What Science Has to Say About How to Use AI -

Relying too much on generative AI tools can negatively impact creativity and even cognitive ability, according to new research

Relying too much on generative AI tools for certain tasks can negatively impact creativity and even cognitive ability, according to new research.

Working the right way with AI is like borrowing 40 points of IQ,” wrote Greg Shove, CEO of Section, a New York–based business education company focused on AI, in a post on the company’s website. But the evidence is mounting that using AI the wrong way — by leaning too heavily on it — can reduce brainpower, memory, and motivation.

Some of that evidence came earlier this year in a working paper, “Your Brain on ChatGPT,” released by Nataliya Kosmyna, a research scientist at MIT Media Lab, which reported the results of a study that measured the brain activity of three groups of students as they wrote essays over a period of months. The first group worked on their own, the second had access to Google search, and the third used ChatGPT. Researchers found the brain-only group showed the strongest neural connectivity and the group using ChatGPT registered the lowest brain activity of all three groups — and that activity dropped over time. The study’s core takeaway, according to the MIT Media Lab, is that “overreliance on AI may reduce brain activity and memory recall.”

Other studies have shown similar results, including a joint Microsoft and Carnegie-Mellon study, cited by Shove, that showed that while gen AI can accelerate routine tasks, it also can reduce our ability to exercise critical thinking. Research that appeared in a recent Harvard Business Review story found that using gen AI for tasks like brainstorming ideas and writing emails can improve task performance, but workers reported being bored when returning to tasks without AI.

Claim or renew your subscription to Convene.
Want deep-dive insights on events delivered to your inbox? Sign up for our newsletters.

How to Stay Sharp

The good news is that the MIT essay-writing experiment points toward a remedy: When the study participants who had relied on their brains only later used ChatGPT, their brains showed a burst of connectivity. “The order of tool use matters,” wrote Jiunn-Tyng Yeh, a researcher at Duke University, in a LinkedIn post summing up the research. “Starting with one’s ideas and then layering AI support can keep neural circuits firing on all cylinders, while starting with AI may stunt the networks that make creativity and critical reasoning uniquely human.”

Thinking first is a point that Junior Tauvaa, PCMA’s chief business officer, makes about PCMA’s AI platforms. “Before jumping into Spark or Destinaitor, stop and ask yourself: What’s my real objective — engagement, new audiences, revenue, or longer content life? Upload your questions, then keep questioning. The more you guide the AI, the more it works for you.”

Tauvaa’s recommendations mirror the advice that Shove gives to those who aspire to be “AI drivers,” as opposed to “AI passengers” who “happily delegate their cognitive work to AI.” Drivers, he wrote, “will insist on directing AI. They’ll do some original thinking themselves, use AI to pressure-test their assumptions, and rigorously check what it says. They won’t accept AI-generated results they can’t validate. They’ll look to AI for a conversation — not the answer.”

Barbara Palmer is Convene’s deputy editor.