OmniScientist: An Omni-Modal Omni-Discipline AI Sci...

OmniScientist: An Omni-Modal Omni-Discipline AI Sci... | AI Research

Key Takeaways

OmniScientist is an end-to-end AI system designed to automate the entire scientific research lifecycle, from generating hypotheses to writing final manuscrip...
Recent advances in foundation models have enabled AI scientists to automate increasingly complete research workflows, from hypothesis generation and code execution to manuscript preparation.
Yet workflow coverage alone does not provide access to the full evidence on which scientific discovery depends.
Existing systems typically reason over text, code, labels, or precomputed summaries, leaving scientifically decisive spatial, temporal, cross-channel, and procedural relations unavailable to the agent.
We introduce OmniScientist, an end-to-end, omni-modal AI scientist that conducts multidisciplinary research directly from heterogeneous raw evidence.

Paper AbstractExpand

Recent advances in foundation models have enabled AI scientists to automate increasingly complete research workflows, from hypothesis generation and code execution to manuscript preparation. Yet workflow coverage alone does not provide access to the full evidence on which scientific discovery depends. Existing systems typically reason over text, code, labels, or precomputed summaries, leaving scientifically decisive spatial, temporal, cross-channel, and procedural relations unavailable to the agent. We introduce OmniScientist, an end-to-end, omni-modal AI scientist that conducts multidisciplinary research directly from heterogeneous raw evidence. A perception layer and 3 autonomous agents for ideation, experiment, and writeup operate within a deterministic pipeline, allowing observations to shape research questions, experimental decisions, and final claims throughout the research lifecycle. By running idea, rigour, and claim checks in code, the system enforces novelty screening, statistical validity, execution provenance, and numerical traceability. We evaluate OmniScientist on 36 real-data cases spanning 5 discipline families, 4 families of scientific evidence, and modalities including images, signals, audio, video, 3-D structures, trajectories, tables, formulae, and graphs. The system completes the full path from raw data to a compiled manuscript in all 36 cases and achieves a mean overall paper score of 6.3 with the reference reasoning backbone. In paired comparisons against a blind variant that receives only precomputed scalar features, direct perception improves all 7 evaluation dimensions and wins 85% of head-to-head judgments. These results show that lifecycle-wide perception is essential for evidence-grounded scientific discovery and provides a practical path toward broadly capable AI scientists.

OmniScientist is an end-to-end AI system designed to automate the entire scientific research lifecycle, from generating hypotheses to writing final manuscripts. Unlike previous AI scientists that rely on precomputed summaries or text-based data, OmniScientist processes raw, heterogeneous scientific evidence—such as images, waveforms, 3-D structures, and signals—directly. By maintaining access to these raw artifacts throughout the research process, the system preserves spatial, temporal, and procedural relations that are often lost when data is reduced to text or scalar features.

How the System Works

The framework operates through a deterministic pipeline that coordinates a perception layer with three autonomous agents: ideation, experiment, and writeup.

Perception Layer: This layer categorizes raw evidence into four families—perceptual, symbolic, quantitative-statistical, and procedural—and uses specific tools to inspect them. It prioritizes native numerical analysis (like trend points or FFT peaks) and only renders visual plots when spatial or structural patterns are necessary.
Autonomous Agents: Each agent uses a ReAct loop to interleave reasoning and action. The ideation agent formulates hypotheses, the experiment agent executes code and inspects results, and the writeup agent compiles the final paper.
Control Pipeline: To ensure scientific integrity, the system enforces code-based checks at every stage. These checks screen for novelty, verify statistical validity, ensure execution provenance, and prevent "HARKing" (hypothesizing after results are known).

Research Scope and Evidence

The authors, Bobo Li, Hao Fei, Tianjie Ju, Mong-Li Lee, and Wynne Hsu, evaluated OmniScientist on a 36-case demonstration suite. These cases span five discipline families—Physical Sciences, Earth & Space, Life & Medical, Agricultural & Ecological, and Engineering & Information—and utilize diverse modalities including audio, video, graphs, and formulae. The system is designed to be discipline-agnostic; adding a new field requires only a new specification file rather than custom research code.

Performance and Impact

In testing, OmniScientist successfully completed the full research path from raw data to a compiled manuscript in all 36 cases. The generated papers achieved a mean score of 6.3 out of a potential maximum on a 7-dimensional rubric.
To isolate the impact of direct perception, the researchers compared the full system against a "blind" variant that only received precomputed scalar features. The full system outperformed the blind variant in all seven evaluation dimensions and won 85% of head-to-head judgments. This suggests that lifecycle-wide perception is a critical component for evidence-grounded scientific discovery, as it allows the agent to notice anomalies and support claims that would otherwise be obscured by data reduction.

Considerations

The authors note that while the system automates the research workflow, it requires a control structure to manage the risks associated with a broader search space, such as data leakage and repeated testing. The framework addresses these through its deterministic pipeline and code-enforced constraints, which are intended to guarantee that all reported claims are traceable and supported by the raw evidence.