Gloria Mark

In the AI Story

Mark's methodology distinguished her work from earlier attention research that relied primarily on self-report or laboratory tasks. She embedded herself in real workplaces, using software to log every application switch, every email check, every transition between windows, while simultaneously observing workers and measuring physiological stress indicators. The combination produced a rich, ecologically valid picture of how knowledge workers actually use their attention in modern offices. Her findings consistently showed that workers switched tasks an average of every three minutes, that most switches were internally generated rather than externally imposed, and that the fragmentation produced measurable stress and performance degradation even when workers believed they were managing their attention competently.

The 23-minute recovery finding came from studies tracking not just when people returned to interrupted tasks but when they regained the performance levels they'd achieved before interruption. This temporal granularity revealed that resumption and recovery are distinct: people return quickly (often within seconds) but recover slowly (typically requiring tens of minutes). The distinction is critical for AI-augmented work, where the builder returns immediately to evaluate an agent's next output but operates with partially reconstructed context and accumulated residue from previous switches. She's back on task but not back at capacity.

Mark's research on multitasking and stress demonstrated that rapid task-switching produces measurable physiological stress responses — elevated heart rate, increased cortisol — independent of task difficulty. The stress arises not from the tasks themselves but from the switching: the cognitive overhead of disassembly and reassembly, the frustration of interrupted goals, the pressure of managing multiple simultaneous demands. AI tools, by increasing the frequency and diversity of switching, intensify these stress responses. The builder monitoring five agents experiences the switching stress of five simultaneous projects without the breaks that slower, pre-AI production provided between switches. The result is chronic stress activation that Mark's framework predicts will produce the exhaustion that Berkeley researchers and countless builders have reported.

Mark's 2025–2026 work has begun examining AI-specific attention dynamics, including how AI assistants fragment attention differently than conventional digital tools. Preliminary findings suggest that because AI tools respond instantly and produce high-quality outputs on demand, they create a compulsion loop stronger than email or social media: the builder feels pulled to check what the agent has produced, and each check generates cognitive demands (evaluation, decision-making) that prevent the attention restoration that conventional breaks might provide. The tool that should reduce cognitive load paradoxically increases it by creating more opportunities for the attention-fragmenting interactions that Mark's career has documented as systematically harmful.

Origin

Mark earned her PhD in psychology from Columbia University and joined UC Irvine's Department of Informatics in 2000. Her research program emerged from the collision between cognitive psychology and human-computer interaction, examining how digital tools reshape the temporal structure of knowledge work. The 23-minute finding appeared in her 2008 CHI conference paper and gained wide recognition through Cal Newport's citation in Deep Work and through media coverage in the 2010s documenting the attention crisis. Attention Span (2023) positioned her as a public intellectual on attention economics, and her ongoing AI research extends her framework to the most consequential attention-fragmenting technology yet deployed at scale.

Key Ideas

Naturalistic methodology. Mark's combination of activity logging, observation, and physiological measurement in real workplaces produced ecologically valid findings that laboratory studies could not replicate.

23-minute recovery time. The canonical finding that returning to performance after interruption takes 23:15 on average, documenting that resumption and recovery are temporally distinct processes.

Self-generated switching. Most task switches are internally initiated rather than externally imposed, suggesting that fragmentation arises from work structures that make switching feel necessary or rewarding.

Stress physiology. Rapid switching produces measurable stress responses independent of task difficulty, with chronic switching producing chronic stress activation that manifests as exhaustion and burnout.