An AI-assisted law game raised test scores

In a six-week Chinese university trial, 84 law students were randomised equally to an offline mystery game or lectures. After baseline adjustment, game participants scored 3.59 points higher on a 100-point course test (95% confidence interval 2.90–4.29).[1]

Published on 28 October, the study used Gemini 3 Pro to draft seven characters for The Case of the Speluncean Explorers. Researchers revised the scripts; students debated in character without a live chatbot.[1]

One instructor taught both groups. The result tests the complete teaching approach, not AI alone; later retention was not assessed.[1]

The fictional case has a long history. Lon Fuller’s 1949 text asks judges to decide the fate of four survivors who killed and ate a companion after becoming trapped underground. Its judges disagree about how the law should work: whether to follow a statute’s wording, interpret its purpose or defer questions of mercy to another authority.[2]

That disagreement gives the scenario its teaching potential. The challenge is to explain why a rule should apply, then face a rival interpretation of the same predicament. A persuasive answer needs a reason that survives an objection, rather than merely a remembered legal term.[2]

Report an error or suggest a correction