A lively debate about AI breakthrough ChatGPT was the subject at UC San Diego’s recent on-campus event “ChattingGPT,” which featured a series of panels discussing the future of AI chatbots. The event was sponsored by the San Diego Supercomputer Center, Halıcıoğlu Data Science Institute (HDSI) and the Institute for Practical Ethics at UC San Diego.
ChatGPT – a so-called ‘Large Language Model’ (LLM) created by OpenAI – is novel in its accessibility. It’s free and anyone in most countries with an internet connection can use it.
Teaching Professor, HDSI
UC San Diego
“LLMs are costly to operate, and so previously they were kept under wraps and used internally. Or, maybe they were difficult to work with – you had to be a software engineer to access them,” said UC San Diego and HDSI Teaching Professor Dr. Justin Eldridge, PhD.
“It’s an incredibly useful tool for information, but that information isn’t always accurate,” added Dr. Eldridge.
In front of the room, Dr. Eldridge asked ChatGPT to provide papers on feline laryngitis. (If you weren’t aware, felines don’t get laryngitis.) Still, the chatbot perceived it as true – even though it was false – and provided a list of papers to support it, Dr. Eldridge added. “It has a problem with hallucination – LLMs will make something up.”
Massive Experiment
“We are being used as subjects in a massive experiment without our consent and I think that’s an enormous ethical challenge,” said HDSI and Philosophy Professor Dr. David Danks, PhD. “You might be thinking this is just math and just code and ethics doesn’t have anything to do with it. I want to push back on that.”
Professor, HDSI
UC San Diego
As Dr. Danks explained, generative AI can provide a host of benefits, like enhancing creativity and reducing menial labor. Yet the potential pitfalls could be detrimental. “These systems can massively increase the scale and magnitude of harm,” he added.
That harm may be intentional if it spits out disinformation or biased content. It may also be unintentional, as is the case with presenting hallucinations as truths.
“For ChatGPT, ‘success’ means that it predicted the correct next word – not whether it told the truth. It just doesn’t care. It’s training mostly on true things, but it will very confidently make things up,” Dr. Danks said.
Deskilling Concerns
Dr. Danks also raised concerns over deskilling the population through advancements in generative AI. He cited the spell-check function on your word-processing software, which has sped up and simplified the editing process but has arguably set some of us back with our spelling capabilities – now sometimes called ‘deskilled.’
Ethical concerns about AI lead to medical concerns about whether patient safety can remain paramount during an AI revolution.
Chief Medical Officer
UC San Diego Health
UC San Diego Health Chief Medical Officer Dr. Chris Longhurst, MD is cautiously optimistic. He said his hospital recently announced that it’s one of the first partners in the country to test the use of operationalized GPT inside its electronic health record system, but only for a very specific use case.
“We’ve become overwhelmed with messages from patients,” Dr. Longhurst explained.
Clinicians don’t always have enough time to respond to every patient so the GPT system will draft messages that clinicians can then edit for content and tone before they’re sent to patients.
“We don’t want any hallucinations going to our patients, so the editing is critical,” Dr. Longhurst added.
UC San Diego
FOUNDED: 1960
CHANCELLOR: Pradeep Khosla
FACULTY: 3,085
ENROLLMENT: 43,000 students
WEBSITE: ucsd.edu
NOTABLE: UC San Diego was ranked third among the nation’s top public colleges, according to the Forbes 2022 list of America’s Top Colleges. UC San Diego was No. 17 out of the complete list of close to 500 four-year private and public colleges in the country.
