STARKVILLE — Usually, when Deborah Lee uses ChatGPT for a recipe or a meal plan, the generative artificial intelligence software gives her an answer almost instantaneously.

But one day, as the director of research impact and AI strategy at Mississippi State University Libraries, Lee decided to show her students what happens when you give the software a hypothetical scenario – something it typically tackles with ease. That’s when Lee asked the software how the Mississippi State Bulldogs could take home the SEC Championship this year.

“I really thought I broke ChatGPT with this,” Lee said with a laugh on Wednesday to listeners at the Oktibbeha County Heritage Museum, as the latest speaker in its community event series.

Lee said the query took ChatGPT almost five minutes, and when the answer came back, it was full of strategies that were no longer possible with the conference’s 2024 division unification. And at the end of the plan, there was something she had never seen – a “Reality Check.”

“The one thing about AI, is it will never tell you you’re wrong,” Lee said. “It will never say ‘that’s a stupid question.’ But this is probably about as close as I could get.”

Generative artificial intelligence is a subset of artificial intelligence that uses probabilities and predictions to create answers to requests, Lee said. While it is not yet at the point that it can replace people, it can approximate human thinking using language learning models, Lee said.

ChatGPT is one of the most commonly referenced models, though it exists in a field of competitors like Microsoft Copilot, Google’s Gemini, xAI’s Grok and more, Lee said. Since its emergence in 2022, artificial intelligence has become more commonly used over time, in everything from spellcheck and streaming service recommendations to automobiles and doctor’s offices. Lee also anticipated it only becoming more prevalent in the near future.

“Do you use it? Or do you not use it?” Lee said. “Only you can answer that question. I would argue that it’s going to be harder and harder to not use it. Because it’s going to be a part of the next phone you buy. I promise it’s going to be a part of the next laptop you buy.”

Lee demonstrated some of the potential ways listeners could use ChatGPT in everyday life, from writing a limerick or an appeal to creating an itinerary for a trip.

Lee also showed an example where she asked ChatGPT how someone can maintain a glucose level of less than 110 throughout the day, through activity and meals. The software answered Lee’s request with a general overview of best practices, followed by offering to produce a sample daily eating and activity plan to meet Lee’s needs. With an extra click, Lee had a full breakdown of recommendations with timestamps for foods to eat and activities to keep her glucose levels maintained throughout the day.

Caution with AI

The daily breakdown also included something Lee did not expect, she said. ChatGPT excluded seafood from her recommended daily plan, as it “remembered” that she had asked to avoid seafood in a search almost two years ago.

“What you’re doing with these models, it’s not private,” Lee said. “It feels private because you’re in your home on your computer, but anything you put in is added to its training database.”

Lee also warned that, since language learning models are designed to be helpful, their goal is to produce an answer to the user’s query. Sometimes, if a direct answer does not exist, language learning models will begin to “hallucinate” or make up answers that appear definitive.

Lee gave the example of a reading list included in a syndicated special section “Heat Index” produced by a third-party company and distributed in May by the Chicago Sun Times.

“They were real authors, but they were fake … titles,” Lee said. “They were not real book titles. It’s not even like they associated the wrong book with the wrong author, they just made up the titles. And it was a very public humiliation for the Chicago Sun Times … but it’s a classic example of AI hallucination.”

Chicago Sun-Times issued a correction, followed by an explanation and an apology from Chicago Public Media CEO Melissa Bell on May 29, which acknowledged instances when fact checking should have been used.

Because of the potential for AI hallucination or the inclusion of outdated information in search responses, Lee recommended following up generative AI searches with more research through authoritative sources. She also recommended not replacing a conversation with a doctor or other expert with an AI search query.

“You just have to be aware when you’re using this to double check anything that I would consider ‘mission critical,’” Lee said. “Something that’s really important. The recipe to fix chicken for this week, it’s not really important, so it’s OK if it kind of fudges on that. But if I’m doing something medical in nature, and I’m actually going to make an important decision based on the information, I’m going to verify that information someplace else.”

Lee also said it is important to remember that ChatGPT and other generative artificial intelligence are not humans and cannot replace human interaction, referencing a lawsuit filed by the parents of 16-year-old Adam Raine alleging ChatGPT contributed to their son’s suicide. The lawsuit was filed in the California superior court on Tuesday, according to reports from CNN.

“Computers can do a lot of things, but they are not humans. They can’t infer. It is not a substitute for human interaction,” Lee said.

READER FAVORITES

• 

  ‘Bears, beets, brownies’

  By Ashlynd King • 6 hours ago