The Need for Verification in AI-Driven Scientific Discovery
2025/09/01 by Cristina Cornelio, Cornelio, Cristina, Takuya Ito +7 · 4 citations
Decision Sciences · Materials Science · Medicine · #Artificial Intelligence (cs.AI) #Artificial Intelligence in Healthcare and Education #FOS: Computer and information sciences #Machine Learning in Materials Science #Scientific Computing and Data Management
paper · doi:10.48550/arxiv.2509.01398
openalex publication_date 2025/09/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Abstract
Artificial intelligence (AI) is transforming the practice of science. Machine learning and large language models (LLMs) can generate hypotheses at a scale and speed far exceeding traditional methods, offering the potential to accelerate discovery across diverse fields. However, the abundance of hypotheses introduces a critical challenge: without scalable and reliable mechanisms for verification, scientific progress risks being hindered rather than being advanced. In this article, we trace the historical development of scientific discovery, examine how AI is reshaping established practices for scientific discovery, and review the principal approaches, ranging from data-driven methods and knowledge-aware neural architectures to symbolic reasoning frameworks and LLM agents. While these systems can uncover patterns and propose candidate laws, their scientific value ultimately depends on rigorous and transparent verification, which we argue must be the cornerstone of AI-assisted discovery.
Citations
- When is a System Discoverable from Data? Discovery Requires Chaos
- The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies
- AlphaEvolve: A coding agent for scientific and algorithmic discovery
- Spurious Rewards: Rethinking Training Signals in RLVR
- Applications of Modular Co-Design for De Novo 3D Molecule Generation
- Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens
- Scientific Hypothesis Generation and Validation: Methods, Datasets, and Future Directions
- LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models
- Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces!
- The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search
- Agentic AI for Scientific Discovery: A Survey of Progress, Challenges, and Future Directions
- A Neural Symbolic Model for Space Physics
- Accelerating scientific discovery with Co-Scientist
- Nature Language Model: Deciphering the Language of Nature for Scientific Discovery
- Discovering Symbolic Cognitive Models from Human and Animal Behavior
- Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2
- Towards Scientific Discovery with Generative AI: Progress, Opportunities, and Challenges
- Proposing and solving olympiad geometry with guided tree search
- MOOSE-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses
- The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
- AtomAgents: Alloy design and discovery through physics-aware multi-modal multi-agent artificial intelligence
- KAN: Kolmogorov-Arnold Networks
- LLM-SR: Scientific Equation Discovery via Programming with Large Language Models
- Can large language models reason and plan?
- BioXP-0.5B: Explainable Medical-AI via RL-GRPO
- Scaling deep learning for materials discovery
- Large Language Models are Zero Shot Hypothesis Proposers
- Lie Point Symmetry and Physics Informed Networks
- NEWTON: Are Large Language Models Capable of Physical Reasoning?
- SNIP: Bridging Mathematical Symbolic and Numeric Realms with Unified Pre-training
- Large Language Models for Automated Open-domain Scientific Hypotheses Discovery
- SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models
- Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks
- ChemCrow: Augmenting large-language models with chemistry tools
- Galactica: A Large Language Model for Science
- Draft, Sketch, and Prove: Guiding Formal Theorem Provers with Informal Proofs
- Neurosymbolic Programming for Science
- Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering
- Semantic Probabilistic Layers for Neuro-Symbolic Learning
- End-to-end symbolic regression with transformers
- Biological Sequence Design with GFlowNets
- Scientific Machine Learning through Physics-Informed Neural Networks: Where we are and What's next
- Scientific Machine Learning Through Physics–Informed Neural Networks: Where we are and What’s Next
- Neuro-Symbolic Inductive Logic Programming with Logical Neural Networks
- Highly accurate protein structure prediction with AlphaFold
- Coordinate Independent Convolutional Networks -- Isometry and Gauge Equivariant Convolutions on Riemannian Manifolds
- Measuring Mathematical Problem Solving With the MATH Dataset
- E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials
- Extracting Training Data from Large Language Models
- Neural Networks Enhancement with Logical Knowledge
- Discovering Symbolic Models from Deep Learning with Inductive Biases
- LGML: Logic Guided Machine Learning
- Are Ideas Getting Harder to Find?
- Learning Compositional Rules via Neural Program Synthesis
- Lagrangian Neural Networks
- Integrating Deep Learning with Logic Fusion for Information Extraction
- Bayesian Symbolic Regression
- DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators
- DRUM: End-To-End Differentiable Rule Mining On Knowledge Graphs
- Fine-Tuning Language Models from Human Preferences
- Embedding Symbolic Knowledge into Deep Networks
- A Logic-Driven Framework for Consistency of Neural Models
- Augmenting Neural Networks with First-order Logic
- Hamiltonian Neural Networks
- AI Feynman: a Physics-Inspired Method for Symbolic Regression
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
- Inductive Learning of Answer Set Programs from Noisy Examples
- HOUDINI: Lifelong Learning as Program Synthesis
- A Semantic Loss Function for Deep Learning with Symbolic Knowledge
- Learning Explanatory Rules from Noisy Data
- Neuro-Symbolic Program Synthesis
- On Identifiability of Nonlinear ODE Models and Applications in Viral Dynamics
- Evaluating Sakana's AI Scientist: Bold Claims, Mixed Results, and a Promising Future?
Cited by
Related