Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Kevin Murphy
• Venue: arXiv: cs.AI
• Published: 2026-08-10
• Collected: 2026-08-11
• URL: http://arxiv.org/abs/2608.09696v1
• PDF: https://arxiv.org/pdf/2608.09696v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The text introduces the Model Discovery Agent (MDA), a system designed to efficiently discover mechanistic, causal models from minimal interventions using a large language model (LLM) and Bayesian methods. The MDA proposes candidate structures, identifies them through experiments, and uses these identified mechanisms for further predictions. It outperforms the state-of-the-art in three different benchmarks: physics, chemistry, and biology.

What the Document Says

• The Model Discovery Agent (MDA) is presented, which combines a large language model (LLM) with Bayesian methods to discover mechanistic world models from few interventions.
• The MDA operates in the M-open setting, where the truth may lie outside the current hypothesis class, and a predictive check flags inadequacy, causing the proposer to expand the hypothesis space.
• The discovery step proposes a mechanism, which is identified by designed experiments in the design step.
• Identified mechanisms improve predictions and enable further discoveries from residuals.
• MDA sets a new state-of-the-art (SOTA) in data-efficient model learning and reliable interventional forecasting ability across three benchmarks: physics, chemistry, and biology.

Key Concepts Mentioned

• Model Discovery Agent (MDA)
• Large Language Model (LLM)
• Mechanistic world models
• Interventional 'what if' questions
• Bayesian machinery
• Sequential Monte Carlo (SMC)
• Simulation-based inference (SBI)
• Value-of-information (VoI)
• Experiment design
• Data efficiency
• M-open setting
• Predictive check
• State-of-the-art (SOTA)

Claims Made by the Document

• MDA discovers latent mechanistic world models from few interventions
• MDA sets a new SOTA in data-efficient model learning and reliable interventional forecasting ability

Evidence Used

• The text describes the MDA's components and processes for discovering and designing mechanisms
• Results from three benchmarks demonstrate MDA's performance

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Predicting the answer to interventional ``what if'' questions --- the outcome of an action never taken --- requires a \emph{mechanistic}, causal model, not a curve fit; and learning such a model requires \emph{experiments}, because passive data leaves its mechanisms unidentified. Experiments are expensive, so the central problem is \emph{data efficiency}. We present the Model Discovery Agent (MDA), which couples a large language model (LLM), used as a \emph{proposer} of candidate structures, with standard Bayesian machinery --- sequential Monte Carlo (SMC) for parameter and structure posteriors, simulation-based inference (SBI) for intractable likelihoods, and value-of-information (VoI) for experiment design --- to discover latent mechanistic world models from few interventions. MDA operates in the M-open setting: when the truth lies outside the current hypothesis class, a predictive check flags the inadequacy and the proposer expands the hypothesis space with a new model whose parameters are then identified by designed experiments. We show that \emph{discovery and design reinforce}: the design step identifies the mechanism the discovery step proposes, and the identified mechanism improves predictions, enabling further discoveries from the remaining unexplained residuals. On three different benchmarks --- covering physics (\DPbench, \citep{wiemann2026discoverphysics}), chemistry (\CHEMbench, \citep{kabra2026autoscilab}) and biology (\HHbench, a new partially observed single-neuron electrophysiology benchmark we create) --- we show that MDA sets a new SOTA in terms of data-efficient model learning and reliable interventional forecasting ability.