Judgment models work best with input evaluation not simulation
Code Owns the Simulation, Jev Owns the Evaluation
Artificial IntelligenceMachine Learning
Summary
Some AI models can decide what’s best by simply judging the options described in the input, without needing extra explanation. The authors tested one such model, Jev, and found it can reliably pick the right choice when all the information is given. But Jev struggles when it must imagine or predict things not directly stated, like guessing what an opponent will do next. The research suggests combining these judgment models with computer code that can handle predictions or simulations leads to better decision-making overall.
What this means in practice
- •For robotics engineers: Use code-driven simulation combined with judgment models for more accurate robot decision-making, improving task success where prediction of future states is needed.
- •For game ai developers: Improve game agents by supplementing judgment models with explicit opponent action simulation to avoid suboptimal strategy choices.
Authors
Yaodong Yang, Hongyao Tang, Yi Ma, Xingyu Fan, Weixun Wang, Jinpeng Li, Tianpei Yang
Abstract
Judgment models such as \jev{} return, in a single call and without reasoning text, a probability for each described option. This makes them attractive as an agent's action-selection layer, but it is unclear which decisions they can be trusted with. We test \jev{} on reflection tests, one-shot matrix games, the text game ALFWorld and robot control, and find a sharp boundary. \jev{} succeeds when the right option can be judged from what the input describes, which we call \emph{evaluation}. Specifically, it solves 99\% of the counterintuitive Cognitive Reflection Test questions. However, it fails when the right option depends on \emph{simulation} (i.e., predicting something not in the input), such as the opponent's action or the subgoal that must come first. In games, \jev{} plays suboptimally as if its rational opponent acted at random, because the opponent's action is not given. In ALFWorld, \jev{} favors commands that mention an object or place named in the task description. For example, given the task ``put a clean knife in the drawer'', \jev{} carries an unwashed knife straight to the drawer instead of first washing it at the sink. Surprisingly, many of these failures are not due to a lack of knowledge. Asked separately what the opponent will do, \jev{} usually answers correctly, and it responds well given the opponent's action. It fails when one call must both perform the simulation and evaluate based on it. This suggests letting code make the prediction or simulation. When code supplies it, such as a lookahead in ALFWorld and physics simulation in robot control, \jev{} becomes an expert controller through its general evaluation ability.