- AI has the tendency to “hallucinate,” or make up information that it thinks is real—and even seems legitimate, or plausible. But while people criticize AI for its ability to hallucinate, scientists celebrate this side effect. They say it’s helped make new inventions and discoveries possible.
In October, David Baker was one of three scientists—and the only American among them—to win the Nobel Prize in chemistry. Baker, from the University of Washington, won for using computer software to invent a new protein.
In an interview with the New York Times before the prize announcement, Baker said AI hallucinations were central to “making proteins from scratch,” adding they helped his lab to design around 10 million “all brand-new” proteins that “don’t occur in nature.”
“Things are moving fast,” he said. “Even scientists who do proteins for a living don’t know how far things have come.”
AI hallucinations happen when artificial intelligence makes up information that it believes is true. And for many folks, at first glance, that information will likely appear to be factual, as AI can roll out fake information that looks highly plausible. But while this has become a topic of criticism for many people, Baker’s example highlights its potential usefulness.
“The public thinks it’s all bad,” Amy McGovern, principal investigator for the federal AI institute AI2ES, told the NYT. “But it’s actually giving scientists new ideas.”
James Collins, a biological engineering professor at MIT, also recently praised hallucinations for accelerating his research into new types of antibiotics. “We’re exploring,” he said. “We’re asking the models to come up with completely new molecules.”
On the topic of molecules, the Nobel Prize committee didn’t seem fazed by Baker’s use of AI, or its hallucinations, in his research. Instead, it said his team was able to produce “one imaginative protein creation after another.”
Baker hopes AI can help him create more proteins “that actually do things,” like break up the abnormal protein structures many believe to be associated with neurodegenerative diseases like Alzheimer’s.
