Researchers created a novel experimental setup using AI models trained only on historical text—specifically, one trained on everything published before 1930—to study how people's memories and perceptions of the past get distorted by everything they've experienced since. The core insight is simple: we can't actually talk to people from the past or run experiments in historical societies, but we can talk to AI trained on historical data, which has no knowledge of what came after. They ran a preregistered experiment where participants interacted with this 1930-bounded AI model versus a modern AI (GPT-5.5 in their framing), then measured whether the interaction changed how people viewed the morality of past societies. The historically-bounded model reduced what psychologists call the "illusion of moral decline"—the widespread bias that people were more moral in the past than they are today. The results showed that talking to the historical AI prompted more reflection and actually shifted participants' views about moral values in the 1930s compared to the control group. This matters because our normal understanding of history gets filtered through everything that's happened since—wars, social movements, scandals, moral progress on some issues and regression on others. That filtering makes it nearly impossible to see what people actually believed and valued at a given moment. By interacting with an AI that genuinely doesn't know about the intervening decades, people got a different kind of window into historical thinking. The researchers frame this as a new methodology they call "Time Machine Experiments," which borrows from design fiction and speculative design traditions in HCI but turns those thought experiments into actual measurable studies. Rather than just imagining what it would be like to talk to someone from 1930, they made it possible and tracked what happens when people do. The approach opens up broader possibilities: you could create historically-bounded models for different time periods and use them to study how temporal knowledge gaps reshape human perception, reasoning, and behavior across all kinds of domains beyond morality.