A new study from researchers at MIT and Princeton AI Lab, published via arXiv, puts hard numbers behind something many of us have quietly suspected. People are not only underestimating how often they lean on AI, they’re also dramatically overestimating what they get out of it. The researchers call it the “efficiency-gain illusion,” and it describes a cognitive trap that could reshape how we think about AI’s actual contribution to productivity.
The numbers behind the illusion
The study, titled “The efficiency-gain illusion: People underestimate the rate of AI use and overestimate its benefits on simple tasks,” ran three pre-registered experiments with a combined 2,691 participants. The tasks were deliberately basic: arithmetic, spell-checking, the kind of work most people can do without breaking a sweat.
Participants consistently believed AI was saving them meaningful time and effort on these simple tasks, even when the actual gains were marginal. In one modeled analysis, using a copy-paste function with AI reduced the average completion time from 102.0 seconds to 66.2 seconds. Participants perceived the benefit as being far greater than that 35-second reality. Their subjective sense of efficiency gains surpassed what actually happened, creating a distorted picture of AI’s usefulness that then informed their future decisions about when to deploy it.
The feedback loop problem
Researchers identified a feedback loop where initial reliance on AI for simple tasks encouraged further reliance, which deepened the misjudgments about productivity. Each time a participant used AI and felt like it helped, they became more likely to reach for it again. Not because the evidence supported doing so, but because the feeling of efficiency became self-reinforcing.
The researchers also found systematic underestimation of AI usage rates among participants. People didn’t just overestimate the benefits. They also underestimated how frequently they were turning to AI in the first place, which makes the feedback loop even harder to break.
The productivity paradox, revisited
What the MIT and Princeton research adds is a behavioral explanation for part of this gap. If individual users are systematically overestimating AI’s benefits on routine tasks, then the collective productivity data may never match the collective enthusiasm. The gains feel real at the individual level, but they don’t fully materialize in the numbers.