This week at SXSW London, everyone was talking about AI. In a session titled “Have we lost control of our brains?”, Gloria Mark gave the blunt answer: yes. Mark, a psychologist at the University of California, Irvine, has spent about 30 years studying how people interact with digital technologies, and her message is that today’s attention drop is not just a nuisance. It is a measurable, compounding shift in how humans process information, and she thinks AI chatbots are pushing in the wrong direction.

Mark’s evidence starts with attention spans shrinking as technology becomes more woven into daily life. About two decades ago, she began asking how device use might affect attention. Using what she calls “living laboratories,” she monitored adult volunteers’ attention, mood, and behavior with sensors and trackers. In 2003, she found the average user could focus on one thing for around two and a half minutes. When she repeated the experiment in 2012, attention spans had shrunk to around 75 seconds on average. In research she conducted between 2014 and 2020, attention spans shrank further still to a mere 47 seconds on average.

If that sounds like a slow bleed, Mark argues it also comes with a sharper cost: stress and performance drag. She described experiments where participants wore heart rate monitors, and the results showed a direct correlation between switching attention quickly and stress rising. She also tied distraction to harder execution. When you keep flipping your attention, it takes longer to complete any single task. “It’s not great for performance,” she said, and “not great for our emotional well-being.” The punchline for leaders is uncomfortable: even if AI makes knowledge work feel faster, it may be training the conditions for slower, more stressful work.

Mark’s concern is not limited to adults. She connected adult attention research to the current regulatory and legal atmosphere around digital addiction. A few months ago, Meta, which owns Facebook and Instagram, and Google’s YouTube were ordered to pay millions of dollars in damages to a 20-year-old woman who accused the companies of creating products that led her to develop a childhood addiction. She also pointed to a recent settlement involving Meta, brought by a rural school district in Kentucky. That district accused Meta of designing addictive products harmful to students and sought more than $60 million to cover mental health costs. Mark noted that around 1,200 other school districts are taking similar legal action against social media companies.

Yet she did not frame digital tech as universally harmful. Social media can help some people build connections, including for marginalized groups. A 2024 survey of LGBTQ+ teenagers found that while some described rejection and fear, others described a sense of belonging, friendships, and identity development. Importantly, Mark said the evidence on broad impacts is inconclusive. “There have been lots and lots of studies, and the evidence is to date inconclusive,” she told me, adding that large, long-term studies might be needed to clarify what is really happening.

Where the AI story gets sharper is in the concept of “depth of processing,” which Mark described as the mental effort you put into evaluating and summarizing information. When people are actively engaged with content, they process it on a deep level, and that makes it more likely they learn it, understand it, and retain it. Her claim is that this depth disappears when most people use AI bots like ChatGPT, Claude, and Gemini to write, summarize, or evaluate for them. “You’re deferring your cognitive work to AI,” Mark said, and she views that as “not good for us.” The risk she points to is cognitive atrophy over time, comparing minds to muscles: if you do not constantly exercise them, they can atrophy.

The second-order implication she raised is that weaker critical thinking can make misinformation easier to accept. She also warned about AI-powered “synthetic companions,” arguing that relationships require time, effort, and understanding. If those ingredients are replaced by a sycophantic bot, the emotional intelligence muscle risks atrophying, and she referenced surveys that suggest emotional intelligence is already declining. She then stacked the broader trajectory: if things continue, she said attention spans diminish, loneliness rises, boredom rises, emotional intelligence decreases, and studies also suggest purpose is decreasing.

So what does course correction look like, according to Mark? She keeps coming back to effort. The more effort you put into something, the deeper the satisfaction. That means choosing the book over the skimmable summary. It means meeting friends in person when possible. It means trying not to use GPS in places you can probably manage without it. She said she “love[s] technology” and does not think we can give it up, but she argues we need new life routines that preserve active mental work. For executives building AI products or policies around them, the strategic stake is clear: if users increasingly outsource thinking, the benefits of AI might come paired with a long-term hollowing out of attention, judgment, and emotional skills, and the resulting social and regulatory pressure could be just as real as the performance promises.