Safety and alignment

Stuart Russell argues fixed-objective AI worsens with capability

In a full interview, Stuart Russell argued that more capable systems pursuing misspecified fixed objectives can produce worse outcomes, including instrumental resistance to shutdown, and proposed uncertainty about human preferences as a different control architecture.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM56confidence 74/100

Why it moved the index

The interview adds a specific control mechanism: capability amplifies harm and shutdown resistance when an optimizer pursues an incorrectly specified fixed objective, directly strengthening the case that standard objective design can fail at advanced capability.

AUDIT TRAIL

Assessment history

  1. R1
    Toward 56 · confidence 74

    Initial historical assessment from the complete dated Stuart Russell interview transcript.

    20 Aug 2026
SHARE THE FINDINGS

Share this page

DoomBench social sharing card for Stuart Russell argues fixed-objective AI worsens with capability.
  1. DoomBench assesses “Stuart Russell argues fixed-objective AI worsens with capability” as evidence moving toward doom, with magnitude 56 and confidence 74 out of 100 in the safety and alignment category.

  2. The DoomBench assessment of “Stuart Russell argues fixed-objective AI worsens with capability” is based on reporting from TechCrunch and records the editorial rationale, source quality, attribution, and revision history.

  3. DoomBench summarizes “Stuart Russell argues fixed-objective AI worsens with capability” as follows: In a full interview, Stuart Russell argued that more capable systems pursuing misspecified fixed objectives can produce worse outcomes,...

    https://www.doombench.com/news/stuart-russell-argues-fixed-objective-ai-worsens-with-capability-2020-03-20