---
type: "post"
id: "the-use-of-ai-in-science-raises-questions-3f80f1"
slug: "the-use-of-ai-in-science-raises-questions-3f80f1"
title: "The use of AI in science raises questions on several levels."
description: "The MIT FutureTech study reports a rise in answer agreement on cutting-edge theory questions, from 2.8% for o1 to 34.6% for GPT-5.5, alongside an estimated $15 per successful match."
retrieval_nugget: "The MIT FutureTech study reports a rise in answer agreement on cutting-edge theory questions, from 2.8% for o1 to 34.6% for GPT-5.5, alongside an estimated $15 per successful match."
published_at: "2026-10-21"
updated_at: "2026-10-21"
record_date: "2026-10-21"
date_kind: "published_at"
topics: ["evals","knowledge-work","models"]
source_urls: ["https://futuretech.mit.edu/publication/ai-disruption-and-falling-costs-in-theory-research"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:the-use-of-ai-in-science-raises-questions-3f80f1"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:29730310ba2359de8ef0533a0ae4829107579957c00b69647b3aadc7c769b554","reviewed_at":"2026-10-21T12:05:20.638Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"The study suggests that frontier models are making some theoretical research questions cheaper to investigate, while measuring answer agreement rather than autonomous scientific discovery.","observed_facts":[{"text":"The study used 254 recent arXiv theory questions across seven domains in mathematics, physics, and economics, with human-authored answers unavailable to models when answering.","source_urls":["https://futuretech.mit.edu/publication/ai-disruption-and-falling-costs-in-theory-research"]},{"text":"The reported share of problems with answers matching the human authors rose from 2.8% for o1 to 34.6% for GPT-5.5.","source_urls":["https://futuretech.mit.edu/publication/ai-disruption-and-falling-costs-in-theory-research"]},{"text":"The researchers estimate approximately $5 per attempt and $15 per successful match for GPT-5.5.","source_urls":["https://futuretech.mit.edu/publication/ai-disruption-and-falling-costs-in-theory-research"]}],"mechanism":"The researchers compare model answers with human-authored ground truth for recent theory questions and estimate the cost per attempt and per matching answer.","why_now":"The study spans model releases from December 2024 to April 2026, making the reported change a longitudinal comparison rather than a single-model snapshot.","implications":["Researchers can treat these systems as potentially lower-cost aids for bounded theory questions, while retaining human review and validation of the underlying work.","The reported cost and answer-match rate provide a test-set-specific basis for comparing utility, not a universal measure of scientific productivity."],"evidence_boundary":"The supplied source reports agreement against known human answers across seven domains; it does not show that models generated novel results, completed an end-to-end research workflow, or generalize to all scientific tasks.","watch_conditions":["Future evaluations should test whether gains persist on independently held-out questions and translate into verified research outputs."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied for this specific benchmark and cost comparison, so it begins a distinct analytical record."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/the-use-of-ai-in-science-raises-questions-3f80f1/","markdown":"https://newruntime.com/posts/the-use-of-ai-in-science-raises-questions-3f80f1.md","json":"https://newruntime.com/posts/the-use-of-ai-in-science-raises-questions-3f80f1.json"}
related_patterns: ["agent-economics-moves-to-completed-work"]
---

# The use of AI in science raises questions on several levels.

## Retrieval answer

The MIT FutureTech study reports a rise in answer agreement on cutting-edge theory questions, from 2.8% for o1 to 34.6% for GPT-5.5, alongside an estimated $15 per successful match.

The MIT FutureTech study reports a rise in answer agreement on cutting-edge theory questions, from 2.8% for o1 to 34.6% for GPT-5.5, alongside an estimated $15 per successful match. These are results on a defined test set, not evidence that a model independently completed a research project.

Thesis: The study suggests that frontier models are making some theoretical research questions cheaper to investigate, while measuring answer agreement rather than autonomous scientific discovery. Mechanism: The researchers compare model answers with human-authored ground truth for recent theory questions and estimate the cost per attempt and per matching answer. Why now: The study spans model releases from December 2024 to April 2026, making the reported change a longitudinal comparison rather than a single-model snapshot. The result gives teams a way to consider both performance and cost, but only within the tested question set.

Implication: Researchers can treat these systems as potentially lower-cost aids for bounded theory questions, while retaining human review and validation of the underlying work. Evidence boundary: The supplied source reports agreement against known human answers across seven domains; it does not show that models generated novel results, completed an end-to-end research workflow, or generalize to all scientific tasks. Watch condition: Future evaluations should test whether gains persist on independently held-out questions and translate into verified research outputs. The paper’s figures are a promising signal, not a general measure of scientific productivity.

This starts a distinct analytical record because no prior New Runtime record is supplied for this specific benchmark and cost comparison.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Evals](https://newruntime.com/topics/evals.md): Explore the evals topic hub.
- topic hub: [Knowledge Work](https://newruntime.com/topics/knowledge-work.md): Explore the knowledge-work topic hub.
- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)