Researchers from Microsoft and Carnegie Mellon University recently published a study looking at how using generative AI at work affects critical thinking skills.
“Used improperly, technologies can and do result in the deterioration of cognitive faculties that ought to be preserved,” the paper states.
When people rely on generative AI at work, their effort shifts toward verifying that an AI’s response is good enough to use, instead of using higher-order critical thinking skills like creating, evaluating, and analyzing information. If humans only intervene when AI responses are insufficient, the paper says, then workers are deprived of “routine opportunities to practice their judgment and strengthen their cognitive musculature, leaving them atrophied and unprepared when the exceptions do arise.”
In other words, when we rely too much on AI to think for us, we get worse at solving problems ourselves when AI fails.
In this study of 319 people, who reported using generative AI at least once a week at work, respondents were asked to share three examples of how they use generative AI at work, which fall into three main categories: creation (writing a formulaic email to a colleague, for example); information (researching a topic or summarizing a long article); and advice (asking for guidance or making a chart from existing data). Then, they were asked if they practice critical thinking skills when doing the task, and if using generative AI makes them use more or less effort to think critically. For each task that respondents mentioned, they were also asked to share how confident they were in themselves, in generative AI, and in their ability to evaluate AI outputs.
About 36% of participants reported that they used critical thinking skills to mitigate potential negative outcomes from using AI. One participant said she used ChatGPT to write a performance review, but double checked the AI output for fear that she could accidentally submit something that would get her suspended. Another respondent reported that he had to edit AI-generated emails that he would send to his boss — whose culture places more emphasis on hierarchy and age — so that he wouldn’t commit a faux pas. And in many cases, participants verified AI-generated responses with more general web searches from resources like YouTube and Wikipedia, possibly defeating the purpose of using AI in the first place.
In order for workers to compensate for the shortcomings of generative AI, they need to understand how those shortcomings happen. But not all participants were familiar with the limits of AI.
“Potential downstream harms of GenAI responses can motivate critical thinking, but only if the user is consciously aware of such harms,” the paper reads.
In fact, the study found that participants who reported confidence in AI used less critical thinking effort than those who reported having confidence in their own abilities.
While the researchers hedge against saying that generative AI tools makes you dumber, the study shows that over reliance on generative AI tools can weaken our capacity for independent problem-solving.
