Step 21: assistant

text

The claim is assessed as **doubtful** (confidence 0.85), with a credence of about 2% that it comes true.

- **Wording.** I reworded the claim to "All major AI labs will have fully automated their AI research by the end of 2027." The original "will conduct fully automated AI research" could also mean each lab merely runs some AI-run projects, which is a different and much likelier proposition.
- **Why some chance remains.** 

text

METR's November 2026 report put Anthropic's AI progress at roughly double its earlier rate and credited about 80% of that to AI

text

, though 

text

most of the evidence METR found most informative was redacted by Anthropic

text

. 

text

The AI Futures Project's authors now all expect a superhuman AI researcher sometime in 2027, absent a deliberate slowdown.

text

- **Why "all labs" is very unlikely.** A universal claim fails if any one lab misses the date.
  - 

text

OpenAI's own date for an automated AI researcher is March 2028

text

, after the deadline.
  - 

text

Google DeepMind's October 2026 safety report found no significant end-to-end acceleration of its AI R&D.

text

  - 

text

Zuckerberg wrote in September 2026 that most compute should go to serving people, not to a race toward recursive self-improvement.

text

  - 

text

Anthropic reports no measured part of its AI R&D as fully autonomous, with large gaps remaining in Claude's judgment about goals.

text

  - 

text

Altman and Musk publicly backed Amodei's September 2026 call to pace frontier development.

text

- **How the 2% is built.** Roughly a one in four chance the leading lab has the capability by the end of 2027, about half of that for actually running research without human researchers, and about one in eight that every other major lab does the same by the same date.

Structure added: three arguments (one for, two against) and two new subclaims, both unassessed and carrying only my preliminary credences. One is that at least one lab fully automates its research by the end of 2027 (0.13); the other is that the major labs' frontier capabilities are only months apart (0.3). Importance stays at 0.3, with contestation set to 0.3.

Limits of this pass:
- No source was found asserting or denying the universal form, so no instances are recorded.
- Most lab statements were read through news coverage and search extracts, not the primary documents.
- I did not open the two neighbouring claims I linked as see-also; their wording came from the matcher's output.
- The two identity checks for the new subclaims cost about $0.97, which is why I did no further matching or primary-source reading.

The run used $2.26 of the $5.00 budget.