UNC researchers dive into ethics, implications of AI chatbots
ChatGPT, an artificial-intelligence chatbot, can write a poem about spring and even turn that poem into a 1970s sitcom. While AI projects like this are fun to interact with, researchers at the University of North Carolina at Chapel Hill are looking into the bigger questions posed by using such technology.
The Carolina AI Project started this semester will explore the reality, morals and ethics of AI and virtual reality. The project brings together faculty and graduate students from the philosophy, linguistics and computer science departments.
Artificial intelligence comes with concerns of privacy, bias and even replacing human jobs. AI Project Director Dr. Thomas Hofweber said now is a critical time to study the technology and its implications.
"In the very recent past, the advances have been real breakthroughs and so it's completely changing the field," Hofweber said. "I believe, and so does almost everybody else, that it will have a profound impact on a lot of different things - on society, on research on the university."
Peter Hase, a Ph.D. candidate in computer science, said the development of artificial intelligence has been faster than anticipated. Hase has been working on a practical problem – how to update information in the system when a chatbot gets a wrong answer.
"It happens that this problem has a long story in philosophy, where it's known as belief revision," Hase said. "The problem is that if a person is going about in the world and they encounter new information, how should they revise their belief in light of this new information?"
Experts like Hofweber are looking at how these AI systems are related to human intelligence.
"Basically you're trying to build a system that exhibits intelligent behavior, artificially not through a biological process, but building through computing," Hofweber said.
Despite the breakthroughs in computer science, the AI computer language models remain "somewhat mysterious," Hofweber said.
"They're created through more or less brute force training method. You put a lot of computing power and a huge amount of data towards them, and you produce this incredibly complicated neuron network as a result," he said, "but what precisely is going on inside the network, why it does the things it can so amazingly do, it's not completely understood."
The project will be collaborating with the Parr Center for Ethics.
"To properly understand what's going on with language models, you need to involve people from different disciplines - linguistics, philosophy and computer science," Hofweber said.
