Thinking fast and slow in AI: The role of metacognition (2021)
- ID
- 29445
- Status
- summarized
- Published
- 28 Sep 2026, 11:23 AM
- Fetched
- 29 Sep 2026, 2:53 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://arxiv.org/abs/2110.01834
- Source URL
- https://hnrss.org/best
Summary
- Score
- 5.0
- Created
- 29 Sep 2026, 2:54 AM
- Tags
- Audience
- developersai_ml_learnersai_agent_usersvibe_coders
What happened
This is the 2021 arXiv paper "Thinking Fast and Slow in AI: the Role of Metacognition" by Marianna Bergamaschi Ganapini, Murray Campbell, Francesco Fabiano, Lior Horesh, Jon Lenchner, Andrea Loreggia, Nicholas Mattei, Francesca Rossi, Biplav Srivastava and Kristen Brent Venable, resurfaced on Hacker News (163 points, 67 comments). It proposes a multi-agent architecture where incoming problems are handled either by "system 1" fast agents that react from past experience, or by "system 2" slow agents deliberately activated when optimal solutions are needed beyond what system 1 can deliver, with both backed by a world model and a model of "self" holding past actions and solver skills. The text contains no benchmarks, code, datasets, or results — it is a position/architecture argument, not an implementation.
Why it matters
The concrete thing here is the escalation trigger: the paper's design puts the decision to spend slow reasoning on a separate "self" model that tracks past actions and solver skills, rather than routing everything through one model. If you are building agents, that is the same lever as choosing between a cheap fast model and an expensive reasoning model per request — but the paper gives no measurements, so it won't tell you when escalation pays off. Treat it as a vocabulary source for your own routing design, not as evidence for a specific threshold.
Discussion angle
If your agent had a real "self model" recording which solvers succeeded on which problem types, what would you actually log, and would that beat a simple static rule like "escalate to the reasoning model only on tool-call failure"? Ask the room whether anyone has measured their own fast-vs-slow split, or whether it is still guesswork.