MIT and Stanford create AI avatars that change people's major life decisions
Anthropic
ElevenLabs
Researchers from MIT and Stanford have developed a multimodal neural network system that creates a digital twin of a person 30 years older. In experiments with 192 volunteers, interacting with such an avatar increased willingness to change initial life choices by up to 34.2%. The key factor was not visual realism but logical persuasiveness.
Scientists from MIT Media Lab, Stanford University, and UCLA created a system that builds a 30-year-older interactive avatar of a user. The experiment involved 192 people aged 18-28 facing real decisions about career, education, or relocation. Participants uploaded a photo, voice recording, and filled a questionnaire about values, dreams, and fears. The aging was done by Nano Banana and LivePortrait, voice cloned by ElevenLabs, and biography generated by Claude Sonnet 4.5. Showing two avatars (each advocating one original option) increased readiness to change initial opinion from 10.8% to 34.2%. A third avatar offering an alternative path (e.g., 'professor of bioengineering') was chosen in 20% of cases, 7 times more than control. Users rated logical reasoning as most valuable, not visual or emotional factors. Ethical concerns were raised: the avatar could be used for hidden manipulation by commercial entities. Researchers proposed four principles: balanced generation of multiple avatars, transparency, right to dispute avatar details, and strengthening personal responsibility.
- Сокращения
- MIT = Massachusetts Institute of Technology — Массачусетский технологический институт
- UCLA = University of California, Los Angeles — Калифорнийский университет в Лос-Анджелесе
Source: Habr — хаб ИИ —
original