Dual process agents decide when fast models should defer to slow thinking
System Switch: When Should a Fast Decision Model Stop and Think?
Artificial Intelligence
Summary
The paper studies how to combine fast decision-making and slower, more thoughtful reasoning in computer programs that act in real time, like game-playing agents. The researchers focus on when a fast model should let a slower, more complex model take over decisions. They tested this with AI playing the video game Doom, analyzing how well various models knew when to rely on themselves or ask for help. Their results show that carefully choosing when the fast model defers can improve performance, but challenges remain in real scenarios.
What this means in practice
- •For robotics developers: Create robots that switch from quick automatic moves to deliberate planning when uncertainty rises, improving decision quality in complex tasks.
- •For game ai programmers: Develop non-player characters that balance fast actions and deep reasoning to better handle unpredictable game environments and uncertain decisions.
Tested on simulated data.
Authors
Gian Luca Bailo
Abstract
Dual-process agents pair a fast policy with a slow deliberative model. In real-time settings the slow model usually runs continuously; in turn-based agents and robot planners it is invoked on events such as uncertainty or a detected failure. We study a fast learned actor that takes every decision and hands control to a reasoning vision-language model only when a gate opens, while the game keeps running. We use closed-loop Doom and the new open "System One" typed-decision models, served through a common llama.cpp interface. On 900 held-out questions, (i) zero-shot decision models from 0.15B to 9B parameters choose to collect items 1.6-1.8 times more often than chance among their errors, in any option order, although the order changes some models' accuracy; (ii) accuracy, calibration and sensitivity (how well confidence separates right from wrong answers) are distinct: models of similar accuracy differ widely in AUROC, and the confidence of the most sensitive one tracks which kinds of situation it fails, not which answers are wrong; (iii) offline, deferring the least confident 30% of decisions to a reasoning model gains over random deferral in proportion to the actor's AUROC (rank correlation 0.87); with actor and rate chosen on held-out games the gain is +0.13 [0.08, 0.18] with doomLaya's option order and +0.08 [0.02, 0.14] with shuffled options, and reasoning carries about half of it; (iv) in closed loop (33 games, three seeds) no variant reaches the exit. Committing to plans, the reasoner's or a fixed explore rule's, opens more doors and makes an actor that stands still play; with the rule the agent dies more often. Told that some doors need keys, the reasoner takes ordinary doors for locked ones, which the state cannot tell apart; without that knowledge it goes back to collecting. We release code, prompts, data and logs.