Time pressure makes people take mental shortcuts and miss AI mistakes, cutting the accuracy of human–AI teams; task complexity and erroneous AI advice further harm performance, with interactions that diverge from classic dual-process expectations.

Main Finding

Time pressure pushes humans from systematic to heuristic processing when evaluating AI outputs, reducing their ability to distinguish correct from faulty AI advice and thereby lowering human–AI team performance. While both greater task complexity and higher AI faultiness also independently degrade performance, their interactions with time pressure are more nuanced than standard dual-process predictions.

Key Points

• Experimental evidence (n = 228) shows a clear cognitive shift under time pressure consistent with the heuristic-systematic model.
• Under time pressure participants:
  • Rely more on heuristics (shortcuts) and less on effortful analytic evaluation.
  • Exhibit reduced discrimination between correct and incorrect AI responses, which lowers accuracy.
• Task complexity and AI faultiness each independently reduce human–AI performance.
• Interactions:
  • The combined effects of time pressure with complexity or AI faultiness are not simply additive; some conditions show attenuated or unexpected interaction patterns relative to standard theoretical expectations.
• Practical consequence: design features or workflows that reduce time pressure or support systematic evaluation (e.g., clear explanations, confidence indicators, decision aids) can improve outcomes.
• Limitations (noted by study): online setting, task types constrained to knowledge-evaluation tasks, and effect sizes/statistical details not reported here.

Data & Methods

• Theoretical framing: dual-processing frameworks, specifically the heuristic–systematic model.
• Design: online between-subjects experiment with two main treatments:
  • Time pressure: participants had limited time to evaluate AI responses.
  • Control: participants had unlimited time.
• Manipulations within tasks:
  • Task complexity varied across items.
  • AI output correctness varied (correct vs. faulty responses).
• Sample: n = 228 participants.
• Outcome measures:
  • Accuracy/performance in evaluating AI responses.
  • Ability to discriminate between correct and incorrect AI advice (presumably measured via hit/false-alarm rates or similar metrics).
• Main analytic approach: comparison of performance across treatment cells to test for main effects and interactions (specific statistical tests and effect sizes not provided in text).

Implications for AI Economics

• Valuation and deployment of AI tools:
  • Productivity gains from AI depend on task context and time constraints; payoff estimates should account for degraded human–AI performance under time pressure.
  • Cost–benefit analyses of automation should include cognitive frictions (time pressure-induced heuristic use) that reduce effective accuracy.
• Design and productization:
  • Product features that mitigate heuristic reliance (fast, salient explanations; confidence scores; prioritized highlighting of critical errors) increase the effective value of AI in time-pressured settings.
  • UI/UX and workflow design can be economically valuable—investments that reduce decision time pressure or scaffold quick systematic checks can raise output quality.
• Labor markets and task allocation:
  • Tasks with tight time constraints may see less complementarity between AI and workers unless interfaces or team processes mitigate heuristic shifts; this influences substitution vs. augmentation dynamics.
  • Assignment algorithms and staffing models should incorporate time-pressure sensitivity: allocate more time or human review to high-complexity or high-stakes tasks.
• Incentives, contracts and regulation:
  • Contracts and performance metrics should account for error amplification under time pressure; incentives for speed may reduce overall output quality when AI is involved.
  • Regulatory or audit regimes for AI use in high-pressure domains (healthcare triage, financial trading, emergency response) should mandate safeguards that counteract time-pressure effects.
• Modeling suggestions for AI economics research:
  • Incorporate cognitive-state parameters (e.g., propensity to use heuristics under time constraints) into models of human–AI team productivity and adoption.
  • Consider non-linear interactions between time pressure, task complexity, and AI reliability when forecasting adoption, pricing, or welfare impacts.
• Policy recommendation:
  • Encourage workplace practices that either (a) provide sufficient evaluation time for AI-assisted decisions or (b) supply explainability/confidence tools that allow near-systematic processing under time constraints.

Overall, the study suggests that accurate economic assessment of AI technologies must account for human cognitive dynamics under time pressure, as these materially affect realized benefits, optimal design choices, and policy prescriptions.