AI & HUMAN JUDGMENT
Advice has a checking cost.
Try an AI adviser role, or “AI seat,” in a game. Write down what information the adviser receives and what it may recommend, then have a person follow that procedure. Include a possible error: how could a player catch it, and what would checking cost?
Cognitive offloading: delegate a task, such as a calculation, while checking the inputs and deciding how to use the result. Cognitive surrender: accept the AI’s judgment with little scrutiny.
In Last Convoy, ask what you checked before accepting the adviser’s claim. A correct outcome alone cannot show whether you exercised judgment. After play, discuss what checking revealed and whether it was worth the time.
Concepts: Steven D. Shaw and Gideon Nave, Thinking—Fast, Slow, and Artificial (January 11, 2026, working paper). Teaching application informed by Red Teaming Associates’ human-in-the-loop approach. These game prompts are discussion aids, not a validated measure of cognitive surrender.
A better answer can conceal a learning gap.
If AI supplies the plan or explains every outcome, players may miss the practice of weighing evidence themselves. Record your reasoning before receiving advice. Check it against the rules, then explain why you accepted or rejected it.
Teaching Wargaming guidance for classroom discussion. Source: MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training, Report (August 13, 2026), section 2.7, “Augmentation not automation.”
What stays with people?
Players decide how to use advice and check whether its evidence supports their choice. A person following an adviser procedure can help you test the role. The exercise cannot establish how a live AI system would behave.