Devi Shridhar warns AI could create a cognitive sedentary lifestyle

From screen time to thinking time

Shridhar’s streaming analogy is not an argument that every new technology is harmful. It is an argument about what happens when convenience becomes the default.

As countries become wealthier, more people move from physical work to sedentary jobs and spend more of their free time inactive. The same relationship appears in low- and middle-income countries: as GDP per capita rises, physical activity falls.

The health costs are now familiar:

Exercise has consequently become a separate health-maintenance activity. In countries such as the UK, it is increasingly associated with higher socioeconomic status.

Shridhar sees a parallel forming around AI. Using AI and average daily screen time are both rising, while students increasingly rely on generative systems to produce assessed work. She assumes that almost every student at her university uses some form of generative AI when preparing assignments. That makes it harder to know what students actually understand.

A 2026 Higher Education Policy Institute survey found that:

The wider education data is moving in an uncomfortable direction. A recent OECD report found that student performance in member countries continued to fall, with average mathematics and reading scores at their lowest levels since records began. The report’s authors linked weak results to more screen time, greater digitisation of lessons and the disappearance of children’s habit of reading for pleasure.

Those figures do not prove that AI caused the decline. They do show why replacing intellectual effort with automation is not a neutral change.

What the brain may lose

The early evidence is limited, but it points toward a specific concern: outsourcing thought could change how intensely and broadly the brain engages with a task.

In the study Your Brain on ChatGPT, researchers measured electrical brain activity using electroencephalography, or EEG. Students who wrote essays with generative AI showed weaker and less widely distributed connections between brain regions than students who worked entirely independently.

That result is preliminary, not a verdict on AI. But it sits beside a broader body of evidence about mental stimulation. A study of more than 19,000 adults over 50 found that regular participation in activities that stimulate thinking, such as crosswords or Sudoku, was associated with a lower risk of dementia and better cognitive function later in life.

The basic analogy is straightforward. Older adults need strength training to preserve muscle mass. Shridhar argues that the brain also needs regular work: without use, its abilities may decline.

The unanswered issue is not whether AI can complete a task. It is whether people will continue doing enough of the difficult parts themselves to preserve concentration, reading, writing, problem-solving and social interaction.

The incentive problem

I think Shridhar’s strongest point is less about neuroscience than incentives. The companies building these systems benefit when more human activity can be automated, and organisations looking to cut costs benefit for the same reason. Neither has an obvious commercial reason to prioritise the long-term cognitive effects of reduced effort.

That does not make the technology inherently harmful. It does mean that claims of universal benefit should be treated carefully. A tool that makes work easier for an individual may also make institutions less willing to pay for human judgment, teaching or attention.

What I would want to know is not only whether AI helps people perform better today, but what happens after years of using it for the parts of thinking that once provided daily practice. The risk is not a dramatic moment when people suddenly stop thinking. It is a gradual shift in what counts as necessary mental work—and, eventually, in who still has the ability to do it unaided.