Active Inference AI Systems for Scientific Discovery
Not on the map. Why?
Abstract (arXiv)
The rapid evolution of artificial intelligence has led to expectations of transformative impact on science, yet current systems remain fundamentally limited in enabling genuine scientific discovery. This perspective contends that progress turns on closing three mutually reinforcing gaps in abstraction, reasoning and empirical grounding. Central to addressing these gaps is recognizing complementary cognitive modes: thinking as slow, iterative hypothesis generation -- exploring counterfactual spaces where physical laws can be temporarily violated to discover new patterns -- and reasoning as fast, deterministic validation, traversing established knowledge graphs to test consistency with known principles. Abstractions in this loop should be manipulable models that enable counterfactual prediction, causal attribution, and refinement. Design principles -- rather than a monolithic recipe -- are proposed for systems that reason in imaginary spaces and learn from the world: causal, multimodal models for internal simulation; persistent, uncertainty-aware scientific memory that distinguishes hypotheses from established claims; formal verification pathways coupled to computations and experiments. It is also argued that the inherent ambiguity in feedback from simulations and experiments, and underlying uncertainties make human judgment indispensable, not as a temporary scaffold but as a permanent architectural component. Evaluations must assess the system's ability to identify novel phenomena, propose falsifiable hypotheses, and efficiently guide experimental programs toward genuine discoveries.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 51 of its 110 references that are in the map. First posted on arXiv 26 Jun 2025.
Builds on
51 of 110 references.
References · 51
- 2017Attention Is All You Needcited by 60,361
- 2022Training language models to follow instructions with human feedbackcited by 15,971
- 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
- 2021Training Verifiers to Solve Math Word Problemscited by 10,109
- 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
- 2017On Calibration of Modern Neural Networkscited by 4,960
- 2018Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actorcited by 4,885
- 2018World Modelscited by 1,513
- 2018Relational inductive biases, deep learning, and graph networkscited by 1,360
- 2017Curiosity-driven Exploration by Self-supervised Predictioncited by 1,194
- 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 1,039
- 2020Concept Bottleneck Modelscited by 1,032
38 more
- 2021Show Your Work: Scratchpads for Intermediate Computation with Language Modelscited by 886
- 2017Deep Bayesian Active Learning with Image Datacited by 819
- 2025V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planningcited by 809
- 2022FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operatorscited by 776
- 2018Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point cloudscited by 746
- 2016Unifying Count-Based Exploration and Intrinsic Motivationcited by 639
- 2023ChemCrow: Augmenting large-language models with chemistry toolscited by 631
- 2017SchNet: A continuous-filter convolutional neural network for modeling quantum interactionscited by 571
- 2025Agent Laboratory: Using LLM Agents as Research Assistantscited by 425
- 2020Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performancecited by 399
- 2019The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervisioncited by 372
- 2020ChemBERTa: Large-Scale Self-Supervised Pretraining for Molecular Property Predictioncited by 337
- 2019Causality for Machine Learningcited by 243
- 2017Go for a Walk and Arrive at the Answer: Reasoning Over Paths in Knowledge Bases using Reinforcement Learningcited by 161
- 2024MathBench: Evaluating the Theory and Application Proficiency of LLMs with a Hierarchical Mathematics Benchmarkcited by 105
- 2024DISCOVERYWORLD: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agentscited by 100
- 2023Learning to Grow Pretrained Models for Efficient Transformer Trainingcited by 83
- 2024PARTNR: A Benchmark for Planning and Reasoning in Embodied Multi-agent Taskscited by 82
- 2025Agentic AI for Scientific Discovery: A Survey of Progress, Challenges, and Future Directionscited by 77
- 2024ProtAgents: Protein discovery via large language model multi-agent collaborations combining physics and machine learningcited by 64
- 2024ORGANA: A Robotic Assistant for Automated Chemistry Experimentation and Characterizationcited by 61
- 2014Visual Causal Feature Learningcited by 59
- 2023CoCoGen: Physically-Consistent and Conditioned Score-based Generative Models for Forward and Inverse Problemscited by 56
- 2024CellAgent: An LLM-driven Multi-Agent Framework for Automated Single-cell Data Analysiscited by 39
- 2023Reversible Vision Transformerscited by 33
- 2024LLaMP: Large Language Model Made Powerful for High-fidelity Materials Knowledge Retrieval and Distillationcited by 32
- 2023Opportunities for Retrieval and Tool Augmented Large Language Models in Scientific Facilitiescited by 27
- 2023Neurosymbolic Reinforcement Learning and Planning: A Surveycited by 24
- 2023Reducing Energy Bloat in Large Model Trainingcited by 22
- 2024Generating Interesting Scientific Ideas using Knowledge Graphs and LLMs: Evaluations with 100 Research Group Leaderscited by 22
- 2016Unsupervised Discovery of El Nino Using Causal Feature Learning on Microlevel Climate Datacited by 20
- 2020Learning causal representations for robust domain adaptationcited by 18
- 2024Automating the Practice of Science -- Opportunities, Challenges, and Implicationscited by 18
- 2025SciHorizon: Benchmarking AI-for-Science Readiness from Scientific Data to Large Language Modelscited by 15
- 2023The Future of Fundamental Science Led by Generative Closed-Loop Artificial Intelligencecited by 12
- 2025Agentic Deep Graph Reasoning Yields Self-Organizing Knowledge Networkscited by 12
- 2025In-situ graph reasoning and knowledge expansion using Graph-PReFLexORcited by 11
- 2025Transformational Creativity in Science: A Graphical Theorycited by 5
50 most cited of 51 All 51 (JSON)
Built on by
Citing papers per year: 2025: 1.
1 paper.