The 25 AI avatars were put into the virtual town of Smallville and given prompts on the back end to simulate human behaviors. The researchers did this using the model for OpenAI’s popular ChatGPT chatbot, GPT 3.5. The avatars were able to “produce believable individual and emergent social behaviors” to complete the tasks given to them.
What happens when you give AI avatars the run of a virtual town?
Answer: Shockingly, nothing out of the ordinary.
What happens when you tell AIs to act like humans? Well, they act like humans, according to researchers from Stanford University and Google. The team wanted to know what would happen if they created a bunch of AI avatars and then put them together in a virtual town and told them to interact. So they did just that.
The 25 AI avatars were put into the virtual town of Smallville and given prompts on the back end to simulate human behaviors. The researchers did this using the model for OpenAI’s popular ChatGPT chatbot, GPT 3.5. The avatars were able to “produce believable individual and emergent social behaviors” to complete the tasks given to them.
They were able to debate elections, plan parties, go on dates and even have impromptu conversations with little to no prompting. It all sounds a bit mundane, to be honest. Even the glitches, or "hallucinations" in AI terms, were nothing to write home about. After all, forgetting the occasional event is pretty normal for everyday humans, too.
The 25 AI avatars were put into the virtual town of Smallville and given prompts on the back end to simulate human behaviors. The researchers did this using the model for OpenAI’s popular ChatGPT chatbot, GPT 3.5. The avatars were able to “produce believable individual and emergent social behaviors” to complete the tasks given to them.