Original title: Time Machine Experiments: Using Historically-Bounded AI for Inquiry into the Human Mind
Hiromu Yakura, Robin Schimmelpfennig, Ezequiel Lopez-Lopez, Alejandro H. Artiles +4 · September 2026
People often think the past was better than the present. We cannot observe history directly. Instead, we rebuild it using memories and stories that are colored by everything that has happened since. To test if talking to the past changes our views, researchers ran an experiment. They used a historically-boundedAn AI that is limited to information from before a certain point in time, in this case 1930. AI. This is a computer model trained only on text written before 1930, so it knows nothing about later events. People chatted with this system as if speaking to someone from that era. The researchers found that this conversation reduced the illusion of moral declineThe belief that people were more moral in the past than they are now, which may not be objectively true., which is the false belief that people long ago were more moral than people today. Talking to a simulated past altered how people judged history.
Imagine you always thought your hometown was friendlier ten years ago. But you only remember it through the lens of everything that happened since—a big move, a new highway. Now suppose you could talk to someone who left town exactly ten years ago and never heard any updates. Would their version of the past change your feeling? In this paper, researchers built a chat AI that only knows the world up to 1930, so when people interact with it, they get a perspective untouched by later history. That interaction measurably reduced their tendency to see the past as morally better.
This shows we can use AI to experimentally test how our view of the past is twisted by knowing what came later. Historians, psychologists, and anyone curious about moral bias might use this to understand why people often think the past was better. It turns a thought experiment into a real experiment.
The abstract presents this as a proof-of-concept with a single historical boundary (pre-1930). The effect was only measured on the illusion of moral decline, not other beliefs about the past. Generalizing to other time periods or different historical scenarios requires further study.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Time Machine Experiments: Using Historically-Bounded AI for Inquiry into the Human Mind — Hiromu Yakura, Robin Schimmelpfennig, Ezequiel Lopez-Lopez et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.