Brian Christian has spent 20 years thinking about how artificial intelligence will change our minds and our world. His 2020 book The Alignment Problem<a href="https://www.nytimes.com/2024/01/31/books/review/artificial-intelligence-best-books.html” rel=”nofollow noopener” target=”_blank”>has been praised as one of the best on AI, and like many scientists, he sometimes feels euphoric about the technology’s advances. After all, when everyone has a fleet of autonomous agents on standby to speed up research, it’s hard not to feel as if a life-changing discovery is just a few prompts away.
He also keeps a notebook on his desk and writes in it every day. The practice keeps him grounded and tamps down a gnawing suspicion that too much time working with AI could cause him to lose his own scholarly edge.
He has good reason to worry — and need only look to his most recent research for proof.
Christian was part of a team behind a landmark study that found that spending just 10 minutes relying on an AI tool significantly impairs our ability to focus and persist with difficult tasks. A preliminary draft made waves earlier this year, and the team presented the updated, peer-reviewed paper this week at the Conference on Language Modeling.
“It’s a striking finding, but in a way it supplies ammunition for a story that a lot of us kind of feel in our gut,” said Christian, a research fellow with the UC Berkeley Center for Human-Compatible AI. “I experience it too.”
To test how quickly people’s focus and persistence plummeted, researchers designed a series of randomized controlled trials and recruited 1,222 participants online.
For the first experiment with 354 participants, one group was tasked with solving 15 basic fraction problems without help, while another had ChatGPT open on the side to use for assistance and to even ask for the answer. The group with AI access started off more accurate than the group without it. Then, after 12 problems, researchers took away the AI tool. Almost immediately, the people in that group stopped solving questions accurately.
A similar thing happened when they ran the experiment on a second, larger group of 667 participants. Whereas the AI users got answers wrong or gave up entirely when the AI assistance was withdrawn, the group without AI stuck with the task and were more successful at the end.
Researchers then ran a third experiment with 201 participants and an SAT reading comprehension prompt. The results echoed those from the fraction problems: When the AI tool was removed, persistence and accuracy dropped.
“You almost couldn’t ask for a clearer story,” Christian said.
The findings backed up what anecdotes and surveys have increasingly found in recent years: An overreliance on AI for answers erodes our cognitive fitness. When AI produces answers in seconds, we get used to that speed; anything that takes longer feels inefficient. And relying on AI also removes what the team called “productive struggle,” which can lead people to find their research strengths and passions.
“We’ve built these systems to be helpful, but are they being helpful to us in a way that’s actually helping us?” Christian said. “We showed, ironically, that they are often helping us in ways that are kind of unhelpful.”
The team included scholars from Carnegie Mellon University, the Massachusetts Institute of Technology, the University of Oxford and UCLA. Outlets from WIRED to Psychology Today covered the study when early findings were released in the spring. The New York Timespublished a story last week about the paper and similar research.
Frontier AI companies have acknowledged how offloading tasks to chatbots can reduce one’s baseline understanding and mastery of a subject. Instead of defaulting to quick answers, AI responses could default to being more instructive, like a tutor, Christian said.
“We need to work toward a future in which human abilities are, as much as possible, augmented rather than supplanted,” he said.
Christian said that’s especially true with research and academia. He worries about the way overreliance on AI tools in science could jeopardize what makes for a good scientist in the first place: staying close to data, being immersed in the material, and working as part of a team. Something significant could be lost when that team of 10 developing the ideas that lead to a breakthrough is whittled down to two or three researchers and a chatbot.
“This is not a story about fractions and SAT problems,” Christian said. “This is something that is happening to human knowledge and human expertise from top to bottom and from grade-school students all the way to the leading experts in the world.”
