It selects, sustains, and reallocates processing among competing signals; it is shaped by goals, salience, and selection history.
Psychology & Human Performance · AI and attention
The Attention Crisis
AI, overload, and the collapse of focus.
A psychology and performance brief on attention fragmentation, AI acceleration, and practical discipline.
Core thesis: AI does not create a new human psychology. It accelerates existing attention vulnerabilities by increasing the volume, velocity, salience, and actionability of cognitive inputs.
PsychologyAIPerformanceDigital Discipline
Executive brief
The modern mind is not weak. It is outnumbered.
Attention is not one mental thing. It is a constrained control system that selects, sustains, and reallocates processing among competing signals. Digital overload works by multiplying the number of signals. AI intensifies the pattern by making each signal more actionable.
The practical answer is not abstinence. It is design discipline: fewer interruptive inputs, protected cognitive windows, purposeful AI sessions, verification rules, and team norms that treat focus as infrastructure.
- Attention is a constrained control system shaped by goals, salience, selection history, working memory, cognitive load, and residue.
- The strongest evidence does not say screens destroyed attention universally; it says interruptions, notification streams, and unguided multitasking reliably degrade many immediate tasks.
- AI accelerates the attention problem by increasing the volume, velocity, salience, and actionability of cognitive inputs.
- AI productivity gains can coexist with fragmentation because faster generation often creates more review, verification, and decision burden.
- The discipline response should be structural: reduce interruption load, batch AI use, protect decision windows, and train verification.
Even after the visible interruption ends, part of cognition can remain attached to the prior task, making re-entry costly.
More drafts, summaries, recommendations, and candidate actions mean more scanning, triage, and verification.
The best evidence is not anti-technology in general; it is anti-ambient, poorly timed, unstructured interruption.
AI can increase bounded-task output while increasing oversight, review burden, and shallow acceptance risk.
Friction, defaults, focus windows, communication norms, and verification protocols matter more than motivational slogans.
AI lowers the cost of generation, but raises the burden of judgment.
How attention actually works
Focus is a control system under pressure.
The research foundations converge on a practical model: attention is pulled by intention, salience, history, limited working memory, task load, and the residue of unfinished work.
Top-down attention keeps task goals active and suppresses irrelevant inputs.
Badges, alerts, motion, novelty, and social cues compete for priority.
Past reward and repeated checking make some signals easier to select again.
The inner workspace maintains goals, context, and subtask state.
Switching, searching, decoding, and verifying can consume capacity before the real task begins.
Unfinished, interrupted, or time-pressured tasks keep part of attention behind.
Digital overload before AI
The attention economy was already expensive.
The evidence does not support a cartoon claim that all screens ruin attention. It supports a sharper claim: frequent, salient, poorly timed interruptions and unstructured multitasking degrade many real tasks.
Microsoft 365 reporting describes an average employee spending more time communicating than creating.
The productive work share gets squeezed when search, meetings, pings, and status loops expand.
Microsoft describes a high share of unscheduled or ad hoc meetings in the modern workday.
Workers in Microsoft reporting said they spent too much time searching for information.
How AI accelerates fragmentation
More intelligence can create more cognitive surface area.
AI can reduce search and drafting burden, but it can also create more objects that require attention, verification, response integration, and stewardship.
- AI lowers generation cost.
- More drafts, summaries, suggestions, alerts, and candidate actions appear.
- Salience rises while the number of decisions expands.
- Task switching, review load, and verification fatigue increase.
- Resumption costs and attention residue accumulate.
- Output may rise, but deep judgment can weaken.
The solution is not less intelligence. It is better discipline around intelligence.
| AI or platform feature | Why it fragments attention | Likely consequence | What to measure |
|---|---|---|---|
| Push alerts, badges, mentions | External attentional capture and social pressure. | More switching and more strain. | Notifications per hour, response latency, perceived strain. |
| Auto-summaries and inline suggestions | More candidate artifacts to scan, accept, reject, or revise. | Faster throughput but more micro-evaluation. | Acceptance rate, edit rate, error rate, review time. |
| AI search and retrieval | Search friction falls while verification burden rises. | Less time gathering; more time validating. | Search time saved versus verification time added. |
| Personalized recommender feeds | Salience is optimized by prior behavior and engagement. | Rabbit holes, compulsive checking, reduced agency. | Session starts, session length, non-profiled feed opt-outs. |
| Prompt rewriting and auto-optimization | The system can distance users from their own intent. | Worse task-model fit in some tasks. | Quality by mode, override frequency, user intent mismatch. |
| Ambient AI inside chat or email | Cognition is embedded inside already interruptive channels. | Always-on partial engagement. | After-hours messages, ad hoc meeting load, focus-time hours. |
Evidence matrix
The strongest evidence is specific, not theatrical.
The record is serious but conditional. Interruption timing, notification frequency, smartphone structure, and verification design matter more than blanket screen panic.
| Topic | Study or source | Design | Main finding | Analytic takeaway |
|---|---|---|---|---|
| Media multitasking | Wiradhany & Nieuwenstein, 2017 | Meta-analysis | Weak overall association after small-study corrections. | Broad claims should be cautious. |
| Sustained attention | Rioja et al., 2023 | Three-sample study | Medium negative link between media multitasking and sustained attention. | Sustained attention appears more vulnerable than some other constructs. |
| Smartphone presence | Parry, 2024 | Systematic review and meta-analysis | Reliable pooled effect mainly for working memory. | The brain-drain effect is not universal across all outcomes. |
| Work notifications | Ohly et al., 2023 | Field experiment | Fewer notification interruptions improved performance and lowered strain. | Reduce pings before training willpower. |
| Interruption timing | Hirsch et al., 2025 | Experiment | High-workload timing increased resumption costs. | Protect high-load moments, not only total hours. |
| Decision quality | Wu, Gao, and Liu, 2023 | Experiment | Interruptions during evaluation and selection harmed performance. | Protect decision windows. |
| Classroom devices | Deng et al., 2025 | Two randomized classroom trials | Unguided smartphone use was worse than bans; guided use performed best. | Structure beats laissez-faire and may beat blanket prohibition. |
| Mobile internet blocking | Castelo et al., 2025 | Randomized trial | Blocking mobile internet improved well-being, mental health, and sustained attention. | Connectivity itself can be a causal part of the problem. |
Performance, learning, creativity, and decision quality
AI can improve throughput while weakening depth.
The key trade is not simple productivity versus distraction. It is whether the workflow preserves enough cognitive room for recall, judgment, creativity, and review.
Fragmentation decomposes work into costly episodes. Work time may remain high while context-rich processing falls.
Off-task and unstructured phone use damages recall and academic performance, while guided classroom use can improve outcomes.
AI can lift individual creative output while making collective outputs more similar unless users bring strong metacognitive strategy.
The key risk is not only hallucination. It is accepting plausible outputs without enough contextual evaluation.
Practical discipline framework
Build an Attention Defense System.
Environmental controls beat self-control alone. The discipline system has to operate at the individual, team, and organizational levels.
- Disable nonessential push alerts.
- Batch communication checks.
- Block mobile internet during focus windows.
- Use AI in deliberate sessions, not ambient grazing.
Measure: Notifications/hour, uninterrupted focus minutes, verification time.
- Create communication windows and protected deep-work blocks.
- Define a narrow urgent channel.
- Use AI review protocols for drafting, coding, and decisions.
- Separate generation meetings from evaluation meetings.
Measure: Ad hoc meeting share, after-hours messages, rework rate.
- Run interruption audits.
- Consolidate duplicate tools and channels.
- Approve fewer AI systems and integrate them better.
- Train AI literacy, verification, and confidence calibration.
Measure: Tool count, channel count, search burden, error severity.
Member research
Unlock the full Attention Crisis report with Basic.
Inspect the evidence matrix, AI fragmentation model, discipline framework, population differences, policy implications, and selected references.
- Evidence matrix
- AI feature comparison
- Attention Defense System
- Policy implications
- Selected references
Basic unlocks the full online report, full books, and all interactive research reports.
Optional PDF copy is separate. Basic unlocks the online report now.