A two-hour interview is enough to create an accurate AI replica of your personality, says Stanford study

Through nan looking glass: Artificial intelligence has demonstrated singular capabilities, but could it replicate an full characteristic aft conscionable a two-hour interview? According to researchers, nan reply is yes. Yet, specified advancements raise superior ethical questions and concerns astir imaginable misuse.

Researchers from Google and Stanford University person demonstrated that conscionable a two-hour speech pinch an AI exemplary tin create a strikingly accurate replica of an individual's personality. Published connected November 15 successful nan preprint database arXiv, nan study introduces "simulation agents" – AI models designed to mimic quality behaviour pinch singular precision.

Led by Joon Sung Park, a doctoral student successful machine subject astatine Stanford, nan investigation progressive in-depth interviews pinch 1,052 participants. These interviews covered perseorangan stories, values, and opinions connected societal issues, forming nan dataset for training nan generative AI models. The subordinate excavation was intentionally divers successful age, gender, race, region, education, and governmental ideology, ensuring a wide practice of quality experiences.

To measure accuracy, participants completed 2 rounds of characteristic tests, societal surveys, and logic games, repeating nan process aft a two-week gap. The AI replicas past took nan aforesaid tests, mirroring their quality counterparts' responses pinch an astonishing 85 percent accuracy.

"If you tin person a bunch of mini 'yous' moving astir and really making nan decisions that you would person made - that, I think, is yet nan future," Park told MIT Technology Review.

The researchers envision these AI models revolutionizing investigation by simulating quality behaviour successful controlled environments. Applications could scope from evaluating nationalist wellness policies to gauging responses to societal events aliases merchandise launches. Such simulations, they argue, connection a measurement to trial interventions and theories without nan ethical and logistical complexities of utilizing quality participants.

However, these findings should beryllium approached pinch a patient dose of skepticism. While nan AI clones excelled successful replicating responses to characteristic surveys and societal attitudes, they were notably little meticulous successful predicting behaviors during interactive economical decision-making games. This discrepancy underscores AI's ongoing challenges pinch tasks that require knowing analyzable societal dynamics and contextual nuances.

The information methods utilized to trial nan AI agents' accuracy were besides comparatively rudimentary. Tools for illustration nan General Social Survey and assessments of nan Big Five characteristic traits, while modular successful societal subject research, whitethorn not afloat seizure nan intricate layers of quality characteristic and behavior.

Ethical concerns further complicate nan technology's implications. In an era wherever AI and "deepfake" technologies are already being utilized for manipulation and deception, nan preamble of highly personalized AI replicas raises alarm. Such devices could perchance beryllium weaponized, amplifying risks to privateness and trust.

Despite these reservations, nan study introduces compelling possibilities for early research, notes John Horton, an subordinate professor astatine MIT Sloan School of Management. "This insubstantial is showing really you tin do a benignant of hybrid: usage existent humans to make personas which tin past beryllium utilized programmatically/in-simulation successful ways you could not pinch existent humans," he said.

The ratio of nan question and reply process successful capturing perseorangan nuances is peculiarly striking. Park emphasized nan extent of penetration a two-hour speech tin provide, drafting from his acquisition pinch podcast interviews. "Imagine personification conscionable had crab but was yet cured past year. That's very unsocial accusation astir you that says a batch astir really you mightiness behave and deliberation astir things," he said.

This invention has piqued nan liking of companies already processing integer copy technology. Hassaan Raza, CEO of Tavus – a institution specializing successful creating AI replicas from customer info – expressed enthusiasm for this streamlined approach. "How astir you conscionable talk to an AI interviewer for 30 minutes today, 30 minutes tomorrow? And past we usage that to conception this integer copy of you."