A panel discussion at the University of Texas at Dallas Tuesday cleared up misconceptions about the viral AI chatbot, ChatGPT, but also noted concerns about its environmental impact and data privacy. Presented in partnership with The Dallas Morning News, the event featured four UTD scientists and was moderated by Adithi Ramakrishnan, science reporter at The News.
Ramakrishnan started the discussion by asking if students can use ChatGPT to write essays for them.
ChatGPT is a kind of artificial intelligence called a “pattern matching tool,” said Gopal Gupta, a computer science professor at UTD. It’s been trained on a large amount of data and text, including code and information from the internet, and works by predicting the next word in a sentence.
Although it can construct a well-written essay, Jessica Ouyang, an assistant computer science professor at UTD, said it can also make mistakes.
“For example, it might switch the names of two characters from the book you’re supposed to be writing an essay on,” she said. “And unless you are critically reading the essay that it has written for you, you may not realize or notice that.”
ChatGPT appears personable — it answers questions with “I” — and conversational. In some cases, it has seemed to express feelings of love and hate. But Ouyang said that isn’t really true: “Don’t worry, ChatGPT is not going to develop sentience and come after us.”
What actually happens is the chatbot remembers the conversation as it occurs. So if the user types a loving, or rude remark, ChatGPT will respond in a similar manner based on the data it’s been trained on.
Gupta compares ChatGPT to a precocious 8-year-old, listening to adult conversations without the experience or knowledge to make sense of them. The chatbot, unlike a human, is unable to reason, he said. “It’s got absolutely no thinking, any reasoning or any kind of logical inference.”
ChatGPT can be a useful tool, Gupta said, but without a human, it can’t function.
“AI is not going to replace your job,” said Dale MacDonald, an associate dean of research and creative technologies at UTD. “A human who can use AI will replace your job.”
ChatGPT can boost productivity and efficiency by combining ideas that already exist, Gupta said.
“Maybe you can combine results in five medical journal papers and come up with some new diagnosis or new treatment plan,” he said.
But it can’t be trusted with crunching numbers, said Xinya Du, an assistant computer science professor at UTD.
When asked “what’s 2 plus 2?” ChatGPT is only able to answer “4″ because of previous answers on the internet.
“We can’t really rely on the AI model to handle these calculations for important analysis,” Du said.
MacDonald even wondered how useful ChatGPT can be.
“This is still at a very high hype state,” he said. “And until people come up with actually good reasons to use it, it may not bear the investment.”
Beyond the practicality of ChatGPT, there are also concerns about its environmental impact.
“There is a lot of environmental concerns with the carbon footprint of training these models,” Ouyang said. “The amount of electricity they use is staggering.”
Gupta and Ouyang also warned about data privacy when using technologies like ChatGPT. Ouyang doubts that OpenAI, the company behind ChatGPT, is selling user data, but said that privacy is still a major concern.
“It’s always a concern that anything you put on the internet could be collected,” she said. “And how would you ever know?”
Still, ChatGPT can be helpful, both at work and in our personal lives. Ouyang, for example, said that she uses it to strategize for the role-playing game, Dungeons & Dragons.
