Nearly 20% of participants who chatted with three AI-generated future selves said they would choose a previously unconsidered path, versus less than 3% in a control group that only imagined the future.
The MIT-led team tested almost 200 people facing hard decisions, using aged images, detailed questionnaires and Claude Sonnet 4.5 to build avatars that generated realistic future “memories.”
Researchers say the avatars may strengthen future self-continuity—helping people connect present choices to long-term goals—but the study tracked stated preferences, not whether participants later followed through.
Critics say the effect could reflect novelty or manipulation rather than better judgment, and note AI advice can mirror creators’ biases, including steering lower-income users toward safer, shorter-term choices.
The findings add to a broader debate over whether AI can improve difficult decisions or simply repackage them, especially when choices hinge on conflicting values rather than missing information.