Boffins build AI agents that respond like real people

Researchers from major universities and Google DeepMind have developed a generative AI architecture that can simulate the behavior of over 1,000 real individuals by leveraging qualitative interviews and large language models. This approach allows AI agents to respond to social science survey questions in a manner closely matching the actual responses of the people interviewed, achieving 85% accuracy. The technique could pave the way for creating custom AI agents to interact with others on behalf of individuals, even potentially after their death.