The Salem Center for Policy launched Friedman A.I., a large language model based on the philosophy of the late economist Milton Friedman on Feb. 29. 

Friedman A.I. collects and analyzes what the late economist Milton Friedman said in his lifetime to predict his perspective on present-day matters, said Nicholas Hallman, associate professor of accounting. Hallman said the center actively receives feedback from a Substack post to review and improve the chatbot’s responses.

“(We) arrived at the idea that it might be fun to resurrect, as it were, an iconic data-driven, data-focused, famous economist and Milton Friedman’s name was at the top of that list,” said Hallman, a senior scholar at the Salem Center.

Hallman said while the Salem Center is currently fine-tuning the Friedman chatbot, there’s potential for more economist chatbots.

“I would love to see us have a Samuelson chatbot or Keynes chatbot or somebody whose views appear, at least on the surface, to be diametrically opposed to Friedman’s and host our own simulated debates,” said Hallman.

Computer science professor Raymond Mooney said these large language models could be used as fun and educational tools but it’s important to be skeptical of their responses. He said large language models tend to “hallucinate,” or make up, information and sometimes contradict themselves when responding because the models learn off of patterns of text they previously processed and produce responses accordingly, although they may not be correct.

“Sometimes large language models surprise you (with) how smart they are, and sometimes they surprise you (with) how stupid they are,” said Mooney, the director of the Artificial Intelligence Laboratory at the University. “(The hallucinations) will sound plausible, but if you look into it, it’s just garbage.”

Because the bot is still in its early phases, these inaccuracies and contradictions are part of the process. Still, Hallman is excited about getting to use Friedman’s wisdom and apply it to new topics. 

“Some people that use the Milton Friedman chatbot don’t like the fact that the bot will sometimes obviously know about things that happened after the real Milton Friedman died,” Hallman said. “As an example, somebody asked Milton Friedman to comment on Taylor Swift. …  I don’t mind at all if he comments on the popularity of Taylor Swift and the sort of economic impact of a Taylor Swift concert in a city and what that means — I see that as a good thing.”

Editor’s Note: This story was corrected to establish that Friedman A.I. was launched on Feb. 29, not March 5. The Texan regrets this error.