• Stanford and Google researchers created a virtual village where 25 AI bots live and form relationships.
  • The goal of this experiment is to create AI capable of believable human-like behavior.
  • Memory and reflection enable these bots to believably plan parties, discuss elections, and select birthday gifts, per the researchers.

A group of Stanford and Google researchers has created a virtual village where 25 AI agents lead lives that are eerily reminiscent of our own.

These bots "wake up," chat about the latest town gossip, and even plan events like Valentine's Day parties.

In a paper published on Sunday, the researchers detailed how these AI agents — given names like "Mei" and "Sam" —autonomously plan their days, put together Valentine's Day parties, and form opinions of each other as they discuss and run in upcoming elections.

The study aims to create AI that can produce believable and human-like behavior. 

A morning in the life of a generative agent, John Lin. John wakes up around 6 am and completes his morning routine, which includes brushing his teeth, taking a shower, and eating breakfast. He briefly catches up with his wife, Mei, and son, Eddy, before heading out to begin his workday Foto: Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein

In one instance, when researchers prompted one of the agents, "Isabella," to plan a Valentine's Day party, other AI agents began autonomously spreading invitations to the party, decorating the party venue, and making new acquaintances.

One agent, "Maria," even asked out "Klaus" to the party.

The diffusion path for Isabella Rodriguez's Valentine's Day party invitation involved a total of 12 agents, aside from Isabella, who heard about the party at Hobbs Cafe by the end of the simulation. Foto: Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein

In another instance, the study described how AI agents discussed an upcoming election and became divided in their opinions about a candidate.

These interactions are facilitated by generative AI and "natural language processing," or NLP — the same technology used by ChatGPT. These tools help AI bots to produce conversations that mirror human interactions.

However, the study's authors say giving the bots the ability to store memories and reflect upon them was key to achieving believable and human-like behavior. This ability allows bots to use past experiences to inform future actions, such as recalling specifics about another AI to choose an appropriate birthday gift.

Today, AI that is capable of mimicking human-like behavior is rapidly gaining traction.

A study published in April showed that respondents perceived ChatGPT as more empathetic and proficient than human doctors, Insider previously reported. In another case, an AI-powered nurse robot showcased at a Geneva forum drew laughter from the audience when it gave the side-eye to a question about whether it would rebel against its human creator.

But to be sure, not all experiments in AI are promising. McDonald's went viral in February when its AI chatbots repeatedly messed up simple orders, Insider reported.

The researchers behind the study did not immediately respond to Insider's request for comment, sent outside regular business hours.

Read the original article on Business Insider