AI systems lack executive function, not scale: four manager-layer strategies to fix it
Designer-Material633 · reddit · 2026-11-19
The author argues, via cognitive science, that models looping, abandoning goals, and failing to commit is a control problem, not a compute problem — current models are workers without managers.
- No goal maintenance under noise: models treat every tool output and retrieved doc as equally salient, diluting the original objective (humans use prefrontal goal representations).
- No hierarchical control: flat reactive decision-making means vague prompts like "explore options" cause unbounded search; specific prompts like "commit to option A and justify" produce completion.
- No working-memory pruning: humans actively discard irrelevant context; models accumulate it indefinitely, creating interference rather than intelligence.
Four implementable strategies: (1) restate the single goal at the top of every run and reject outputs that don't serve it; (2) structure prompts as Goal → Plan → Execute → Report, forcing commitment before action; (3) prune context aggressively after each step, keeping only relevant summaries; (4) add an inhibition layer stating what NOT to do (e.g. "don't browse more than 3 sources"). Conclusion: we don't need bigger models, we need better managers.
Related event: Reddit Post: AI's Missing Piece Is Executive Function, Not Compute(3 posts)→
More from AGI Musings
- Stanford professor: AI takes students from problem to working prototype in 80 minutes — HariSubramonyam · 2026-11-19
- Berggruet Winner Thomas Nagel Dismisses AI Consciousness Talk as 'Just Ridiculous' — Dr_Atoosa · 2026-11-19
- Dwarkesh Claims China Is Only 12 Hours — Not 12 Months — Behind Frontier AI Labs — teortaxesTex · 2026-11-19
- Researcher calls METR/Anthropic '2x speedup' RSI analysis statistically dubious — jessi_cata · 2026-11-19
- Physician: AI competition now measured in weeks, GPT-6 versions 7 days apart — ramos_casals · 2026-11-19
- Why pro-handover rhetoric fails: people don't trust ASI's loyalty to labs — 1a3orn · 2026-11-19