural networks, which can lend itself to causal interpretation.137
• Neuro-Symbolic AI & Knowledge Graphs: Baidu’s ERNIE series of LLMs is known for integrating external knowledge graphs to enhance reasoning and performance.89 They conduct research on various aspects of knowledge graph representation and reasoning.94
• Physics-Informed AI & Scientific Discovery: Baidu’s open-source PaddlePaddle platform 139 includes PaddleHelix for biocomputing (drug discovery, molecular property prediction, mRNA vaccine design 48) and the forthcoming PaddleScience for broader scientific computing, enabling the incorporation of physics constraints.
• Complex Adaptive Systems: Research includes modeling systems like Baidu Encyclopedia using system dynamics 141 and developing advanced Transformer architectures like the Inner Thinking Transformer (ITT) for adaptive token routing in complex tasks.142 Their Baidu Intelligent Edge (BIE) platform extends cloud AI capabilities to edge devices for local data analysis and low-latency computation.143
• Quantum Computing: Baidu established a quantum computing research center in 2018, developing the “Qian Shi” quantum computer and the “Liang Xi” quantum software-hardware integration solution, which includes the Paddle Quantum machine learning toolkit.144 Their research focused on Quantum AI, Algorithms, and Architecture.144 However, in January 2024, Baidu announced the donation of its quantum computing laboratory and equipment to the Beijing Academy of Quantum Information Sciences (BAQIS) 147, signaling a shift in its direct hardware development strategy, though it continues to collaborate with BAQIS, for instance, on an IP alliance.150 Baidu’s 20-F filings mention investment in AI and, historically, quantum computing.155

• Nvidia: Primarily an enabler of AI and scientific computing through its GPU hardware and software stacks (CUDA, cuQuantum).

• AI for Scientific Discovery: Their BioNeMo platform provides models and frameworks for drug discovery and biomolecular research.47
• Quantum Computing: Nvidia’s strategy is not to build quantum processors but to accelerate the quantum ecosystem through hybrid quantum-classical computing. Their cuQuantum SDK facilitates the simulation of quantum circuits on GPUs, and CUDA-Q is a platform for hybrid systems.157 They have launched the NVIDIA Accelerated Quantum Center to integrate quantum hardware with AI supercomputers and foster collaboration.157

• Intel: Focuses on both AI hardware/software and a distinct approach to quantum computing.

• Fundamental AI Research: The Intel AI Lab conducts research in ML for semiconductor design, efficient ML architectures, and computational chemistry, exploring techniques like Transformers and GNNs for multi-objective optimization.158
• Quantum Computing: Intel is pursuing silicon spin qubits, with developments like the “Tunnel Falls” 12-qubit chip. They collaborate with academic and national labs (e.g., LQC at University of Maryland) to democratize access to their quantum hardware for research, aiming for a full-stack commercial quantum system.159

• AMD: A major player in high-performance computing hardware, supporting AI and quantum research.

• Fundamental AI Research: AMD’s Research and Advanced Development (RAD) group works on next-generation processor architectures and AI/HPC workloads. They support open science, as seen with Essential AI’s research using AMD Instinct GPUs.29
• Quantum Computing: AMD Instinct GPUs are used in supercomputers (like Frontier at Oak Ridge) to accelerate quantum simulations and algorithm testing. Their adaptive SoCs and FPGAs are also used in quantum machine control units. They partner with entities like Elevate Quantum and Riverlane.161

• ARM Holdings: Known for its ubiquitous processor architecture, ARM is increasingly focusing on AI enablement from edge to cloud.

• Fundamental AI Research: Provides CPU and GPU IP optimized for AI, along with software like the Arm NN SDK. They engage in partnerships and support open-source AI frameworks.27
• Quantum Computing: In collaboration with Equal1, ARM has tested its Cortex processor at cryogenic temperatures (3.3 Kelvin), aiming to integrate classical compute capabilities within quantum systems (QSoCs) for applications in quantum AI and drug discovery.164

• Other Technology Firms:

• Samsung: Conducts AI research focused on language, voice/sound, and vision AI to enhance its consumer devices.165 While Qiskit is used for materials science and QML 166, this refers to the general IBM framework rather than a specific Samsung quantum program.
• LG: Has joined the IBM Quantum Network to explore applications of quantum computing in AI, connected cars, and IoT.10 Their LG AI Research division is also working on protein structure prediction using their EXAONE LLM for drug discovery.44
• Toshiba: Is a significant player in Quantum Key Distribution (QKD).167 Their Cambridge Research Laboratory (CRL) is also investing in Embodied AI, focusing on agents that can interact physically and learn cooperatively, using NLP for pose estimation and developing efficient world model training.168
• NEC: Focuses on quantum annealing machines for solving complex combinatorial optimization problems, with a history of developing solid-state quantum bits.169 Their European labs also conduct research in data science, computational biology, and materials informatics.169
• Hitachi: Engages in broad AI research spanning data science (ML, RL, DL, knowledge graphs), intelligent vision (multimodal AI, generative AI), and media intelligence processing (domain-specific LLMs, LLM agents) for applications in logistics, anomaly detection, and natural sciences.171 They have also launched a $400 million fund targeting disruptive technologies including quantum computing and AI.172
• Cisco: Their AI Research team works on Generative AI for networking automation, human-computer interaction, and foundational representations for networking.173 In quantum, they have developed a prototype entanglement chip for scalable, distributed quantum computing and established Cisco Quantum Labs for quantum networking research.174
• Broadcom: Collaborates with UC Berkeley’s Sky Computing Lab on open AI ecosystems (vLLM, SAISI) 175 and is developing quantum-resistant network encryption solutions.176
• Panasonic: Conducts general R&D in image recognition, data analysis, robotics, and AI/IoT platforms.178 No specific deep tech initiatives for fundamental causal discovery or quantum were found in the provided material.
• Quantinuum: (Though not a traditional “electronics conglomerate,” it is a major integrated quantum company formed from Honeywell Quantum Solutions and Cambridge Quantum). They are leaders in trapped-ion quantum computing and develop quantum software, including “Quantum Origin” for quantum random number generation and tools for quantum chemistry and AI/ML (compositional intelligence).179

4.2. Governmental Laboratories and National Initiatives

National strategies and government-funded labs play a crucial role in fostering long-term, high-risk research in deep tech.

• USA:

• The National Quantum Initiative (NQI), established by the NQI Act of 2018, is a coordinated federal program to accelerate QIS R&D for economic and national security.186 It involves agencies like the National Science Foundation (NSF), Department of Energy (DOE), National Institute of Standards and Technology (NIST), with increasing roles for NASA, Department of Defense (DOD), and the Intelligence Community (IC).5 The White House Office of Science and Technology Policy (OSTP) hosts the National Quantum Coordination Office (NQCO) and the National Quantum Initiative Advisory Committee (NQIAC) to guide these efforts.5 The NQI focuses on quantum computing, networking, and sensing, with an emphasis on international cooperation and workforce development.5
• NASA is actively researching quantum sensing for Earth observation (e.g., Quantum Gravity Gradiometer pathfinder, Rydberg radiometers) 8 and quantum computing for analyzing large scientific datasets and physics-aware QNN modeling.8 NASA is also a partner in the Google Quantum AI Lab.119
• DARPA runs initiatives like the Quantum Benchmarking Initiative (QBI) to assess the feasibility of utility-scale fault-tolerant quantum computers.131
• The DOE Office of Science has a roadmap for QIS, supporting R&D in quantum computing, sensing, and networks, and highlighting the need for breakthroughs in enabling technologies.188 DOE National Laboratories like Argonne, Oak Ridge, and Brookhaven are part of collaborations such as the Deepspeed4Science Initiative (with AMD and Microsoft) for large-scale scientific discovery.29
• The Defense Innovation Unit (DIU) is working on transitioning quantum sensing technologies for applications like inertial sensors and gravimeters for navigation.37
• NIST plays a role in standards development, including for post-quantum cryptography and validating technologies like Quantinuum’s Quantum Origin QRNG.174
• Think tanks like the Special Competitive Studies Project (SCSP) 189 and the Information Technology and Innovation Foundation (ITIF) 7 provide analyses and policy recommendations regarding US competitiveness in AI and quantum technologies, often highlighting the strategic competition with China.

• China:

• China has made AI development a national priority, with substantial government funding, policy incentives, and public-private partnerships aimed at achieving global leadership.6 The Ministry of Science and Technology (MOST) designates “national high-tech enterprises” to foster homegrown AI champions.191
• In quantum technologies, China has invested heavily (over $15 billion claimed in public funding for R&D 7) and has demonstrated leadership in quantum communication (e.g., the Beijing-Shanghai QKD network and the Micius satellite 7). It is also strong in quantum sensing research output and patents.7
• There is a trend towards increasing state control over quantum R&D, evidenced by major tech companies like Alibaba and Baidu donating their quantum research labs and equipment to government-backed academic institutions such as Zhejiang University and the Beijing Academy of Quantum Information Sciences (BAQIS).7
• BAQIS is a key government-run institute, now housing Baidu’s former quantum lab, and collaborates on initiatives like quantum computing IP alliances.150
• The China Academy of Information and Communications Technology (CAICT) is involved in developing technical standards for quantum technologies.194

• Europe (EU, UK, Germany, France):

• The European Union has strategic initiatives like the Quantum Flagship program (a €1 billion investment 195) and funds research through programs like Horizon Europe.169 The European Commission actively analyzes the role of AI in science and advocates for increased funding and infrastructure.6
• The United Kingdom has a strong research base, with entities like the Quantum Science and Technology Hub at the University of Portsmouth focusing on quantum sensing.36 The Alan Turing Institute serves as a national center for AI and data science, fostering collaborations for AI in scientific discovery.196 Reports highlight the need for growing private capital to support its deep tech ecosystem.197
• Germany hosts significant research activities, including NEC Laboratories Europe in Heidelberg (focusing on data science, computational biology, materials informatics, human-centric AI) 169, and numerous Max Planck Institutes conducting fundamental research in areas like intelligent systems, causal inference, and complex systems AI.198
• France boasts prominent research institutions like INRIA (National Institute for Research in Digital Science and Technology), CNRS (National Centre for Scientific Research), Sorbonne University, PSL University, and Institut Polytechnique de Paris, all active in AI and related deep tech fields.201

• Japan:

• Key institutions include RIKEN AIP (Center for Advanced Intelligence Project), which focuses on AI for scientific discovery, causal inference, and complex systems modeling.203 RIKEN also collaborates with companies like NTT on quantum computing cloud services.206
• NTT has a dedicated Physics of Artificial Intelligence (PAI) Group exploring fundamental AI mechanisms, LLM uncertainty, and in-context learning, alongside research in optical quantum computing and the IOWN (Innovative Optical and Wireless Network) concept.206
• The Japanese government has pledged significant investment in semiconductor and AI-related technology sectors.208

• South Korea:

• Leading universities like KAIST, Seoul National University (SNU), POSTECH, and research institutes like KIST are active in AI for scientific discovery, causal inference methods, and deep learning for complex systems.113 SNU, for instance, collaborates with LG on AI for protein prediction.44

• Israel:

• Renowned institutions such as the Weizmann Institute of Science, Technion – Israel Institute of Technology, Tel Aviv University, and the Hebrew University of Jerusalem are significant contributors to fundamental AI research, including causal representation learning and machine learning for scientific discovery.210

• Singapore:

• The Agency for Science, Technology and Research (A*STAR), through its Centre for Frontier AI Research (CFAR), focuses on strategic AI research pillars including AI for Science, Artificial General Intelligence, Resilient & Safe AI, Sustainable AI, and Theory and Optimisation in AI.213 NUS and NTU are also key academic players.

• Indonesia:

• Institutions like Bandung Institute of Technology, University of Indonesia, and the National Research and Innovation Agency (BRIN) are involved in AI research for complex systems and data science.214 BINUS University’s Bioinformatics & Data Science Research Center (BDSRC), in collaboration with NVIDIA AI R&D Center, works on bioinformatics, LLMs, transformers, and GNNs.214

A notable pattern in the global deep tech landscape is the contrast between “national champion” strategies, particularly visible in China’s consolidation of quantum efforts within state-backed institutions 7, and “open ecosystem” approaches prevalent in the US and parts of Europe. The US model, while featuring substantial government funding through initiatives like the NQI 5, thrives on a dynamic interplay between large corporate R&D (Google, Microsoft), a vibrant startup scene often emerging from universities, and the widespread use of open-source tools.3 European strategies often involve collaborative, publicly funded consortia.169 The long-term effectiveness of these differing models in achieving fundamental breakthroughs in causal discovery remains an open question, suggesting that the socio-economic structure supporting research is as critical as the technological pursuits themselves.

4.3. Academic Powerhouses and University Consortia

Universities are the bedrock of theoretical innovation and talent development in these advanced fields.

• USA: Stanford University, MIT, and UC Berkeley (Sky Computing Lab, active in open AI ecosystems 175) are consistently at the forefront. The University of Maryland hosts the Qubit Collaboratory (LQC), partnering with Intel on quantum chip access.159 Caltech (Prof. John Preskill, a key theorist in quantum computation 121), the University of Chicago’s Pritzker School of Molecular Engineering (AI-driven quantum materials synthesis 216), Yale University (van Dijk Lab, focusing on multiscale foundation models for biomedicine, spatiotemporal dynamical systems, neural operator learning, and causal inference 50), the University of Connecticut (quantum sensing collaboration with Google 40), and Columbia University (causal AI in healthcare 112, Deepspeed4Science initiative 29) are also major contributors.
• China: Tsinghua University is prominent in AI research, including applications in psychology using LLMs for science 114 and hierarchical causal discovery algorithms.222 Peking University and Zhejiang University (notable for its collaboration with Alibaba in quantum computing after receiving its lab 147 and research in physics-informed AI for acoustics 57) are key players. The University of Science and Technology of China (USTC) in Hefei is a world-leading center for quantum research, responsible for breakthroughs like the Micius satellite, and the Jiuzhang and Zuchongzhi quantum computers.193
• UK: The University of Portsmouth’s Quantum Science and Technology Hub is active in quantum sensing.36 The University of Cambridge hosts leading groups like the van der Schaar Lab, which focuses on Causal Deep Learning, AutoML, and time series analysis for healthcare and medical knowledge discovery.26 The University of Oxford, University College London (UCL), Imperial College London, and the University of Edinburgh are also significant centers for AI and scientific discovery research, including multimodal AI applications.60
• Germany: The Max Planck Society, with institutes like the MPI for Intelligent Systems and MPI for Dynamics and Self-Organization, conducts cutting-edge research in areas including causal inference, machine learning for first principles, and complex systems AI.198 Technical University of Munich, Heidelberg University, and Karlsruhe Institute of Technology (KIT) are also leading academic institutions.
• France: A strong academic ecosystem includes INRIA, CNRS, Sorbonne University, PSL University, and Institut Polytechnique de Paris, contributing to geometric deep learning, AI for fundamental science, and complex adaptive systems modeling.201 Ecole Normale Superieure is also noted for relevant research.198
• Japan: The University of Tokyo and Kyoto University are major research universities. RIKEN AIP is a key national center for AI research, including AI for scientific discovery and causal inference methods.203 The National Institute of Advanced Industrial Science and Technology (AIST) also contributes.
• South Korea: KAIST (Korea Advanced Institute of Science and Technology), Seoul National University (SNU), Pohang University of Science and Technology (POSTECH), and the Korea Institute of Science and Technology (KIST) lead research in AI for scientific applications, including causal inference and deep learning for complex systems.113
• Israel: The Weizmann Institute of Science, Technion, Tel Aviv University, and the Hebrew University of Jerusalem are hubs for fundamental AI research, including causal representation learning and machine learning for scientific discovery.210
• Canada: The University of Waterloo’s Institute for Quantum Computing (IQC) is notable for work such as developing qudit-based quantum computers for simulating particle physics theories.227
• Russia: Skolkovo Institute of Science and Technology (Skoltech), Moscow State University, and HSE University are involved in research on complex systems modeling, causal discovery algorithms, and AI for scientific applications.228
• Singapore: The National University of Singapore (NUS) and Nanyang Technological University (NTU), along with A*STAR’s CFAR, are driving AI research, with CFAR focusing on AI for Science and Theory/Optimization in AI.99
• Indonesia: Academic contributions come from institutions like Bandung Institute of Technology, University of Indonesia, and BINUS University (Bioinformatics & Data Science Research Center focusing on LLMs, transformers, GNNs in bioinformatics).214

4.4. Synthesis of Global Research Strengths

The following table (Table 2) provides a comparative overview of the leading entities and their strategic emphasis in the domain of AI for fundamental causal discovery.

Table 2: Comparative Overview of Leading Entities in AI for Fundamental Causal Discovery

 

Country/Region	Leading Corporate Labs (Key Focus Areas)	Leading Govt. Labs/Initiatives (Strategic Focus)	Leading University Hubs (Key Research Themes)	Overall National Strategic Emphasis
USA	Google: Causal AI (methodology, AI Co-Scientist), Neuro-Symbolic (NSM, AlphaGeometry), PINNs, CAS modeling, Quantum AI & Sensing. 2 Microsoft: Causal AI Suite, Neuro-Symbolic (NLP, VQA), PINNs (Fusion), Multi-agent CAS, Azure Quantum (Topological Qubits), AI for Science. 3 Nvidia: Enabler (BioNeMo, cuQuantum, CUDA-Q). 47 Intel: AI for chip design, Silicon spin qubits. 158 AMD: AI/HPC hardware, Open science support. 160	NQI, DOE, NSF, NASA, DARPA: Fundamental QIS (computing, sensing, networking), AI for science, exascale computing, democratizing quantum hardware access. 5	Stanford, MIT, UC Berkeley, Yale (van Dijk Lab), Caltech, UChicago, UConn, Columbia: Foundational ML, GDL, causal inference, AI for biomedicine, quantum algorithms & materials. 40	Strong emphasis on both fundamental theory and applied innovation, driven by corporate R&D, university research, and significant government funding. Robust open-source culture. Leadership in LLMs, Causal AI toolkits, and quantum hardware/software ecosystems.
China	Baidu: ERNIE (LLM+KG), PaddlePaddle (AI platform, PaddleHelix for bio-AI, PaddleScience), Past quantum research (Qian Shi, Liang Xi – now donated). 89 Huawei, Tencent, Alibaba (past quantum): Large-scale AI models, cloud platforms. 193 Origin Quantum: Full-stack quantum computing (Wukong chip, QPanda). 233 DeepSeek: Advanced LLMs. 191	MOST, BAQIS, CAICT: National AI and Quantum strategies, state-led consolidation of quantum R&D, standards development. 6	Tsinghua University, Peking University, USTC, Zhejiang University: LLMs for science, causal discovery algorithms, quantum communication & computing breakthroughs. 57	Rapidly advancing with strong state-directed investment in AI and Quantum. Leading in quantum communication, significant progress in LLMs and quantum computing hardware. Focus on national champions and integrating AI into industrial applications.
EU (UK, Germany, France as examples)	NEC Labs Europe (Germany): Data science, computational biology, materials informatics. 169 ARM (UK HQ): AI-enabled CPU/GPU IP, quantum processor testing. 27 Toshiba CRL (UK): Embodied AI, QKD. 167	EU Quantum Flagship, Horizon Europe: Collaborative funding for quantum and AI research. 169 National initiatives in UK, Germany, France supporting AI and quantum ecosystems.	UK: Cambridge (van der Schaar Lab – Causal DL), Oxford, UCL, Edinburgh, Portsmouth (Quantum Sensing). 36 Germany: Max Planck Institutes (Causal inference, ML for first principles), TU Munich, Heidelberg. 198 France: INRIA, CNRS, Sorbonne, PSL (GDL, AI for science, CAS modeling). 201	Strong academic research base. Collaborative, publicly funded projects are common. Emphasis on trustworthy AI, industrial applications, and building sovereign capabilities in strategic deep tech areas.
Japan	NTT: Quantum computing (optical), IOWN, Physics of AI Group. 206 Hitachi, NEC, Toshiba, Panasonic: Corporate AI R&D in various sectors, quantum annealing (NEC), QKD (Toshiba). 167	National strategies for AI and semiconductor development. 208	RIKEN AIP, UTokyo, Kyoto University, AIST: Foundational AI, AI for scientific discovery, causal inference, complex systems. 203	Strong in fundamental research and specialized hardware/materials. Growing focus on AI for societal challenges and integrating AI with existing industrial strengths.
South Korea	Samsung, LG: AI for device enhancement, applied AI (e.g., LG’s protein prediction AI, quantum exploration with IBM). 10	Government support for AI and semiconductor industries.	KAIST, SNU, POSTECH, KIST: AI for scientific discovery, causal inference, deep learning for complex systems. 113	Focus on applied AI, particularly in electronics and manufacturing, with growing academic strength in fundamental AI research.
Israel	Limited large corporate labs in this specific deep tech for causality space compared to US/China, but strong startup ecosystem.	Government support for high-tech R&D.	Weizmann Institute, Technion, Tel Aviv University, Hebrew University: Leading research in ML theory, causal representation learning, AI for science. 210	Highly innovative R&D ecosystem with strong academic research and a vibrant startup scene, particularly in AI algorithms and cybersecurity.
Russia	Limited information in snippets on corporate labs focusing on these specific advanced AI/causal discovery areas.	State support for AI and technology development.	Skoltech, Moscow State University, HSE University: Research in complex systems modeling, causal discovery, AI applications. 228	Academic research in foundational mathematics and AI, with applications in specific national priority areas.
Singapore	Limited information on local corporate labs in snippets, but global tech companies have presence.	Strong government push for AI adoption and R&D (e.g., AI Singapore).	NUS, NTU, A*STAR (CFAR): AI for Science, Theory/Optimization in AI, deep learning for complex systems. 99	Strategic focus on becoming an AI hub, with significant investment in research talent and infrastructure. Strong in applied AI.
Indonesia	Limited information on local corporate labs in snippets.	Government initiatives for digital transformation and AI capacity building.	Bandung Institute of Technology, University of Indonesia, BRIN, BINUS University (BDSRC): Bioinformatics, data science, LLMs, applied ML. 214	Developing AI capabilities, with a focus on applications relevant to national development goals.

5. The Quantum Intersection: Quantum Technologies as a Deep Tech Enabler

While the primary focus of this report is on classical deep learning and mathematical methods for causal discovery, the burgeoning field of quantum technologies—quantum computing and quantum sensing—represents a parallel deep tech frontier with potential long-term synergies. Many of the same leading organizations are investing in both AI and quantum, recognizing their potential to revolutionize scientific discovery and the modeling of complex systems.

5.1. Quantum AI and its Potential for Complex Modeling and Causal Inference

Quantum Artificial Intelligence (QAI) explores how quantum mechanical principles can enhance machine learning algorithms or, conversely, how AI can optimize quantum systems.

• Quantum Machine Learning (QML): This field investigates quantum algorithms that could offer speedups or other advantages for tasks relevant to CAS modeling and causal inference. Areas of interest include quantum algorithms for solving linear systems of equations (potentially useful in fitting large-scale models), quantum principal component analysis, quantum support vector machines, and quantum methods for optimization problems that are intractable for classical computers.166 Baidu’s Paddle Quantum, for instance, was developed for QML research, including combinatorial optimization.144 The integration of quantum computing with causal AI is also an emerging trend.4 Many research groups are exploring how quantum computers might accelerate the training of AI models or enable the processing of classically intractable datasets.237
• Quantum Neural Networks (QNNs): These are models that use quantum circuits as components of neural networks, or are entirely based on quantum principles. Theoretical advantages could include enhanced representational power or more efficient learning.8 An exciting recent development is the Quantum Weighted Tensor Hybrid Network (QWTHN) or Quantum Tensor Hybrid Adaptation (QTHA), proposed by researchers including from Origin Quantum, which uses QNNs and tensor networks for parameter-efficient fine-tuning of large language models, leveraging quantum superposition to potentially overcome classical rank limitations and reduce trainable parameters significantly while improving performance.239 QPINNs (Quantum Physics-Informed Neural Networks) also merge quantum computing with physics-informed machine learning to solve or discover differential equations.104
• Solving Complex Optimization Problems in CAS: Many problems in causal discovery (e.g., structure learning) and CAS modeling involve navigating vast search spaces. Quantum annealing, pursued by companies like NEC 169, and other quantum optimization algorithms aim to find optimal or near-optimal solutions to such computationally hard problems more efficiently than classical heuristics.

The current research stage for QML in direct causal discovery is largely theoretical and basic. However, the potential for quantum computers to tackle the enormous computational challenges posed by high-dimensional CAS data makes this a compelling long-term prospect.

5.2. Quantum Sensing: Novel Data Sources for Understanding Fundamental Phenomena

Quantum sensing leverages quantum mechanical effects like superposition and entanglement to make measurements of physical quantities with precision surpassing classical limits.38

• Principles: Quantum sensors can detect minute changes in gravitational fields, electromagnetic fields, temperature, and other physical parameters by probing the delicate quantum states of atoms, photons, or other quantum systems.36
• Applications: This technology is yielding new ways to gather high-resolution data about physical, biological, and environmental systems.

• Geophysics and Navigation: Quantum gravimeters and inertial sensors (using atom interferometry) promise drift-free navigation in GPS-denied environments and high-precision mapping of Earth’s gravitational field for resource discovery or geodetic studies.8 Google’s collaboration with UConn explores using qubits themselves as gravity sensors.40
• Medical Diagnostics: Quantum sensors are being developed for magnetoencephalography (MEG) and magnetocardiography (MCG) with enhanced sensitivity, potentially revolutionizing brain and heart diagnostics.243 Quantum-enhanced MRI aims for molecular and even atomic-scale imaging, aiding in drug discovery and understanding biological processes at a fundamental level.243
• Environmental Monitoring: Precise measurements of atmospheric composition, temperature, or pollutants could be enabled by quantum sensors, providing critical data for climate models and environmental science.39
• Fundamental Science: Quantum sensors are also tools for fundamental physics research, such as searching for dark matter or testing gravitational theories.41

• Data for CAS: The ultra-sensitive and novel types of data generated by quantum sensors could reveal previously unobservable signals or dynamics within CAS. For example, mapping subtle changes in local environmental fields or tracking molecular interactions in vivo could provide new inputs for AI-driven causal discovery models. AI is also being used to enhance quantum sensing itself, for instance, through intelligent noise reduction and real-time sensor control.246

The development of quantum sensors is in various stages of applied research, with some technologies nearing commercialization. Their integration with AI systems for causal discovery in CAS is a more nascent but highly promising area.

5.3. Overview of Quantum Research in Key Organizations and Countries

Many of the same organizations leading in AI are also investing heavily in quantum technologies, recognizing the long-term transformative potential.

• USA:

• Google Quantum AI: Developed the Sycamore 120 and Willow 45 quantum processors. Active in quantum algorithm development and exploring quantum sensing.40
• Microsoft Azure Quantum: Focused on topological qubits (Majorana 1 chip 131) and providing a cloud platform for quantum computing and applications in chemistry/materials science via Azure Quantum Elements.132
• Intel: Researching silicon spin qubits (Tunnel Falls chip 159).
• AMD & Nvidia: Providing classical HPC hardware and software (e.g., AMD Instinct GPUs 162, Nvidia CUDA-Q 157) that supports quantum research and hybrid quantum-classical systems.
• Government Initiatives (NQI, DOE, NASA, DARPA): Significant funding and coordination for QIS research across computing, sensing, and networking.5

• China:

• Origin Quantum: A leading Chinese company developing full-stack quantum computers, including the “Wukong” superconducting quantum computer, and software like QPanda and VQNet.233 They have also published research on quantum-enhanced LLM fine-tuning.239
• Baidu: Historically researched quantum computing (Qian Shi, Liang Xi, Paddle Quantum 144), but donated its lab to BAQIS.147 Continues to be involved via collaborations and its AI expertise.
• USTC & BAQIS: Leading academic and government-backed research in quantum computing (Jiuzhang, Zuchongzhi 193) and communication. BAQIS is a central institute in China’s quantum strategy.147
• Other companies like QuantumCTek focus on quantum communication.193

• Europe (UK, Germany, France, EU):

• EU Quantum Flagship: A large-scale, long-term research initiative.195
• UK: Strong academic research (e.g., University of Portsmouth in quantum sensing 36). Companies like ARM are exploring quantum device integration.164
• Germany: Home to research labs like NEC Laboratories Europe 169 and Max Planck Institutes.
• France: Academic hubs and startups like Alice & Bob (cat qubits 250).

• Japan:

• NTT: Research in optical quantum computing and the IOWN initiative.206
• NEC: Developing quantum annealing machines.170
• Toshiba: Focus on Quantum Key Distribution.167
• Hitachi: Investing in quantum computing through its new tech fund.172

• Other Notable Players:

• Quantinuum (US/UK): Leader in trapped-ion quantum computing and quantum software (Quantum Origin).179
• Canada: University of Waterloo’s IQC is prominent in quantum information research.227

The global quantum landscape is characterized by intense research and development, with different players focusing on various qubit modalities and aspects of the quantum stack. The intersection with AI is primarily focused on using classical AI to improve quantum systems (e.g., calibration, error correction 237) and exploring QML algorithms, with the application of quantum-enhanced AI to fundamental causal discovery still in its early theoretical stages.

6. Challenges and Future Directions

The ambition to use deep tech and advanced mathematics to uncover objective first fundamentals in CAS is profound, but it faces significant theoretical, computational, and practical challenges. Addressing these will shape the future trajectory of this research frontier.

• Scalability and Computational Complexity:

• Many deep learning models, especially large Transformers and complex GNNs, are computationally expensive to train and deploy. Scaling these to the vast datasets and intricate structures of real-world CAS is a major hurdle.66
• Causal discovery algorithms often involve searching super-exponential spaces of possible causal graphs.77
• Hyperbolic geometry and other advanced mathematical frameworks, while offering representational advantages, can involve more complex computations than Euclidean counterparts.68
• Future Direction: Research into more efficient algorithms, specialized hardware (including neuromorphic and potentially quantum accelerators), model compression, and methods for decomposing complex problems into manageable sub-problems. The work on parameter-efficient fine-tuning of LLMs using quantum-classical hybrids (QWTHN/QTHA) points to novel avenues for efficiency.239

• Data Requirements and Quality:

• Deep learning models typically require large, diverse, and high-quality datasets. For many CAS, such data may be scarce, noisy, incomplete, or expensive to obtain, particularly interventional data.28
• The “omnipresence of signals” is a double-edged sword: while data is abundant, its heterogeneity, varying quality, and inherent biases pose significant integration and interpretation challenges.253
• Future Direction: Development of methods robust to noisy, sparse, or biased data; techniques for effective data fusion from multiple modalities and scales; active learning and optimal experimental design strategies guided by AI to acquire the most informative data; and the use of synthetic data generation, potentially guided by LLMs or physics-informed models, to augment real-world datasets.34

• Interpretability and Explainability (XAI):

• For discovering “objective first fundamentals,” it is not enough for models to be predictive; their reasoning and the causal mechanisms they identify must be understandable and verifiable by human experts.20
• Many deep learning models, particularly large neural networks, operate as “black boxes,” making it difficult to interpret their internal workings or the basis for their predictions.83
• Future Direction: Advancing neuro-symbolic AI to combine learned representations with explicit symbolic reasoning 83; developing inherently interpretable models; creating new XAI techniques tailored for causal discovery in complex systems (e.g., causal attribution methods); and leveraging LLMs to translate complex model outputs into human-understandable explanations.1 Microsoft’s ReX method, using Shapley values for causal discovery, is an example of integrating explainability.21

• Validation of Causal Claims:

• Establishing the ground truth for causal relationships in real-world CAS is often impossible, making direct validation of AI-discovered causal models difficult.28
• Simulations can help, but their fidelity to reality is always a concern.
• Future Direction: Developing robust refutation testing frameworks (as in DoWhy 3); methods for comparing and contrasting causal models derived from different data sources or AI techniques; leveraging any available (even limited) experimental data for partial validation; and establishing benchmarks with well-understood (even if simplified) causal grounds, like Auto-Bench.62

• Integration of Diverse Methodologies:

• As highlighted earlier, no single deep tech method is likely to solve the multifaceted challenge of causal discovery in CAS. The synergistic integration of LLMs, GDL, neuro-symbolic AI, PINNs, and formal causal inference frameworks is crucial.
• Future Direction: Fostering interdisciplinary research and developing modular, interoperable platforms that allow different AI tools and mathematical frameworks to be combined effectively. Standardization of data formats and model interfaces will be important.

• Theoretical Foundations:

• The theoretical underpinnings of why certain deep learning architectures (like Transformers or GNNs) are effective for causal tasks, or how advanced geometries truly map to causal structures, are still being developed.69
• Formalizing concepts like “objective first fundamentals” in the context of data-driven discovery requires further philosophical and mathematical work.
• Future Direction: Continued basic research into the mathematical properties of these models, identifiability conditions for causal discovery in complex settings, and the development of a more rigorous “theory of scientific discovery” for AI.

• Ethical Considerations and Responsible AI:

• As AI systems become more powerful and are used to infer fundamental causal relationships that could inform policy or critical decisions, ensuring their fairness, safety, reliability, and alignment with human values is paramount.27
• The risk of AI models perpetuating or amplifying existing biases, or generating plausible but incorrect causal explanations, must be actively managed.
• Future Direction: Continued development and adoption of Responsible AI principles and practices; research into bias detection and mitigation in causal models; and public discourse on the societal implications of AI-driven scientific discovery.

The journey towards using deep tech to uncover objective first fundamentals in CAS is ambitious and long-term. It requires sustained investment in fundamental research, interdisciplinary collaboration, and a commitment to responsible innovation. While the challenges are substantial, the potential rewards—a deeper understanding of the universe, life, and society, leading to solutions for some of the world’s most pressing problems—are transformative.

7. Conclusion

The exploration of deep technologies—ranging from Large Language Models and Transformers to Geometric Deep Learning, Neuro-Symbolic AI, and Physics-Informed Neural Networks, augmented by advanced mathematical concepts like hyperbolic geometry and topology—is ushering in a new era for scientific inquiry, particularly in the challenging domain of causal discovery within Complex Adaptive Systems. The overarching ambition is to transcend correlational understanding and uncover “objective first fundamentals,” the core generative mechanisms that dictate the behavior of these intricate systems across material, molecular, and observable scales.

Current research indicates a vibrant but fragmented landscape. Significant progress is evident in specialized areas: LLMs show promise as hypothesis generators and knowledge integrators 1; Causal AI toolkits are maturing for practical applications 3; PINNs are effectively embedding physical laws into predictive models 67; and geometric deep learning is offering novel ways to represent complex, hierarchical data.56 However, the application of more abstract mathematical concepts like hyperbolic 3-manifolds or advanced homology for direct causal inference remains largely theoretical. The true power for discovering fundamental causal principles likely lies in the future synergistic integration of these diverse deep tech methodologies, creating hybrid systems that leverage the strengths of each.

Globally, leading corporate labs in the USA (Google, Microsoft, Nvidia, Intel) and China (Baidu, Huawei, Origin Quantum) are heavily investing in both the foundational AI/quantum technologies and their application to scientific discovery. Government initiatives worldwide, notably in the USA, China, and the EU, are providing strategic direction and substantial funding, fostering both national capabilities and international collaborations.5 Universities remain critical hubs for theoretical breakthroughs and talent development across all mentioned geographies. The approaches to fostering these deep tech ecosystems vary, from state-led consolidation in some regions to more open, collaborative, and market-driven models elsewhere, the long-term impact of which on fundamental discovery is yet to be determined.

The intersection with quantum technologies, particularly quantum computing for complex simulations and QML, and quantum sensing for novel data acquisition, presents a longer-term but potentially revolutionary pathway for advancing scientific understanding and CAS modeling.9

The primary challenges ahead are manifold: ensuring the scalability of computationally intensive models, addressing data scarcity and quality issues in complex domains, enhancing the interpretability and explainability of AI-driven causal claims, rigorously validating these claims, and building robust theoretical foundations for these new approaches. Furthermore, the ethical implications of AI-driven discovery and the imperative for responsible innovation must remain central to this pursuit.

Ultimately, the quest for objective first fundamentals through deep tech is not merely a technological endeavor but a scientific and philosophical one. It requires a sustained, interdisciplinary effort to develop AI systems that can not only predict but also explain the causal workings of the complex adaptive systems that shape our world. While the path is arduous and many breakthroughs are still on the horizon, the convergence of these powerful tools and concepts holds the potential to redefine the very nature of scientific discovery.

Works cited

1. Large Language Models for Causal Discovery: Current Landscape and Future Directions, accessed May 9, 2025, https://arxiv.org/html/2402.11068
2. Google’s new AI Co-Scientist: Accelerating Scientific Discovery with …, accessed May 9, 2025, https://wandb.ai/byyoung3/ml-news/reports/Google-s-new-AI-Co-Scientist-Accelerating-Scientific-Discovery-with-AI-Collaboration—VmlldzoxMTQzNDE2NA
3. www.microsoft.com, accessed May 9, 2025, https://www.microsoft.com/en-us/research/wp-content/uploads/2022/11/CausalAISuiteForDecisionMaking.pdf
4. The State of Causal AI in 2025: Summary with Open Source Projects – Sonicviz, accessed May 9, 2025, https://sonicviz.com/2025/02/16/the-state-of-causal-ai-in-2025/
5. Quantum Origin Stories, accessed May 9, 2025, https://www.quantum.gov/quantum-origin-stories/
6. China leads EU and US on using artificial intelligence in science, accessed May 9, 2025, https://sciencebusiness.net/news/ai/china-leads-eu-and-us-using-artificial-intelligence-science
7. China Is Challenging U.S. Leadership in Quantum Technologies; New Report Finds China Leads in Quantum Communication, Tied in Sensing, Behind in Computing, accessed May 9, 2025, https://itif.org/publications/2024/09/09/china-is-challenging-us-leadership-in-quantum-technologies-new-report-finds/
8. Quantum Technology for Earth Science – NASA ESTO, accessed May 9, 2025, https://esto.nasa.gov/quantum/
9. Investing in American leadership in quantum technology: the next frontier in innovation – Microsoft On the Issues, accessed May 9, 2025, https://blogs.microsoft.com/on-the-issues/2025/04/28/investing-in-american-leadership-quantum/
10. LG Joins IBM Quantum Network for Advance Industry Applications of Quantum Computing, accessed May 9, 2025, https://www.lg.com/global/newsroom/news/corporate/lg-joins-ibm-quantum-network-for-advance-industry-applications-of-quantum-computing/
11. Complex Adaptive Systems: Views from the Physical, Natural, and Social Sciences, accessed May 9, 2025, https://books.google.it/books?id=O46dDwAAQBAJ
12. [2502.00012] Lessons from complexity theory for AI governance – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2502.00012
13. Extending the Machine Learning Abstraction Boundary: A Complex …, accessed May 9, 2025, https://research.google/pubs/extending-the-machine-learning-abstraction-boundary-a-complex-systems-approach-to-incorporate-societal-context/
14. SYMBIOSIS: Systems Thinking and Machine Intelligence for Better Outcomes in Society, accessed May 9, 2025, https://arxiv.org/html/2503.05857v1
15. Complex adaptive systems-based framework for modeling the health impacts of climate change – eScholarship, accessed May 9, 2025, https://escholarship.org/content/qt32h6c846/qt32h6c846.pdf
16. Implications of complex adaptive systems theory for the design of research on health care organizations, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC3667498/
17. Fostering Innovation through Complex Adaptive Systems Theory: A Case Study of an Engineering Organization – The Atrium – University of Guelph, accessed May 9, 2025, https://atrium.lib.uoguelph.ca/bitstream/10214/7502/1/Hakimi_Rohan_201308_MASc.pdf
18. Complex adaptive systems-based framework for modeling the health impacts of climate change – PMC – PubMed Central, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC10900873/
19. How ‘causal’ AI can improve your decision-making – IMD Business School, accessed May 9, 2025, https://www.imd.org/ibyimd/artificial-intelligence/how-causal-ai-can-improve-your-decision-making/
20. What is Causal AI? Understanding Causes and Effects – DataCamp, accessed May 9, 2025, https://www.datacamp.com/blog/what-is-causal-ai
21. ReX: Causal Discovery based on Machine Learning and Explainability techniques – arXiv, accessed May 9, 2025, https://arxiv.org/html/2501.12706v1
22. Causal network reconstruction from time series: From theoretical assumptions to practical estimation | Chaos: An Interdisciplinary Journal of Nonlinear Science | AIP Publishing, accessed May 9, 2025, https://pubs.aip.org/aip/cha/article/28/7/075310/386353/Causal-network-reconstruction-from-time-series
23. Causal inference methods for combining randomized trials and …, accessed May 9, 2025, https://research.google/pubs/causal-inference-methods-for-combining-randomized-trials-and-observational-studies-a-review/
24. Deep Learning for Causal Inference – OSF, accessed May 9, 2025, https://osf.io/download/6160f94d42b474012100c43e/
25. Google’s Ai In Causal Inference | Restackio, accessed May 9, 2025, https://www.restack.io/p/causal-ai-answer-googles-ai-in-causal-inference-cat-ai
26. Causal Deep Learning – van der Schaar Lab, accessed May 9, 2025, https://www.vanderschaar-lab.com/wp-content/uploads/2023/02/CDL-1.pdf
27. Arm AI Readiness Index, accessed May 9, 2025, https://www.arm.com/-/media/Files/pdf/report/arm-ai-readiness-index-report-part1.pdf?rev=2f8c6d73c3464702ac91cff6c245372f&revision=2f8c6d73-c346-4702-ac91-cff6c245372f
28. Commentary: Implications of causality in artificial intelligence. Why Causal AI is easier said than done – Frontiers, accessed May 9, 2025, https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2024.1488359/full
29. Responsible AI – AMD, accessed May 9, 2025, https://www.amd.com/en/solutions/ai/responsible-ai.html
30. Governing AI – Microsoft, accessed May 9, 2025, https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-brand/documents/RW1hAdp.pdf
31. Baidu AI Ethics Measures, accessed May 9, 2025, http://esg.baidu.com/detail/560.html
32. Form 10-K for Microsoft Corp filed 07/27/2023, accessed May 9, 2025, https://microsoft.gcs-web.com/static-files/e2931fdb-9823-4130-b2a8-f6b8db0b15a9
33. Neural Symbolic Machines: Learning Semantic Parsers on …, accessed May 9, 2025, https://research.google/pubs/neural-symbolic-machines-learning-semantic-parsers-on-freebase-with-weak-supervision/
34. AlphaGeometry: An Olympiad-level AI system for geometry – Google …, accessed May 9, 2025, https://deepmind.google/discover/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/
35. AI and Complex Adaptive Systems: A Lifecycle Methodology for Verification and Monitoring | ID: 5712m7598 – GW ScholarSpace, accessed May 9, 2025, https://scholarspace.library.gwu.edu/concern/gw_etds/5712m7598
36. Breakthrough in quantum sensing could revolutionise high-precision measurement technologies | University of Portsmouth, accessed May 9, 2025, https://www.port.ac.uk/news-events-and-blogs/news/breakthrough-in-quantum-sensing-could-revolutionise-high-precision-measurement-technologies
37. DIU’s TQS Field Testing To Begin Across Five Critical Areas – Defense Innovation Unit, accessed May 9, 2025, https://www.diu.mil/latest/dius-transition-of-quantum-sensing-tqs-field-testing-to-begin-across-five
38. What is Quantum Sensing? | BAE Systems, accessed May 9, 2025, https://www.baesystems.com/en-us/definition/what-is-quantum-sensing
39. Quantum Advantage in Quantum Sensing by Q-CTRL, accessed May 9, 2025, https://quantumcomputingreport.com/quantum-advantage-in-quantum-sensing-by-q-ctrl/
40. UConn Partners with Google Quantum AI on New Quantum Sensing …, accessed May 9, 2025, https://www.hpcwire.com/off-the-wire/uconn-partners-with-google-quantum-ai-on-new-quantum-sensing-study/
41. Testimony of Dr. Celia Merzbacher Executive Director Quantum Economic Development Consortium SRI International Before the Commit – Congress.gov, accessed May 9, 2025, https://www.congress.gov/119/meeting/house/118196/witnesses/HHRG-119-SY00-Wstate-MerzbacherC-20250507.pdf
42. Metasurface: Unlocking the future of wireless sensing and communication – Microsoft, accessed May 9, 2025, https://www.microsoft.com/en-us/research/blog/metasurface-unlocking-the-future-of-wireless-sensing-and-communication/
43. Causaly Case Study | Google Cloud, accessed May 9, 2025, https://cloud.google.com/customers/causaly
44. ‘The convergence between of AI and the Biology’ LG Creates “Next-Generation Protein Structure Prediction AI” for Drug Discovery – PR Newswire, accessed May 9, 2025, https://www.prnewswire.com/news-releases/the-convergence-between-of-ai-and-the-biology–lg-creates-next-generation-protein-structure-prediction-ai-for-drug-discovery-302371150.html
45. abc.xyz, accessed May 9, 2025, https://abc.xyz/assets/77/51/9841ad5c4fbe85b4440c47a4df8d/goog-10-k-2024.pdf
46. Science AI – Google AI, accessed May 9, 2025, https://ai.google/applied-ai/science/
47. Accelerate AI in Healthcare: NVIDIA BioNeMo + GKE | Google Cloud Blog, accessed May 9, 2025, https://cloud.google.com/blog/products/ai-machine-learning/accelerate-ai-in-healthcare-nvidia-bionemo-gke
48. Artificial intelligence to predict inhibitors of drug-metabolizing enzymes and transporters for safer drug design | Request PDF – ResearchGate, accessed May 9, 2025, https://www.researchgate.net/publication/390876297_Artificial_intelligence_to_predict_inhibitors_of_drug-metabolizing_enzymes_and_transporters_for_safer_drug_design
49. The discovery of subunit-selective GluN1/GluN2B NMDAR antagonist via pharmacophere-based virtual screening – PMC – PubMed Central, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC10854469/
50. Cell2Sentence: Teaching Large Language Models the Language of Biology – PMC, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC11565894/
51. Cell2Sentence: Teaching Large Language Models the Language of Biology – bioRxiv, accessed May 9, 2025, https://www.biorxiv.org/content/10.1101/2023.09.11.557287v3.full-text
52. Google Research – Explore Our Latest Research in Science and AI, accessed May 9, 2025, https://research.google/
53. How we’re using AI to drive scientific research with greater real-world benefit – Google Blog, accessed May 9, 2025, https://blog.google/technology/research/google-research-scientific-discovery/
54. Social impact of AI and how it’s helped society – Google AI, accessed May 9, 2025, https://ai.google/advancing-ai/social-impact/
55. Physics aware downsampling with deep learning for scalable flood …, accessed May 9, 2025, https://research.google/pubs/physics-aware-downsampling-with-deep-learning-for-scalable-flood-modeling/
56. Recurrences Reveal Shared Causal Drivers of Complex Time …, accessed May 9, 2025, https://link.aps.org/doi/10.1103/PhysRevX.15.011005
57. Special Issue: Generative and Physics-Informed Artificial Intelligence for Acoustics, accessed May 9, 2025, https://pubs.aip.org/asa/jasa/pages/cfp030425
58. CausalFormer: An Interpretable Transformer for Temporal Causal …, accessed May 9, 2025, https://www.computer.org/csdl/journal/tk/2025/01/10726725/21dMtID3eUw
59. Graph-Aware Isomorphic Attention for Adaptive Dynamics in Transformers – arXiv, accessed May 9, 2025, https://arxiv.org/html/2501.02393v3
60. (PDF) Towards deployment-centric multimodal AI beyond vision and language, accessed May 9, 2025, https://www.researchgate.net/publication/390545359_Towards_deployment-centric_multimodal_AI_beyond_vision_and_language
61. Multi-Head Attention and Transformer Architecture – Pathway, accessed May 9, 2025, https://pathway.com/bootcamps/rag-and-llms/coursework/module-2-word-vectors-simplified/bonus-overview-of-the-transformer-architecture/multi-head-attention-and-transformer-architecture/
62. Auto-Bench: An Automated Benchmark for Scientific Discovery in LLMs – arXiv, accessed May 9, 2025, https://arxiv.org/html/2502.15224v1
63. arxiv.org, accessed May 9, 2025, https://arxiv.org/abs/2502.15224
64. accessed December 31, 1969, https://arxiv.org/pdf/2502.15224.pdf
65. Integrating Large Language Models in Causal Discovery: A Statistical Causal Approach, accessed May 9, 2025, https://www.alphaxiv.org/overview/2402.01454
66. DISCO: learning to DISCover an evolution Operator for multi-physics-agnostic prediction, accessed May 9, 2025, https://arxiv.org/html/2504.19496v1
67. arXiv:2503.22528v1 [cs.LG] 28 Mar 2025, accessed May 9, 2025, https://www.arxiv.org/pdf/2503.22528
68. Hyperbolic Graph Neural Networks: A Tutorial on Methods and …, accessed May 9, 2025, https://www.researchgate.net/publication/372925746_Hyperbolic_Graph_Neural_Networks_A_Tutorial_on_Methods_and_Applications
69. Hyperbolic Graph Neural Networks, accessed May 9, 2025, http://papers.neurips.cc/paper/9033-hyperbolic-graph-neural-networks.pdf
70. Foundations of Machine Learning – MCML, accessed May 9, 2025, https://mcml.ai/research/areaa/
71. REX: Causal Discovery based on Machine Learning and Explainability techniques – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2501.12706
72. Causal Representation Learning from Multi-modal Biomedical Observations – PMC, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC11952583/
73. arxiv.org, accessed May 9, 2025, https://arxiv.org/pdf/2409.02772
74. Cubical Homology-Based Machine Learning: An Application in Image Classification – MDPI, accessed May 9, 2025, https://www.mdpi.com/2075-1680/11/3/112
75. Category-Theoretical and Topos-Theoretical Frameworks in Machine Learning: A Survey – MDPI, accessed May 9, 2025, https://www.mdpi.com/2075-1680/14/3/204
76. Information Filtering Networks: Theoretical Foundations, Generative Methodologies, and Real-World Applications – arXiv, accessed May 9, 2025, https://www.arxiv.org/pdf/2505.03812
77. A Higher Algebraic K-Theory of Causality[v1] – Preprints.org, accessed May 9, 2025, https://www.preprints.org/manuscript/202501.1242/v1
78. Topological Field Theory of Data: Mining Data Beyond Complex Networks, accessed May 9, 2025, https://www.bcamath.org/sites/default/files/inline-files/Topological%20Field%20Theory%20of%20Data.pdf
79. [1310.3410] Verified computations for hyperbolic 3-manifolds – arXiv, accessed May 9, 2025, https://arxiv.org/abs/1310.3410
80. Differential Geometry Feb 2025 – arXiv, accessed May 9, 2025, http://www.arxiv.org/list/math.DG/2025-02?skip=0&show=250
81. arxiv.org, accessed May 9, 2025, https://arxiv.org/pdf/2501.06934
82. [2405.09610] Learning 3-Manifold Triangulations – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2405.09610
83. Neuro-Symbolic AI in 2024: A Systematic Review – arXiv, accessed May 9, 2025, https://arxiv.org/pdf/2501.05435
84. The Neuro-Symbolic AI Paradigm: Integrating Logic-Based Reasoning with Deep Learning, accessed May 9, 2025, https://www.researchgate.net/publication/390033390_The_Neuro-Symbolic_AI_Paradigm_Integrating_Logic-Based_Reasoning_with_Deep_Learning
85. About Neurosymbolic Artificial Intelligence | Neurosymbolic Artificial Intelligence, accessed May 9, 2025, https://neurosymbolic-ai-journal.com/content/about-neurosymbolic-artificial-intelligence
86. Neuro-Symbolic AI: A Future of Tomorrow – ASEAN Journal on Science and Technology for Development – Universiti Brunei Darussalam, accessed May 9, 2025, https://ajstd.ubd.edu.bn/cgi/viewcontent.cgi?article=1620&context=journal
87. Neuro-symbolic AI – Wikipedia, accessed May 9, 2025, https://en.wikipedia.org/wiki/Neuro-symbolic_AI
88. Neuro-symbolic artificial intelligence | European Data Protection Supervisor, accessed May 9, 2025, https://www.edps.europa.eu/data-protection/technology-monitoring/techsonar/neuro-symbolic-artificial-intelligence_en
89. Import AI 279: Baidu adds knowledge to a language model; US military + AI; how China thinks about AI governance, accessed May 9, 2025, https://jack-clark.net/2022/01/10/import-ai-279-baidu-adds-knowledge-to-a-language-model-us-military-ai-how-china-thinks-about-ai-governance/
90. Neurosymbolic AI Disruption: Chinese Models Reshaping Global Tech – Data Pragmatist, accessed May 9, 2025, https://datapragmatist.com/p/neurosymbolic-ai-the-future-of-intelligent-machines
91. [2502.11269] Unlocking the Potential of Generative AI through Neuro-Symbolic Architectures: Benefits and Limitations – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2502.11269
92. Unlocking the Potential of Generative AI through Neuro-Symbolic Architectures – Benefits and Limitations – arXiv, accessed May 9, 2025, https://arxiv.org/html/2502.11269v1
93. Adaptable and Interpretable Neural Memory Over Symbolic …, accessed May 9, 2025, https://research.google/pubs/adaptable-and-interpretable-neural-memory-over-symbolic-knowledge/
94. (PDF) A Survey on Knowledge Graphs: Representation, Acquisition, and Applications, accessed May 9, 2025, https://www.researchgate.net/publication/351115157_A_Survey_on_Knowledge_Graphs_Representation_Acquisition_and_Applications
95. CoT-RAG: Integrating Chain of Thought and Retrieval-Augmented Generation to Enhance Reasoning in Large Language Models – arXiv, accessed May 9, 2025, https://arxiv.org/html/2504.13534v1
96. [2503.18213] A Study on Neuro-Symbolic Artificial Intelligence: Healthcare Perspectives, accessed May 9, 2025, https://arxiv.org/abs/2503.18213
97. [2408.16806] Physics-Informed Neural Networks and Extensions – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2408.16806
98. Understanding Physics-Informed Neural Networks: Techniques, Applications, Trends, and Challenges – MDPI, accessed May 9, 2025, https://www.mdpi.com/2673-2688/5/3/74
99. Evolutionary Optimization of Physics-Informed Neural Networks: Survey and Prospects, accessed May 9, 2025, https://arxiv.org/html/2501.06572v2
100. Physics-Informed AI – Autodesk Research, accessed May 9, 2025, https://www.research.autodesk.com/projects/physics-informed-ai-seminars/
101. Microsoft Fusion Summit explores how AI can accelerate fusion …, accessed May 9, 2025, https://www.microsoft.com/en-us/research/blog/microsoft-fusion-summit-explores-how-ai-can-accelerate-fusion-research/
102. Award for Inclusion Research recipients, accessed May 9, 2025, https://research.google/programs-and-events/award-for-inclusion-research-program/recipients/
103. Physics-informed AI – SLB, accessed May 9, 2025, https://www.slb.com/products-and-services/delivering-digital-at-scale/software/physics-informed-ai
104. [2503.12244] Quantum physics informed neural networks for multi-variable partial differential equations – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2503.12244
105. [2503.18181] Adaptive Physics-informed Neural Networks: A Survey – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2503.18181
106. [2501.06572] Evolutionary Optimization of Physics-Informed Neural Networks: Survey and Prospects – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2501.06572
107. (PDF) Machine Learning Through Physics–Informed Neural Networks: Progress and Challenges – ResearchGate, accessed May 9, 2025, https://www.researchgate.net/publication/378348600_Machine_Learning_Through_Physics-Informed_Neural_Networks_Progress_and_Challenges
108. A Comparison of Causal Inference Methods for Estimating Sales Lift – Google Research, accessed May 9, 2025, https://research.google/pubs/a-comparison-of-causal-inference-methods-for-estimating-sales-lift/
109. Quantum Causal Networks – Systems Engineering and Operations Research, accessed May 9, 2025, http://seor.vse.gmu.edu/~klaskey/papers/Laskey_QCN.pdf
110. What is Causal AI and Why Does it Matter? – Lifesight, accessed May 9, 2025, https://lifesight.io/blog/causal-ai/
111. Research Report 2024: Global Causal AI Market Set to Soar – GlobeNewswire, accessed May 9, 2025, https://www.globenewswire.com/news-release/2025/01/08/3006403/28124/en/Research-Report-2024-Global-Causal-AI-Market-Set-to-Soar-to-USD-456-8-Million-by-2030-Propelled-by-Demand-in-Decision-Making-Tools.html
112. AI as an intervention: improving clinical outcomes relies on a causal approach to AI development and validation – Oxford Academic, accessed May 9, 2025, https://academic.oup.com/jamia/advance-article/doi/10.1093/jamia/ocae301/7945189
113. (PDF) Improving Predictive Models With Causal Methods — Study Protocol – ResearchGate, accessed May 9, 2025, https://www.researchgate.net/publication/390409216_Improving_Predictive_Models_With_Causal_Methods_–_Study_Protocol
114. ChinaXiv:202401.00108v1, accessed May 9, 2025, https://chinaxiv.org/user/view.htm?uuid=5ba9df8c3bbf40a5a6dea9f4af187203&filetype=pdf
115. Neuro-Symbolic Visual Reasoning – Microsoft, accessed May 9, 2025, https://www.microsoft.com/en-us/research/wp-content/uploads/2020/07/Amizadeh-VQA.pdf
116. Machine Learning for Causal Inference – Google Books, accessed May 9, 2025, https://books.google.com/books/about/Machine_Learning_for_Causal_Inference.html?id=2Y3uzwEACAAJ
117. Robust agents learn causal world models – Google DeepMind, accessed May 9, 2025, https://deepmind.google/research/publications/49666/
118. [2503.05857] SYMBIOSIS: Systems Thinking and Machine Intelligence for Better Outcomes in Society – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2503.05857
119. Quantum Artificial Intelligence Lab – Wikipedia, accessed May 9, 2025, https://en.wikipedia.org/wiki/Quantum_Artificial_Intelligence_Lab
120. Quantum Supremacy Using a Programmable Superconducting Processor – Google Research, accessed May 9, 2025, https://research.google/blog/quantum-supremacy-using-a-programmable-superconducting-processor/
121. Google’s Sycamore Achieves Quantum Supremacy – PostQuantum.com, accessed May 9, 2025, https://postquantum.com/industry-news/google-sycamore/
122. Google Calls for Industry-Academia Alliance to Tackle Quantum Computing’s Scaling Challenges, accessed May 9, 2025, https://thequantuminsider.com/2025/05/05/google-calls-for-industry-academia-alliance-to-tackle-quantum-computings-scaling-challenges/
123. Microsoft Research Blog – Microsoft Research, accessed May 9, 2025, https://www.microsoft.com/en-us/research/blog/
124. Neuro-Symbolic Computation for Inference Reasoning – Microsoft Research, accessed May 9, 2025, https://www.microsoft.com/en-us/research/project/neural-symbolic-computation-in-natural-language-processing/
125. AI for Scientific Discovery – Microsoft’s Vision for the Future | Part 2 – EngageRM, accessed May 9, 2025, https://engagerm.com/ai-for-scientific-discovery-microsofts-vision-for-the-future-part-2/
126. YES AND: A Generative AI Multi-Agent Framework for Enhancing Diversity of Thought in Individual Ideation for Problem-Solving – Microsoft, accessed May 9, 2025, https://www.microsoft.com/en-us/research/wp-content/uploads/2025/03/CHI2025-Yes_And_An_AI_powered_problem_solving_framework_for_diversity_of_thought.pdf
127. Tracing the path to self-adapting AI agents – Microsoft Research, accessed May 9, 2025, https://www.microsoft.com/en-us/research/blog/tracing-the-path-to-self-adapting-ai-agents/
128. Advancing Multi-Agent Systems Through Model Context Protocol: Architecture, Implementation, and Applications – arXiv, accessed May 9, 2025, https://arxiv.org/html/2504.21030v1
129. Control versus Generativity: A Complex Adaptive Systems Perspective on Platforms, accessed May 9, 2025, https://aisel.aisnet.org/cgi/viewcontent.cgi?referer=&httpsredir=1&article=1225&context=icis2014
130. Supporting American leadership in quantum technology – Microsoft On the Issues, accessed May 9, 2025, https://blogs.microsoft.com/on-the-issues/2025/05/07/quantum-technology/
131. Microsoft’s Majorana 1 chip carves new path for quantum computing – Source, accessed May 9, 2025, https://news.microsoft.com/source/features/innovation/microsofts-majorana-1-chip-carves-new-path-for-quantum-computing/
132. Microsoft Quantum | Homepage, accessed May 9, 2025, https://quantum.microsoft.com/
133. What is Azure Quantum? – Azure Quantum | Microsoft Learn, accessed May 9, 2025, https://learn.microsoft.com/en-us/azure/quantum/overview-azure-quantum
134. Investing in American leadership in quantum technology: the next frontier in innovation – Microsoft On the Issues, accessed May 9, 2025, https://blogs.microsoft.com/on-the-issues/2025/04/27/investing-in-american-leadership-quantum/
135. Microsoft Story Labs: Station Q, accessed May 9, 2025, https://news.microsoft.com/stories/stationq/
136. Computer Vision – Baidu Research, accessed May 9, 2025, https://research.baidu.com/Publications?research%5B%5D=55&research%5B%5D=56&research%5B%5D=57&year=0
137. An advanced deep learning framework for simulating information propagation dynamics – Frontiers, accessed May 9, 2025, https://www.frontiersin.org/articles/10.3389/fphy.2025.1524104
138. Baidu’s ERNIE X1 and ERNIE 4.5 Models Explained | Built In, accessed May 9, 2025, https://builtin.com/artificial-intelligence/baidu-ernie-x1-ernie-4-5
139. PaddlePaddle: An Open-Source Deep Learning Framework – viso.ai, accessed May 9, 2025, https://viso.ai/deep-learning/paddlepaddle/
140. Baidu Presents Top 10 Frontier Technology Inventions of 2022, accessed May 9, 2025, https://research.baidu.com/Blog/index-view?id=171
141. Analysis and Simulation of the System Dynamics Model of the Peer Production-Taking Baidu Encyclopedia as an Example – Scientific Research Publishing, accessed May 9, 2025, https://www.scirp.org/journal/paperinformation?paperid=53108
142. Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking – arXiv, accessed May 9, 2025, https://arxiv.org/html/2502.13842v1
143. Intelligent Edge – Baidu AI Cloud, accessed May 9, 2025, https://intl.cloud.baidu.com/product/bie.html
144. Quantum Computing – Baidu Research, accessed May 9, 2025, https://research.baidu.com/Research_Areas/index-view?id=75
145. PaddlePaddle Receives Full Upgrade to Facilitate Integrated Innovation and Lower Thresholds – Baidu Research, accessed May 9, 2025, https://research.baidu.com/Blog/index-view?id=157
146. Baidu Announces Superconducting Quantum Computer and … – InfoQ, accessed May 9, 2025, https://www.infoq.com/news/2022/09/Baidu-quantum-computing-platform/
147. Baidu and Alibaba plan to quit quantum computing research – Physics World, accessed May 9, 2025, https://physicsworld.com/a/baidu-and-alibaba-plan-to-quit-quantum-computing-research/
148. China invests billions in quantum computing, race with US now neck-and-neck – SDx, accessed May 9, 2025, https://www.sdxcentral.com/analysis/china-invests-billions-in-quantum-computing-race-with-us-now-neck-and-neck/
149. Alibaba and Baidu close quantum research labs – DCD – Data Center Dynamics, accessed May 9, 2025, https://www.datacenterdynamics.com/en/news/alibaba-and-baidu-close-quantum-research-labs/
150. Baidu to donate quantum computing lab, equipment to Beijing institute – The Standard, accessed May 9, 2025, https://www.thestandard.com.hk/breaking-news/article/212041/Baidu-to-donate-quantum-computing-lab%2C-equipment-to-Beijing-institute
151. Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move – HPCwire, accessed May 9, 2025, https://www.hpcwire.com/2024/01/05/baidu-exits-quantum-closely-following-alibabas-earlier-move/
152. Baidu to donate quantum computing lab, equipment to Beijing institute, accessed May 9, 2025, https://m.economictimes.com/tech/technology/baidu-to-donate-quantum-computing-lab-equipment-to-beijing-institute/articleshow/106530830.cms
153. Baidu Ends Its Quantum Computing Research and Donates Lab and Equipment to the Beijing Academy of Quantum Information Sciences, accessed May 9, 2025, https://quantumcomputingreport.com/baidu-ends-its-quantum-computing-research-and-donates-lab-and-equipment-to-the-beijing-academy-of-quantum-information-sciences/
154. Baidu Quantum Donates Quantum Computing Lab To BAQIS, Echoing Alibaba’s Move In Extraordinary Move For China Tech, accessed May 9, 2025, https://quantumzeitgeist.com/baidu-donates-quantum-computing-lab-to-baqis-echoing-alibabas-move/
155. Form 20-F for Baidu INC filed 03/22/2023, accessed May 9, 2025, https://ir.baidu.com/static-files/eb1b6454-b72d-4ba6-9836-67804e8a2ed9
156. Form 20-F for Baidu INC filed 03/28/2025 – 百度, accessed May 9, 2025, https://ir.baidu.com/static-files/aa26913e-2e09-4ca2-be14-ce755b68adac
157. NVIDIA’s Quantum Strategy: Not Building the Computer, But the World That Enables It, accessed May 9, 2025, https://thequantuminsider.com/2025/03/21/nvidias-quantum-strategy-not-building-the-computer-but-the-world-that-enables-it/
158. Foundational AI Research Scientist Intel contract employee, accessed May 9, 2025, https://jobs.intel.com/en/job/arizona/foundational-ai-research-scientist-intel-contract-employee/41147/78681319168?source=levels.fyi&ref=levels.fyi&src=levels.fyi&utm_source=levels.fyi
159. Intel’s New Chip to Advance Silicon Spin Qubit Research for Quantum Computing, accessed May 9, 2025, https://newsroom.intel.com/new-technologies/quantum-computing-chip-to-advance-research
160. Pretraining Just Got Smarter: How Essential AI’s Research Redefines the Role of Reasoning in Model Development – AMD, accessed May 9, 2025, https://www.amd.com/en/developer/resources/technical-articles/how-essential-ai-research-redefines-reasoning-in-model-development.html
161. AMD Research and Advanced Development (RAD), accessed May 9, 2025, https://www.amd.com/en/corporate/research.html
162. Quantum Computing – AMD, accessed May 9, 2025, https://www.amd.com/en/solutions/quantum-computing.html
163. Arm: The Technology Foundation for AI Everywhere, accessed May 9, 2025, https://newsroom.arm.com/blog/arm-ai-everywhere
164. ARM’s Quantum Leap: Cortex Processor Tested at 3.3 Kelvin – AInvest, accessed May 9, 2025, https://www.ainvest.com/news/arm-s-quantum-leap-cortex-processor-tested-at-3-3-kelvin-24111010a493f9ecf439afcf/
165. Artificial Intelligence – Samsung Research, accessed May 9, 2025, https://research.samsung.com/artificial-intelligence
166. Qiskit in 2025: Pioneering Quantum Solutions Across Industries – Java Code Geeks, accessed May 9, 2025, https://www.javacodegeeks.com/2025/03/qiskit-in-2025-pioneering-quantum-solutions-across-industries.html
167. Quantum Key Distribution | TOSHIBA DIGITAL SOLUTIONS CORPORATION, accessed May 9, 2025, https://www.global.toshiba/ww/products-solutions/security-ict/qkd.html
168. Toshiba’s leading Cambridge Research Laboratory (CRL) doubles down on cutting edge embodied AI to bring AI into the real world., accessed May 9, 2025, https://www.toshiba.eu/cambridge-research-laboratory/news/toshibas-leading-cambridge-research-laboratory-crl-doubles-down-on-cutting-edge-embodied-ai-to-bring-ai-into-the-real-world/
169. Who We Are – NEC Laboratories Europe, accessed May 9, 2025, https://neclab.eu/about-us/who-we-are
170. Quantum Computing Initiatives – NEC Corporation, accessed May 9, 2025, https://www.nec.com/en/global/quantum-computing/index.html
171. Hitachi AI, Digital, Data Science – Job Matching Navigator, accessed May 9, 2025, https://www.hitachi.co.jp/recruit/en/newgraduate/jm-navi/ai_digital/index.html
172. Hitachi Launches $400 Million Fund Targeting Quantum, AI And Next-Gen Technology, accessed May 9, 2025, https://thequantuminsider.com/2025/02/07/hitachi-launches-400-million-fund-targeting-quantum-ai-and-next-gen-technology/
173. AI Researcher – Cisco AI Software and Platform Group – Cisco Careers, accessed May 9, 2025, https://jobs.cisco.com/jobs/ProjectDetail/AI-Researcher-Cisco-AI-Software-and-Platform-Group/1441439
174. Cisco Unveils Quantum Chip and Research Lab to Scale Quantum Computing – Techerati, accessed May 9, 2025, https://www.techerati.com/news-hub/cisco-unveils-quantum-chip-and-research-lab-to-scale-quantum-computing/
175. Broadcom and UC Berkeley Sky Computing Lab Expand Collaboration to Accelerate Open AI Ecosystems, accessed May 9, 2025, https://news.broadcom.com/artificial-intelligence/broadcom-uc-berkeley-sky-computing-lab-expand-collaboration-to-accelerate-open-ai-ecosystems
176. Broadcom Delivers Industry’s First Quantum Resistant Network Encryption, Enabling Real-time Ransomware Detection – GlobeNewswire, accessed May 9, 2025, https://www.globenewswire.com/news-release/2025/01/28/3016430/19933/en/Broadcom-Delivers-Industry-s-First-Quantum-Resistant-Network-Encryption-Enabling-Real-time-Ransomware-Detection.html
177. Broadcom Delivers Industry’s First Quantum Resistant Network Encryption, Enabling Real-time Ransomware Detection, accessed May 9, 2025, https://investors.broadcom.com/news-releases/news-release-details/broadcom-delivers-industrys-first-quantum-resistant-network
178. Panasonic AI | Panasonic Research and Development on Artificial Intelligence (AI) – Panasonic, accessed May 9, 2025, https://tech-ai.panasonic.com/en/
179. Research Areas – Quantinuum, accessed May 9, 2025, https://www.quantinuum.com/research/research-areas?979754fc_page=5
180. Quantinuum Announces Plans to Build a New Quantum R&D Center in New Mexico, Anchoring the State’s Quantum Technology Revolution, accessed May 9, 2025, https://www.quantinuum.com/press-releases/quantinuum-announces-plans-to-build-a-new-quantum-r-d-center-in-new-mexico-anchoring-the-states-quantum-technology-revolution
181. Quantum Origin | Quantum Random Number Generator – Quantinuum, accessed May 9, 2025, https://www.quantinuum.com/products-solutions/quantum-origin
182. Technical White Paper, accessed May 9, 2025, https://cdn.prod.website-files.com/669960f53cd73aedb80c8eea/66da564bce3f80e2827db453_6670683dc0f4a03b3ac9def3_quantum-origin-technical-whitepaper_r03.pdf
183. Quantinuum’s ‘Quantum Origin’ Becomes First Software Quantum Random Number Generator to Achieve NIST Validation, accessed May 9, 2025, https://www.quantinuum.com/press-releases/quantinuums-quantum-origin-becomes-first-software-quantum-random-number-generator-to-achieve-nist-validation
184. What is Quantum Origin by Quantinuum? | Solution Overview – EM360Tech, accessed May 9, 2025, https://em360tech.com/tech-articles/what-quantum-origin-quantinuum-solution-overview
185. Pioneering the next era of computing – Quantinuum, accessed May 9, 2025, https://www.quantinuum.com/company/about
186. National Quantum Initiative, accessed May 9, 2025, https://www.quantum.gov/
187. Quantum Information Science and Technology White Paper FINAL VERSION 2022.07.26 – National Archives, accessed May 9, 2025, https://www.archives.gov/files/records-mgmt/policy/nara-quantum-information-science-and-technology-whitepaper.pdf
188. Quantum Information Science Applications Roadmap, accessed May 9, 2025, https://quantum.gov/wp-content/uploads/2024/12/DOE_QIS_Roadmap_Final.pdf
189. Who’s Ahead, Who’s Behind, and Where We Are Headed Next in the U.S.-China Technology Competition – SCSP.ai, accessed May 9, 2025, https://www.scsp.ai/wp-content/uploads/2025/01/Gaps-Analysis-2025-Report.pdf
190. Generative AI: The Future of Innovation Power – Special Competitive Studies Project, accessed May 9, 2025, https://www.scsp.ai/wp-content/uploads/2023/09/GenAI-web.pdf
191. Beyond DeepSeek: How China’s AI Ecosystem Fuels Breakthroughs – CNAS, accessed May 9, 2025, https://www.cnas.org/publications/commentary/beyond-deepseek-how-chinas-ai-ecosystem-fuels-breakthroughs
192. Chapter 3 – U.S.-China Competition in Emerging Technologies, accessed May 9, 2025, https://www.uscc.gov/sites/default/files/2024-11/Chapter_3–U.S.-China_Competition_in_Emerging_Technologies.pdf
193. China’s Quantum Computing and Quantum Technology Initiatives – PostQuantum.com, accessed May 9, 2025, https://postquantum.com/quantum-computing/china-quantum/
194. 量子ベンチャー企業の動向, accessed May 9, 2025, https://www8.cao.go.jp/cstp/ryoshigijutsu/kento_wg/6kai/sanko3.pdf
195. Who’s Leading the Quantum Race? Market Share & Investment Stats | PatentPC, accessed May 9, 2025, https://patentpc.com/blog/whos-leading-the-quantum-race-market-share-investment-stats
196. AI at CERN and SKA (17-18 September 2018) · Indico, accessed May 9, 2025, https://indico.cern.ch/event/745580/timetable/?print=1&view=standard
197. State of UK Deep Tech 2024 – Royal Academy of Engineering, accessed May 9, 2025, https://raeng.org.uk/media/by0l0j4h/state-of-uk-deep-tech-2024.pdf
198. events – Fritz Haber Institute – Max-Planck-Gesellschaft, accessed May 9, 2025, https://www.fhi.mpg.de/60031/events
199. Publications | Max Planck Institute for Intelligent Systems, accessed May 9, 2025, https://is.mpg.de/publications?publication_type%5B%5D=Book+Chapter
200. Mathematisches Forschungsinstitut Oberwolfach Foundations and new horizons for causal inference – EMS Press, accessed May 9, 2025, https://ems.press/content/serial-article-files/46804
201. (PDF) Second-order control of complex systems with correlated synthetic data, accessed May 9, 2025, https://www.researchgate.net/publication/337407701_Second-order_control_of_complex_systems_with_correlated_synthetic_data
202. An interdisciplinary bibliometric analysis of models for land-use and transport interactions – arXiv, accessed May 9, 2025, https://arxiv.org/pdf/2102.13501
203. Bioinformatic analysis – UKRI gateway, accessed May 9, 2025, https://gtr.ukri.org/projects?ref=BBS%2FE%2FD%2F20211551&pn=11&fetchSize=10&selectedSortableField=title&selectedSortOrder=DESC
204. Proceedings of the 2nd International Conference on Software Engineering and Machine Learning – ACE Applied and Computational Engineering, accessed May 9, 2025, https://www.ewadirect.com/proceedings/ace/volumes/vol/76/367.pdf
205. arxiv.org, accessed May 9, 2025, https://arxiv.org/pdf/2210.16751
206. Toward Creation of a System Architecture for Quantum Computers | NTT R&D Website, accessed May 9, 2025, https://www.rd.ntt/e/research/JN202309_23082.html
207. NTT Scientists Present Breakthrough Research on AI Deep Learning at ICLR 2025, accessed May 9, 2025, https://ntt-research.com/ntt-scientists-present-breakthrough-research-on-ai-deep-learning-at-iclr-2025/
208. Energy and AI – NET, accessed May 9, 2025, https://iea.blob.core.windows.net/assets/b3a8b37f-32d1-4873-9eca-31cec5895264/EnergyandAI.pdf
209. Causal Inference Meets Deep Learning: A Comprehensive Survey …, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC11384545/
210. Machine Learning Nov 2023 – arXiv, accessed May 9, 2025, http://arxiv.org/list/cs.LG/2023-11?skip=650&show=1000
211. MBZUAI Faculty Brochure | PDF | Artificial Intelligence – Scribd, accessed May 9, 2025, https://www.scribd.com/document/823719509/MBZUAI-Faculty-Brochure
212. SPLASH 2021 – OOPSLA – Conferences – Researchr, accessed May 9, 2025, https://conf.researchr.org/track/splash-2021/splash-2021-oopsla
213. A*STAR CFAR – Agency for Science, Technology and Research (A*STAR), accessed May 9, 2025, https://www.a-star.edu.sg/cfar
214. Personnel – Bioinformatics & Data Science – Binus Research, accessed May 9, 2025, https://research.binus.ac.id/bdsrc/personnel/
215. Bioinformatics & Data Science Research Center (BDSRC) – Binus Research, accessed May 9, 2025, https://research.binus.ac.id/bdsrc/
216. Quantum materials, built by AI robot | Pritzker School of Molecular Engineering, accessed May 9, 2025, https://pme.uchicago.edu/news/quantum-materials-built-ai-robot
217. van Dijk Lab @Yale, accessed May 9, 2025, https://www.vandijklab.org/
218. David van Dijk, PhD, MSc, BSc – Yale School of Medicine, accessed May 9, 2025, https://medicine.yale.edu/profile/david-vandijk/
219. Scaling Large Language Models for Next-Generation Single-Cell Analysis – ResearchGate, accessed May 9, 2025, https://www.researchgate.net/publication/390980280_Scaling_Large_Language_Models_for_Next-Generation_Single-Cell_Analysis
220. vandijklab/brainlm · Hugging Face, accessed May 9, 2025, https://huggingface.co/vandijklab/brainlm
221. Exploring the Frontiers of LLMs in Psychological Applications: A Comprehensive Review – arXiv, accessed May 9, 2025, https://arxiv.org/pdf/2401.01519
222. Hierarchical Causal Discovery From Large-Scale Observed Variables – IEEE Computer Society, accessed May 9, 2025, https://www.computer.org/csdl/journal/tk/5555/01/10877758/247rPYmFOus
223. Causal deep learning – van der Schaar Lab, accessed May 9, 2025, https://www.vanderschaar-lab.com/causal-deep-learning-research-pillar/
224. Research team – van der Schaar Lab, accessed May 9, 2025, https://www.vanderschaar-lab.com/research-team-old/
225. Machine Learning Mar 2023 – arXiv, accessed May 9, 2025, http://arxiv.org/list/cs.LG/2023-03?skip=50&show=2000
226. Publications // van der Schaar Lab, accessed May 9, 2025, https://www.vanderschaar-lab.com/publications/
227. Unconventional quantum computer can help unveil fundamental mysteries of the universe | Waterloo News, accessed May 9, 2025, https://uwaterloo.ca/news/unconventional-quantum-computer-can-help-unveil-fundamental
228. Nonparametric Decomposition of Quasi-periodic Time Series for Change-point Detection, accessed May 9, 2025, https://www.researchgate.net/publication/290440053_Nonparametric_Decomposition_of_Quasi-periodic_Time_Series_for_Change-point_Detection
229. Prof. dr. P.M.A. (Peter) Sloot – University of Amsterdam, accessed May 9, 2025, https://www.uva.nl/en/profile/s/l/p.m.a.sloot/p.m.a.sloot.html
230. Arxiv今日论文| 2025-03-05 – 闲记算法, accessed May 9, 2025, http://lonepatient.top/2025/03/05/arxiv_papers_2025-03-05
231. Exploring Non-Classical correlations in Causal Scenarios – White Rose eTheses Online, accessed May 9, 2025, https://etheses.whiterose.ac.uk/id/eprint/36307/1/PhD_Thesis_Submitted_Version.pdf
232. China’s Generative AI Ecosystem in 2024: Rising Investment and Expectations | The National Bureau of Asian Research (NBR), accessed May 9, 2025, https://www.nbr.org/publication/chinas-generative-ai-ecosystem-in-2024-rising-investment-and-expectations/
233. China’s Origin Wukong quantum computer becomes world’s first to …, accessed May 9, 2025, https://www.chinadaily.com.cn/a/202504/07/WS67f3a2bba3104d9fd381df3e.html
234. AI in China: Top Companies, Innovations & ChatGPT Rivals – HPRT, accessed May 9, 2025, https://www.hprt.com/blog/AI-in-China-Top-Companies-Innovations-ChatGPT-Rivals.html
235. Quantum AI: Harnessing Quantum Computing for AI (2024 Update) – PostQuantum.com, accessed May 9, 2025, https://postquantum.com/quantum-ai/quantum-ai-qai/
236. Redefining Intelligence: The 7 Best Quantum Machine Learning Tools – BlueQubit, accessed May 9, 2025, https://www.bluequbit.io/quantum-machine-learning-tools
237. Quantum and AI: The Future of Computing or Just Hype?, accessed May 9, 2025, https://meetiqm.com/blog/quantum-and-ai-the-future-of-computing-or-just-hype/
238. Your Quick Guide to Quantum and AI: The Future of Computing or Just Hype?, accessed May 9, 2025, https://meetiqm.com/blog/quantum-ai-the-future-of-computing-or-just-hype/
239. Quantum-Enhanced LLM Efficient Fine Tuning – arXiv, accessed May 9, 2025, https://arxiv.org/html/2503.12790v1
240. [2503.12790] Quantum-Enhanced LLM Efficient Fine Tuning – arXiv, accessed May 9, 2025, https://arxiv.org/abs/2503.12790
241. (PDF) Quantum-Enhanced LLM Efficient Fine Tuning – ResearchGate, accessed May 9, 2025, https://www.researchgate.net/publication/389917092_Quantum-Enhanced_LLM_Efficient_Fine_Tuning/download
242. accessed December 31, 1969, https://arxiv.org/pdf/2503.12790.pdf
243. Quantum Sensing – Key Use Cases – PostQuantum.com, accessed May 9, 2025, https://postquantum.com/quantum-sensing/quantum-sensing-use-cases/
244. Quantum Thesis – Main Sequence Ventures, accessed May 9, 2025, https://www.mseq.vc/msv-blog/quantum-thesis
245. Quantum Sensing: Comparing the United States and China – The International Institute for Strategic Studies, accessed May 9, 2025, https://www.iiss.org/globalassets/media-library—content–migration/files/research-papers/2024/02/iiss_quantum-sensing_022024.pdf
246. AI and Quantum Sensing: A Perfect Synergy – PostQuantum.com, accessed May 9, 2025, https://postquantum.com/quantum-sensing/ai-quantum-sensing/
247. A Molecular Approach to Quantum Sensing – PMC – PubMed Central, accessed May 9, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC8161477/
248. OriginQ Cloud – 本源量子, accessed May 9, 2025, https://m.originqc.com.cn/en
249. Top Quantum Computing companies in China by Total Equity | StockViz.com, accessed May 9, 2025, https://stockviz.com/en/ranking-of-china-stocks-in-quantum-computing-by-total-equity
250. Roadmap – Alice & Bob, accessed May 9, 2025, https://alice-bob.com/roadmap/
251. AI for Quantum Error Correction: A Comprehensive Guide to Using Artificial Intelligence to Improve Quantum Error Correction – The Quantum Insider, accessed May 9, 2025, https://thequantuminsider.com/2025/01/06/ai-for-quantum-error-correction-a-comprehensive-guide-to-using-artificial-intelligence-to-improve-quantum-error-correction/
252. Demystifying amortized causal discovery with transformers – OpenReview, accessed May 9, 2025, https://openreview.net/forum?id=lQYi2zeDyh
253. Responsible governance of generative AI: conceptualizing GenAI as complex adaptive systems | Policy and Society | Oxford Academic, accessed May 9, 2025, https://academic.oup.com/policyandsociety/article/44/1/38/7965776
254. Microsoft New Future of Work Report 2024, accessed May 9, 2025, https://www.microsoft.com/en-us/research/uploads/prod/2024/12/NFWReport2024_1.27.2025.pdf