Science Agents

AI Use-cases for Science

Literature

• alphaXiv | Explore: Understand arXiv papers

LLM extract data from papers

• 2024-14: From text to insight: large language models for chemical data extraction

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
• 2026-01: Deep contrastive learning enables genome-wide virtual screening

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

• 2025-01: Accurate predictions on small data with a tabular foundation model (code)

Symbolic Regression

• 2024-09: Symbolic Regression with a Learned Concept Library

Literature Discovery

• FutureHouse
  • hasanyone
  • PaperQA2
• 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

• Bioptimus
• EvolutionaryScale

AI/ML Methods in Science

• 2025-07: Synthetic Scientific Image Generation with VAE, GAN, and Diffusion Model Architectures

Imaging

• 2025-05: Behind the Noise: Conformal Quantile Regression Reveals Emergent Representations (blog: Behind the Noise)

Materials

• 2024-12: Crystal structure generation with autoregressive large language modeling
• 2025-03: All-atom Diffusion Transformers: Unified generative modelling of molecules and materials
• 2022-11: Training-Free Active Learning Framework in Materials Science with Large Language Models

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
• 2026-01: Advancing regulatory variant effect prediction with AlphaGenome

Medicine

See: AI_Agents#Medicine

Successes

• 2025-02: Site-Decorated Model for Unconventional Frustrated Magnets: Ultranarrow Phase Crossover and Spin Reversal Transition

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
• 2026-01: Using Interpretability to Identify a Novel Class of Alzheimer's Biomarkers

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
  • 2025-03: CURIE: Evaluating LLMs On Multitask Scientific Long Context Understanding and Reasoning
  • 2024-07: SPIQA: A Dataset for Multimodal Question Answering on Scientific Papers
  • 2024-10: FEABench: Evaluating Language Models on Real World Physics Reasoning Ability
• 2026-02: Edison: LABBench 2

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

Challenges

• 2026-01: Why LLMs Aren't Scientists Yet: Lessons from Four Autonomous Research Attempts

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
• 2026-02: PaperBanana: Automating Academic Illustration for AI Scientists

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
• 2025-12: Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows
• 2026-01: SciSciGPT: advancing human–AI collaboration in the science of science

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

• 2025-08: Operationalizing Serendipity: Multi-Agent AI Workflows for Enhanced Materials Characterization with Theory-in-the-Loop

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

• 2025-07: BioMARS: A Multi-Agent Robotic System for Autonomous Biological Experiments

Physics

• 2025-12: PhysMaster: Building an Autonomous AI Physicist for Theoretical and Computational Physics Research

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
• 2026-02: Accelerating Scientific Research with Gemini: Case Studies and Common Techniques

Related Tools

Literature Search

• Perplexity
• ArXival

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

• 2025-03: Rxn-INSIGHT: fast chemical reaction analysis using bond-electron matrices (docs)

Science Datasets

• Google Dataset Search
• Awesome Materials & Chemistry Datasets
• NIST Jarvis (simulations)

Genuine Discoveries

• 2025-11: Early science acceleration experiments with GPT-5
• 2025-12: AI can obviously create new knowledge - But maybe not new concepts
• Math:
  • 2023-07: Faster sorting algorithms discovered using deep reinforcement learning
  • 2025-06: AlphaEvolve: A coding agent for scientific and algorithmic discovery
  • 2025-11: Mathematical exploration and discovery at scale
  • 2025-11: Olympiad-level formal mathematical reasoning with reinforcement learning
  • 2025-12: Extremal descendant integrals on moduli spaces of curves: An inequality discovered and proved in collaboration with AI
  • AI Solving Erdős Problems:
    • 2026-01: Erdős Problem #728 and #729 solved by Aristotle using ChatGPT 5.2 Pro
    • 2026-01: Erdős Problem #397 solved by Neel Somani using ChatGPT 5.2 Pro
    • 2026-01: Erdős Problem #205 solved by Aristotle using ChatGPT 5.2 Pro
    • 2026-01: Erdős Problem #281 solved by Neel Somani using ChatGPT 5.2 Pro
    • 2026-01: Google DeepMind: Irrationality of rapidly converging series: a problem of Erdős and Graham
      • Erdős Problem #1051 solved by Google DeepMind Aletheia agent
    • 2026-01: Google DeepMind: Semi-Autonomous Mathematics Discovery with Gemini: A Case Study on the Erdős Problems
      • Attempted 700 problems, solved 13 open Erdős problems: 5 novel autonomous solutions, 8 through existing literature.
  • 2026-01: The motivic class of the space of genus 0 maps to the flag variety
• Physics assistance:
  • 2025-03: Exact solution of the frustrated Potts model with next-nearest-neighbor interactions in one dimension via AI bootstrapping (Weiguo Yin)
  • 2025-12: Relativistic covariance and nonlinear quantum mechanics: Tomonaga-Schwinger analysis
    • Steve Hsu, Theoretical Physics with Generative AI
• Literature exploration:
  • 2025-11: Kosmos: An AI Scientist for Autonomous Discovery (Edison)
    • Nucleotide metabolism in hypothermia
    • Determinant of perovskite solar-cell failure
    • Log-normal connectivity in neural networks
    • SOD2 as driver of myocardial fibrosis
    • Protective variant of SSR1 in type 2 diabetes
    • Temporal ordering in Alzheimer’s disease
    • Mechanism of neuron vulnerability in aging
• Bio design:
  • 2023-07: De novo design of protein structure and function with RFdiffusion
  • 2025-11: Atomically accurate de novo design of antibodies with RFdiffusion
  • 2025-11: AlphaFold: Five years of impact
  • 2026-01: Using Interpretability to Identify a Novel Class of Alzheimer's Biomarkers
• Material Discovery:
  • 2023-11: Scaling deep learning for materials discovery

See Also

• AI agents
• Nanobot.chat: Intelligent AI for the labnetwork @ mtl.mit.edu forum