After two hours of conversation, a new AI model is capable of imitating human personalities with 85% accuracy, potentially revolutionizing the modeling of behavior and opinions.
An AI model was trained to mimic human behavior by Stanford University and Google’s DeepMind, according to a recent scientific study. After only a two-hour interview, computerized versions of real individuals were developed by the Generative Agent Simulator.
The research was done with over 1,000 individuals after which the AI model was able to mimic their personalities with 85% accuracy. After speaking with the AI for a few hours, it will be able to replicate your reactions accurately enough to appear to be thinking exactly like you.
The AI was first read the first few lines of The Great Gatsby by the 1000+ participants. This was the AI equivalent of stretching before working out, it seems. A 2D figure then questioned them about their families, careers, beliefs, lives, and other topics.
The AI was able to create a digital clone in roughly two hours and with an average of 6,491 words.
Also, the clones appeared to be competent. In almost 85% of cases, the AI bots’ responses to questions from personality tests or general surveys mirrored those of their real-world counterparts.
The researchers tested these AI clones using economic games such as the Dictator Game and the Prisoner’s Dilemma, in which players must make decisions regarding resource sharing, cooperation, and trust. Even if it was only just about 60% of the time did the AI make the same choices as the real human, that’s still more than could be explained by chance.
The AI may not be able to utterly trick your best friend or spouse, but based on that brief interview, it’s quite amazing.
Mimicking Human Behavior
Decision-making processes, viewpoints, and even peculiarities of personality could be mimicked by the AI. Consider this: you spend years discovering your own identity and preferences, and an AI does so in a single afternoon. It is, like all good AI presentations, both amazing and a little unsettling.
Stanford and DeepMind believe that the study of human behavior can be advanced by this AI. It also offers a preview of how AI might mimic human behavior in general. Are you curious about how a new health policy might be received by a community? Talk to some generative agents about it. Wondering what consumers will think of a daring product redesign? Consult the AI clones.
It could basically be a focus group that never ends. What happens if an AI has access to years’ worth of data, of course, if it can learn to imitate you this successfully after just one interview? Your Spotify playlists, social media posts, and online purchasing patterns may all train the AI to be you or at least accommodate your preferences—even ones you weren’t aware you had.
This technology is currently in the hands of researchers, and the emphasis is on learning how to make it useful for disciplines like sociology, psychology, and economics, even if there is a lot of room for abuse by con artists and other bad actors. Furthermore, it is a little inevitable that technology will advance in this way. Nothing is more human than making something that makes a concerted effort to resemble ourselves.