Digital twins have long become a fundamental tool, especially in medicine, when testing different types of therapies or in climatology, when evaluating strategies against global warming. But now, according to a study by Stanford and Google scientists, published on arXiv, Digital twins could be created in just 2 hours. And they would have our personality.
Led by Joon Sung Park, the team that signs The study recruited 1,000 people who varied in age, gender, race, region, education and political ideology. Based on interviews with them, the team created digital replicas (AI agents) of those individuals. As a test of how well the agents imitated their human counterparts, participants took a series of personality tests, social surveys, and logic games, twice each, two weeks apart; The officers then completed the same exercises. The results were 85% similar.
“Yes you can having dozens of little ‘you’s running around and making the decisions that you would have made, I think that’s ultimately the future” says Park.
In the study, the replicas are called simulation agents, and the goal is to facilitate social science scientists conducting studies that would be expensive, impractical, or unethical if done with real human subjects. If AI models can be created that behave like real people, the idea is to use them to test the effect of misinformation on social networks, the reaction to injustice or the behavior that causes traffic jams on the road.
“This article shows how you can make a kind of hybrid: use real humans to generate characters that can then be used programmatically/in simulation in ways that could not be used with real humans”, the authors point out in an interview.
Of course, the study also has a dark side. In the same way that AIs capable of generating images facilitate the creation of fakes without people’s consent, these agents pose questions about the ease with which people can create tools to impersonate others onlinesaying or authorizing things they did not intend to say.
Although the information necessary to build these agents can be obtained through data accessible in social networks, emails and history on web pages, An interview allows you to delve into very specific and more personal topics, something that a survey does not achieve.
The authors also highlight that right now agents need two hours, but This time can be reduced to 30 minutes if we have other data available on the Internet.
On the positive side, having a “digital other” to carry out our routine tasks (replying messages, filing your income tax return, making a shopping list, looking for vacation deals…) is very tempting. The problem is that Nobody assures us that this other self is not hackable and makes us say or do things that we have not really said.…Or is it?