SALT LAKE CITY — What is intelligence? After all, the rapidly developing technologies of artificial intelligence and large-scale language models are not only assets in the pursuit of new scientific discoveries, but also intelligence. can be an important resource for exploring the true nature of
That's according to neuroscientist David Eagleman, who currently teaches at Stanford University.
“The question is, if we enter a world with another intelligence, will things get weird?” Eagleman, best-selling author and screenwriter of the television series “The Brain with David Eagleman” reflected with an audience at the University of Utah's Kingsbury Hall on Tuesday. He spoke as part of the Utah Museum of Natural History's 2024 lecture series centered on the nature of intelligence.
Eagleman told the audience that AI appears so smart because it instantly synthesizes information from all documented sources on the internet and uses sophisticated large-scale language models (a type of artificial neural network). I explained that this is simply due to their ability to eloquently generate answers to queries. Through learning the language and syntax, the program can synthesize information into precise statements that mimic natural speech.
“What you have to consider is that absorbing text, absorbing trillions of pages of text, and running large statistical models on them is not intelligence or perception in itself. “Because you don't know what you're talking about,” Eagleman explained, adding that programs like ChatGPT collect the information they collect, using a nearly limitless library of sources to deliver information at incredible speeds. Collect.
To illustrate this point, Eagleman used the China Room thought experiment to argue that no matter how human-like programmers create an AI, it cannot actually be intelligent. .
The premise of the “Chinese Room Thought Experiment” involves participants who speak Chinese fluently and those who know no Chinese at all. The person, who does not know Chinese, is locked in a room full of books with instructions on how to work with Chinese symbols. When a fluent Chinese speaker passes messages in Chinese to people in the room, non-Chinese speakers look up the matching symbols and see which symbols are appropriate responses to those messages. .
Given the wide selection of sources in the room and the thorough representation of written Chinese, Eagleman believes that even non-fluent people can convince fluent people that they know Chinese. I reasoned that I would be able to convince him. However, that does not mean that the non-fluent person actually knows Chinese; by piecing together the information they have access to, they may be able to trick the other person into believing they are fluent in Chinese. It just means that.
“The question is, have we proven that computers can be intelligent? In reality, these computers are just playing a statistical game,” Eagleman said, adding that the Turing test He explained that previous intelligence tests such as the Loveless Test and the Loveless Test test large-scale language models. A capability they have been capable of for a long time.
“What I proposed in the literature a few months ago is a new test of intelligence. If a system is really intelligent, it should be able to make scientific discoveries,” he said. “One of the most important things humans do is science, so the day our AI can make real discoveries will be the day I consider AI smart.”
One of the most important things humans do is science. So the day that AI can make real discoveries is the day I think AI is smart.
–David Eagleman, neuroscientist and author
Eagleman further illustrated his point by classifying two forms of intelligence: Level 1 scientific discovery and Level 2 scientific discovery. The former involves the culmination of existing facts and concepts to find effective solutions, and the latter involves arriving at scientific discoveries through the conceptualization of original ideas.
Examples of Level 2 scientific discoveries cited by Eagleman include Albert Einstein's discovery of the theory of relativity and Charles Darwin's ideas about evolution.
But Eagleman also said he believes we are at the point where AI is proving capable of creativity.
“The human brain is good at remixing information, bending, breaking, and merging ideas. In fact, what we've seen is that these LLMs (large-scale language models) It's perfectly good at that,” Eagleman said. “At this point, we feel like they might be just as creative as we are. What they're bad at is filtering out what humans care about.”
Regarding the near future of technology, which seems to be becoming more sophisticated by the week, Eagleman sees a bright future in which AI makes scientific discoveries alongside its human cohorts and provides humans with valuable personal resources. He said he is doing so. A therapist tailored to your needs is always available for consultation.