AI Predicts Behavior of 1,000+ Humans After 2-Hour Interviews

­This is…terrifying.

We already have generative AI systems that can closely imitate the look and sound of real people. We have AI rewriting (or in many cases, writing) term papers, composing creative arts, and they’ve long-since passed the Turing Test.

But now we have “novel agent architecture” that can accurately predict individual responses with 85% accuracy after a mere two-hour interview.

Researchers from Stanford, Northwestern University, the University of Washington, and Google DeepMind presented their findings, taking 1,052 individuals and “applying large language models to qualitative interviews about their lives, then measuring how well these agents replicate the attitudes and behaviors of the individuals that they represent.”

The researchers took 1,052 test subjects and initiated 2-hour interviews – with an average of 6,491 words per participant – on a broad range of topics, like social sciences (e.g., “Tell me the story of your life”) and their views on current societal issues (e.g., “How have you responded to the increased focus on race and/or racism and policing?”).

The AI agent also dynamically generated follow-up questions.

Researchers then had the AI predict the human subjects’ responses on similar topics, and two weeks later, the subjects returned to answer the same questions.

And in all 1,052 cases, the AI and the human responses matched 85% of the time.

…meaning that if AI can already accurately predict human behavior that well after a mere 2-hour interview, how long until they can imitate humanity as a whole after little or no prompting?

We’ve already seen instances of AI attempting to trick its human creators. Forget the Turing Test – how long before digital AI is absolutely indistinguishable from its meaty masters?