AI from the past alters views of morality
Summary and headline written by AI from the source article. How we work
Researchers conducted a randomized experiment with 240 participants to explore how interacting with an artificial intelligence modeled on a person from 1930 impacts perceptions of the past.
The team created a historically-bounded large language model (LLM) trained only on text from before 1930. Participants then engaged in conversations with this LLM. The study revealed that interacting with the past-trained AI reduced the “illusion of moral decline”, the common tendency to believe that past societies were more virtuous than our own.
Compared to a control group that interacted with a contemporary LLM, those who spoke with the 1930s-era AI reported less of a belief that morality has decreased over time. This suggests that access to a perspective unshaped by later events can influence how people evaluate historical morality. This research introduces a "Time Machine Experiment" paradigm. By manipulating the temporal knowledge available to an AI, scientists can treat time as a variable in experiments.
The team believes this approach opens new possibilities for testing ideas previously confined to thought experiments and pushes the boundaries of what's possible at the intersection of science fiction and scientific inquiry. Further research could explore how different historical periods or specific individuals might shape perceptions.
