Rules, Predictions & Playing the System

60 per cent of the world’s working population earn a living in the informal economy – a place where people make their own rules and defy official laws of their country. Think street food vendors or handicraft makers. Interestingly, in the case of entrepreneurs, they do not outright reject the system but mix and match official regulations with unofficial norms, deciding when to play by the rules and when to bend them to their advantage. Entrepreneurs thereby choose just how ‘formal’ to be. This talk explores how those decisions are made.

Before running an experiment, researchers often commit to their predictions in advance — a practice called pre-registration. But since AI tools like ChatGPT entered the scene, are those predictions coming true more often? We compared over 10,000 pre-registered predictions with what experiments actually found. The answer is yes: with each new generation of AI, predictions are more likely to be confirmed, but the effect is not uniform. It depends on how bold the prediction is and how much is already known about the topic. Why? AI may induce scientists to play it safer and help them design better studies. Or AI may already "know" the answer from the research it has absorbed. Teasing apart these mechanisms matters: if experiments are becoming more predictable, we need to rethink how we allocate lab access and fund experimental research in the behavioral sciences.