Difference between revisions of "Science Agents"
KevinYager (talk | contribs) (→Genuine Discoveries) |
KevinYager (talk | contribs) (→Genuine Discoveries) |
||
| (8 intermediate revisions by the same user not shown) | |||
| Line 58: | Line 58: | ||
===Science Foundation Models=== | ===Science Foundation Models=== | ||
* 2025-08: [https://arxiv.org/abs/2508.15763 Intern-S1: A Scientific Multimodal Foundation Model] | * 2025-08: [https://arxiv.org/abs/2508.15763 Intern-S1: A Scientific Multimodal Foundation Model] | ||
| + | * 2025-11: [https://pubs.aip.org/aip/jcp/article/163/18/184110/3372267/A-foundation-model-for-atomistic-materials A foundation model for atomistic materials chemistry] | ||
| + | * 2025-11: [https://arxiv.org/abs/2511.15684 Walrus: A Cross-Domain Foundation Model for Continuum Dynamics] | ||
===Regression (Data Fitting)=== | ===Regression (Data Fitting)=== | ||
| Line 184: | Line 186: | ||
* 2025-07: [https://arxiv.org/abs/2507.14267 DREAMS: Density Functional Theory Based Research Engine for Agentic Materials Simulation] | * 2025-07: [https://arxiv.org/abs/2507.14267 DREAMS: Density Functional Theory Based Research Engine for Agentic Materials Simulation] | ||
* 2025-11: [https://arxiv.org/abs/2511.02824 Kosmos: An AI Scientist for Autonomous Discovery] | * 2025-11: [https://arxiv.org/abs/2511.02824 Kosmos: An AI Scientist for Autonomous Discovery] | ||
| + | * 2025-11: [https://arxiv.org/abs/2511.08151 SciAgent: A Unified Multi-Agent System for Generalistic Scientific Reasoning] | ||
==Science Multi-Agent Setups== | ==Science Multi-Agent Setups== | ||
| Line 252: | Line 255: | ||
=Genuine Discoveries= | =Genuine Discoveries= | ||
| − | * 2025- | + | * 2025-11: [https://cdn.openai.com/pdf/4a25f921-e4e0-479a-9b38-5367b47e8fd0/early-science-acceleration-experiments-with-gpt-5.pdf Early science acceleration experiments with GPT-5] |
| − | * 2025-11: [https://arxiv.org/abs/2511.02864 Mathematical exploration and discovery at scale] | + | |
| − | * 2025-11: [https://arxiv.org/abs/2511.02824 Kosmos: An AI Scientist for Autonomous Discovery] | + | * '''Math:''' |
| − | ** [https://platform.edisonscientific.com/kosmos/c4bdef64-5e9b-43b9-a365-592dd1ed7587 Nucleotide metabolism in hypothermia] | + | ** 2023-07: [https://www.nature.com/articles/s41586-023-06004-9?utm_source=chatgpt.com Faster sorting algorithms discovered using deep reinforcement learning] |
| − | ** [https://platform.edisonscientific.com/kosmos/1fdbf827-be65-4d97-9b66-bf0da600091a Determinant of perovskite solar-cell failure] | + | ** 2025-11: [https://arxiv.org/abs/2511.02864 Mathematical exploration and discovery at scale] |
| − | ** [https://platform.edisonscientific.com/kosmos/4fb3fbdb-c449-4064-9aa6-ff4ec53131d8 Log-normal connectivity in neural networks] | + | ** 2025-11: [https://www.nature.com/articles/s41586-025-09833-y Olympiad-level formal mathematical reasoning with reinforcement learning] |
| − | ** [https://platform.edisonscientific.com/kosmos/c6849232-5858-4634-adf5-83780afbe3db SOD2 as driver of myocardial fibrosis] | + | * '''Physics assistance:''' |
| − | ** [https://platform.edisonscientific.com/kosmos/abac07da-a6bb-458f-b0ba-ef08f1be617e Protective variant of SSR1 in type 2 diabetes] | + | ** 2025-03: [https://arxiv.org/abs/2503.23758 Exact solution of the frustrated Potts model with next-nearest-neighbor interactions in one dimension via AI bootstrapping] |
| − | ** [https://platform.edisonscientific.com/kosmos/a770052b-2334-4bbe-b086-5149e0f03d99 Temporal ordering in Alzheimer’s disease] | + | * '''Literature exploration:''' |
| − | ** [https://platform.edisonscientific.com/kosmos/28c427d2-be31-48b5-b272-28d5a1e3ea5c Mechanism of neuron vulnerability in aging] | + | ** 2025-11: [https://arxiv.org/abs/2511.02824 Kosmos: An AI Scientist for Autonomous Discovery] |
| + | *** [https://platform.edisonscientific.com/kosmos/c4bdef64-5e9b-43b9-a365-592dd1ed7587 Nucleotide metabolism in hypothermia] | ||
| + | *** [https://platform.edisonscientific.com/kosmos/1fdbf827-be65-4d97-9b66-bf0da600091a Determinant of perovskite solar-cell failure] | ||
| + | *** [https://platform.edisonscientific.com/kosmos/4fb3fbdb-c449-4064-9aa6-ff4ec53131d8 Log-normal connectivity in neural networks] | ||
| + | *** [https://platform.edisonscientific.com/kosmos/c6849232-5858-4634-adf5-83780afbe3db SOD2 as driver of myocardial fibrosis] | ||
| + | *** [https://platform.edisonscientific.com/kosmos/abac07da-a6bb-458f-b0ba-ef08f1be617e Protective variant of SSR1 in type 2 diabetes] | ||
| + | *** [https://platform.edisonscientific.com/kosmos/a770052b-2334-4bbe-b086-5149e0f03d99 Temporal ordering in Alzheimer’s disease] | ||
| + | *** [https://platform.edisonscientific.com/kosmos/28c427d2-be31-48b5-b272-28d5a1e3ea5c Mechanism of neuron vulnerability in aging] | ||
| + | * '''Bio design:''' | ||
| + | ** 2023-07: [https://www.nature.com/articles/s41586-023-06415-8 De novo design of protein structure and function with RFdiffusion] | ||
| + | ** 2025-11: [https://www.nature.com/articles/s41586-025-09721-5 Atomically accurate de novo design of antibodies with RFdiffusion] | ||
| + | ** 2025-11: [https://x.com/GoogleDeepMind/status/1993350293703016451?s=20 ] | ||
| + | * '''Material Discovery:''' | ||
| + | ** 2023-11: [https://deepmind.google/blog/alphafold-five-years-of-impact/ AlphaFold: Five years of impact] | ||
=See Also= | =See Also= | ||
* [[AI agents]] | * [[AI agents]] | ||
* [https://nanobot.chat/ Nanobot.chat]: Intelligent AI for the labnetwork @ mtl.mit.edu forum | * [https://nanobot.chat/ Nanobot.chat]: Intelligent AI for the labnetwork @ mtl.mit.edu forum | ||
Latest revision as of 10:03, 26 November 2025
Contents
- 1 AI Use-cases for Science
- 2 Science Benchmarks
- 3 Science Agents
- 4 AI Science Systems
- 5 Impact of AI in Science
- 6 Related Tools
- 7 Science Datasets
- 8 Genuine Discoveries
- 9 See Also
AI Use-cases for Science
Literature
- alphaXiv | Explore: Understand arXiv papers
LLM extract data from papers
AI finding links in literature
- 2019-07: Unsupervised word embeddings capture latent knowledge from materials science literature
- 2024-11: Large language models surpass human experts in predicting neuroscience results
(Pre) Generate Articles
- 2022-12: Re3: Generating Longer Stories With Recursive Reprompting and Revision
- 2023-03: English essays: Artificial intelligence (AI) technology in OpenAI ChatGPT application: A review of ChatGPT in writing English essay
- 2023-01: Journalism: Collaborating With ChatGPT: Considering the Implications of Generative Artificial Intelligence for Journalism and Media Education
- 2023-07: Science writing: Artificial intelligence in scientific writing: a friend or a foe?
- 2024-02: Wikipedia style: Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models
- 2024-02: LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs (code)
- 2024-08: Scientific papers: The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
- 2024-09: PaperQA2: Language Models Achieve Superhuman Synthesis of Scientific Knowledge (𝕏 post, code)
- 2025-03: Reasoning to Learn from Latent Thoughts
- 2025-03: WikiAutoGen: Towards Multi-Modal Wikipedia-Style Article Generation
- 2025-04: Sleep-time Compute: Beyond Inference Scaling at Test-time
Explanation
- 2025-02: TheoremExplainAgent: Towards Multimodal Explanations for LLM Theorem Understanding (preprint)
- 2025-04: Do Two AI Scientists Agree?
Autonomous Ideation
- 2024-04: ResearchAgent: Iterative Research Idea Generation over Scientific Literature with Large Language Models
- 2024-09: Mining Causality: AI-Assisted Search for Instrumental Variables
- 2024-12: Thinking Fast and Laterally: Multi-Agentic Approach for Reasoning about Uncertain Emerging Events
- 2024-12: LLMs can realize combinatorial creativity: generating creative ideas via LLMs for scientific research
- 2024-12: LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context
- 2025-01: Hypothesis Generation for Materials Discovery and Design Using Goal-Driven and Constraint-Guided LLM Agents
- 2025-02: Agentic Deep Graph Reasoning Yields Self-Organizing Knowledge Networks
- 2025-06: Predicting Empirical AI Research Outcomes with Language Models
- 2025-06: The Ideation-Execution Gap: Execution Outcomes of LLM-Generated versus Human Research Ideas
Adapting LLMs to Science
- 2023-06: Domain-specific chatbots for science using embeddings
- 2024-10: Personalization of Large Language Models: A Survey
- 2024-11: Adapting While Learning: Grounding LLMs for Scientific Problems with Intelligent Tool Usage Adaptation
AI/LLM Control of Scientific Instruments/Facilities
- 2023-12: Opportunities for retrieval and tool augmented large language models in scientific facilities
- 2023-12: Virtual Scientific Companion for Synchrotron Beamlines: A Prototype
- 2023-12: Autonomous chemical research with large language models
- 2024-01: Synergizing Human Expertise and AI Efficiency with Language Model for Microscopy Operation and Automated Experiment Design
- 2024-06: From Text to Test: AI-Generated Control Software for Materials Science Instruments
- 2024-12: VISION: A Modular AI Assistant for Natural Human-Instrument Interaction at Scientific User Facilities
- 2025-01: Large language models for human-machine collaborative particle accelerator tuning through natural language
- 2025-04: Operating Robotic Laboratories with Large Language Models and Teachable Agents
AI/ML Methods tailored to Science
Science Foundation Models
- 2025-08: Intern-S1: A Scientific Multimodal Foundation Model
- 2025-11: A foundation model for atomistic materials chemistry
- 2025-11: Walrus: A Cross-Domain Foundation Model for Continuum Dynamics
Regression (Data Fitting)
- 2024-06: Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data: training on (x,y) pairs enables inferring underlying function (define it in code, invert it, compose it)
- 2024-12: OmniPred: Language Models as Universal Regressors
Tabular Classification/Regression
Symbolic Regression
Literature Discovery
- FutureHouse
- Lumina
- Automated-AI-Web-Researcher-Ollama
- 2025-01: Search-o1: Agentic Search-Enhanced Large Reasoning Models (project, code)
Commercial
- Sakana AI
- Cusp AI: Materials/AI
- Lila AI: Life sciences
- Radical AI: Material simulation/design
- Autoscience (Carl)
- Periodic Labs
- Edison Scientific (drug discovery, spinoff from FutureHouse)
Bio
AI/ML Methods in Science
Imaging
- 2025-05: Behind the Noise: Conformal Quantile Regression Reveals Emergent Representations (blog: Behind the Noise)
Materials
- 2024-12: [https://www.nature.com/articles/s41467-024-54639-7 Crystal structure generation with autoregressive large language modeling
- 2025-03: All-atom Diffusion Transformers: Unified generative modelling of molecules and materials
Chemistry
- 2025-01: Large language models for reticular chemistry
- 2025-02: Image-based generation for molecule design with SketchMol
- 2025-02: Large language models for scientific discovery in molecular property prediction
- 2025-03: Vant AI Neo-1: atomistic foundation model (small molecules, proteins, etc.)
- 2025-04: Compositional Flows for 3D Molecule and Synthesis Pathway Co-design
- 2025-07: General purpose models for the chemical sciences
- 2025-11: ChemTorch: A Deep Learning Framework for Benchmarking and Developing Chemical Reaction Property Prediction Models
Biology
- 2018: AlphaFold
- 2021-07: AlphaFold 2
- 2024-05: AlphaFold 3
- 2023-03: Evolutionary-scale prediction of atomic-level protein structure with a language model (ESMFold)
- 2023-11: Illuminating protein space with a programmable generative model
- 2024-11: Sequence modeling and design from molecular to genome scale with Evo (Evo)
- 2025-01: Targeting protein–ligand neosurfaces with a generalizable deep learning tool (Chroma)
- 2025-01: Simulating 500 million years of evolution with a language model (ESM 3 model)
- 2025-02: Genome modeling and design across all domains of life with Evo 2
- 2025-02: Exploring the structural changes driving protein function with BioEmu-1
- 2025-02: Protein Large Language Models: A Comprehensive Survey
- 2025-03: Vant AI Neo-1: atomistic foundation model (small molecules, proteins, etc.)
- 2025-03: Lyra: An Efficient and Expressive Subquadratic Architecture for Modeling Biological Sequences
- 2025-08: RosettaFold 3: Accelerating Biomolecular Modeling with AtomWorks and RF3
- 2025-09: Generative design of novel bacteriophages with genome language models
- 2025-10: Strengthening nucleic acid biosecurity screening against generative protein design tools
Medicine
See: AI_Agents#Medicine
Successes
AI/ML Methods co-opted for Science
Mechanistic Interpretability
Train large model on science data. Then apply mechanistic interpretability (e.g. sparse autoencoders, SAE) to the feature/activation space.
- Mechanistic interpretability for protein language models (visualizer, code, SAE)
- Markov Bio: Through a Glass Darkly: Mechanistic Interpretability as the Bridge to End-to-End Biology (quick description, background info on recent bio progress)
- 2023-01: Tracr: Compiled Transformers as a Laboratory for Interpretability (code)
- 2024-10: An X-Ray Is Worth 15 Features: Sparse Autoencoders for Interpretable Radiology Report Generation
- 2024-12: Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models
- 2024-12: InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders
- 2025-01: Insights on Galaxy Evolution from Interpretable Sparse Feature Networks
- 2025-02: From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models
- 2025-02: Interpreting Evo 2: Arc Institute's Next-Generation Genomic Foundation Model
Uncertainty
- 2024-10: entropix: Entropy Based Sampling and Parallel CoT Decoding
- 2024-10: Taming Overconfidence in LLMs: Reward Calibration in RLHF
Science Benchmarks
- 2024-07: SciCode: A Research Coding Benchmark Curated by Scientists (project)
- 2024-11: AidanBench: Evaluating Novel Idea Generation on Open-Ended Questions (code)
- 2024-12: LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context
- 2025-01: Humanity's Last Exam
- ScienceAgentBench
- 2025-02: EAIRA: Establishing a Methodology for Evaluating AI Models as Scientific Research Assistants
- 2025-03: BixBench: Novel hypotheses (accept/reject)
- 2025-04: Google: Evaluating progress of LLMs on scientific problem-solving
Science Agents
Reviews
- 2024-10: Empowering biomedical discovery with AI agents
- 2025-01: A review of large language models and autonomous agents in chemistry (github)
- 2025-07: AI4Research: A Survey of Artificial Intelligence for Scientific Research
- 2025-08: From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery
Specific
- 2024-01-13: ORGANA: A Robotic Assistant for Automated Chemistry Experimentation and Characterization (video)
- 2024-06-19: LLMatDesign: Autonomous Materials Discovery with Large Language Models
- 2024-08-12: Sakana AI: AI Scientist; The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery (code)
- 2024-09-09: SciAgents: Automating scientific discovery through multi-agent intelligent graph reasoning (code)
- 2024-09-11: PaperQA2: Language Models Achieve Superhuman Synthesis of Scientific Knowledge (𝕏 post, code)
- 2024-10-17: Rapid and Automated Alloy Design with Graph Neural Network-Powered LLM-Driven Multi-Agent Systems
- 2024-10-28: Large Language Model-Guided Prediction Toward Quantum Materials Synthesis
- 2024-12-06: The Virtual Lab: AI Agents Design New SARS-CoV-2 Nanobodies with Experimental Validation (writeup: Virtual lab powered by ‘AI scientists’ super-charges biomedical research: Could human–AI collaborations be the future of interdisciplinary studies?)
- 2024-12-30: Aviary: training language agents on challenging scientific tasks
- See also: AI Agents > Deep Research
- 2025-04-08: Sakana: The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search (code)
- 2025-07: DREAMS: Density Functional Theory Based Research Engine for Agentic Materials Simulation
- 2025-11: Kosmos: An AI Scientist for Autonomous Discovery
- 2025-11: SciAgent: A Unified Multi-Agent System for Generalistic Scientific Reasoning
Science Multi-Agent Setups
- 2025-01: Agent Laboratory: Using LLM Agents as Research Assistants
- 2025-04: Coordinated AI agents for advancing healthcare (pdf)
AI Science Systems
- 2025-01: Dolphin: Closed-loop Open-ended Auto-research through Thinking, Practice, and Feedback
- 2025-01: Hypothesis Generation for Materials Discovery and Design Using Goal-Driven and Constraint-Guided LLM Agents
- 2025-02: Towards an AI co-scientist (Google blog post: Accelerating scientific breakthroughs with an AI co-scientist)
- 2025-06: The Discovery Engine
- 2025-07: Benchmarking the Discovery Engine (blog)
- 2025-07: Autonomous Scientific Discovery Through Hierarchical AI Scientist Systems
Inorganic Materials Discovery
- 2023-11: Scaling deep learning for materials discovery
- 2023-11: An autonomous laboratory for the accelerated synthesis of novel materials
- 2024-09: HoneyComb: A Flexible LLM-Based Agent System for Materials Science
- 2024-10: Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models (code, datasets, checkpoints, blogpost)
- 2025-01: A generative model for inorganic materials design
- 2025-04: System of Agentic AI for the Discovery of Metal-Organic Frameworks
- 2025-05: The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models
Materials Characterization
Chemistry
- 2023-12: Autonomous chemical research with large language models (Coscientist)
- 2024-09: PNNL ChemAIst V0.2
- 2024-11: An automatic end-to-end chemical synthesis development platform powered by large language models
- 2025-06: Training a Scientific Reasoning Model for Chemistry
- 2025-06: ChemGraph: An Agentic Framework for Computational Chemistry Workflows (code)
Bio
LLMs Optimized for Science
- 2022-11: Galactica: A Large Language Model for Science
- 2024-12: Crystal structure generation with autoregressive large language modeling
- 2025-02: MatterChat: A Multi-Modal LLM for Material Science
- 2025-03: OmniScience: A Domain-Specialized LLM for Scientific Reasoning and Discovery
- 2025-03: Google TxGemma (2B, 9B, 27B): drug development
Impact of AI in Science
- 2024-11:
Artificial Intelligence, Scientific Discovery, and Product Innovation- 2025-05: Retraction: Assuring an accurate research record
- 2025-02: Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation
Related Tools
Literature Search
Data Visualization
- 2024-10: Microsoft Data Formulator: Create Rich Visualization with AI iteratively (video, code)
- Julius AI: Analyze your data with computational AI
Generative
- 2025-03: StarVector 1B, 8B: text or image to SVG
Chemistry
Science Datasets
- Google Dataset Search
- Awesome Materials & Chemistry Datasets
- NIST Jarvis (simulations)
Genuine Discoveries
- Math:
- Physics assistance:
- Literature exploration:
- Bio design:
- Material Discovery:
- 2023-11: AlphaFold: Five years of impact
See Also
- AI agents
- Nanobot.chat: Intelligent AI for the labnetwork @ mtl.mit.edu forum
