nabu.science
MethodologyAboutContact

 

  • Snapshot
  • Key Claims
  • Strengths & Limitations
  • Commentary
  • From the Authors
  • Author Note / Field Feedback
  • Dimension breakdown
  • Impact Profile
Free · About 5 minutes
Report #NBU-EBB72A08·Framework v5.8·Pipeline v0.4·Evaluated 21 Aug 2026·OECD: 5.1 Psychology·Subfield: Safety Research·Basis: Full text·Type: Laboratory experiment

Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender

Steven D Shaw et al.Steven D Shaw, Gideon Nave·2026·10.2139/ssrn.6097646 ↗

Primary claim

Relative to baseline (no System 3 access), accuracy significantly rose when AI was accurate and fell when it erred (+25/-15 percentage points; Study 1), the behavioral signature of cognitive surrender

Supported · randomized accuracy contrasts establish AI-dependent performance, but minimal internal scrutiny is inferred rather than directly measured
In brief

A rigorous multi-study experiment shows reasoning tracks AI accuracy, but its cognitive-surrender construct exceeds what consultation and adoption measures directly identify.

ReliabilityCaveats4 of 4 checks completeRead the entry ↓
Likely to hold — the claim's weakest requirement is that the effect exists at all. Experimental. Every point below is traceable to a quoted passage, but this forecast has not yet been validated against known replication outcomes.
Abstract

People increasingly consult generative artificial intelligence (AI) while reasoning. As AI becomes embedded in daily thought, what becomes of human judgment? We introduce Tri-System Theory, extending dual-process accounts of reasoning by positing System 3: artificial cognition that operates outside the brain. System 3 can supplement or supplant internal processes, introducing novel cognitive pathways. A key prediction of the theory is "cognitive surrender"-adopting AI outputs with minimal scrutiny, overriding intuition (System 1) and deliberation (System 2). Across three preregistered experiments using an adapted Cognitive Reflection Test (N = 1,372; 9,593 trials), we randomized AI accuracy via hidden seed prompts. Participants chose to consult an AI assistant on a majority of trials (>50%). Relative to baseline (no System 3 access), accuracy significantly rose when AI was accurate and fell when it erred (+25/-15 percentage points; Study 1), the behavioral signature of cognitive surrender (AI-Accurate vs. AI-Faulty contrast; Cohen's h = 0.81). Engaging System 3 also increased confidence, even following errors. Time pressure (Study 2) and per-item incentives and feedback (Study 3) shifted baseline performance but did not eliminate this pattern: when accurate, AI buffered time-pressure costs and amplified incentive gains; when faulty, it consistently reduced accuracy regardless of situational moderators. Across studies, participants with higher trust in AI and lower need for cognition and fluid intelligence showed greater surrender to System 3. Tri-System Theory thus characterizes a triadic cognitive ecology, revealing how System 3 reframes human reasoning and may reshape autonomy and accountability in the age of AI.

Key Claims

Primary finding · Supported

Relative to baseline (no System 3 access), accuracy significantly rose when AI was accurate and fell when it erred (+25/-15 percentage points; Study 1), the behavioral signature of cognitive surrender

Likely to hold — the claim's weakest requirement is that the effect exists at all. Experimental. Every point below is traceable to a quoted passage, but this forecast has not yet been validated against known replication outcomes.

randomized accuracy contrasts establish AI-dependent performance, but minimal internal scrutiny is inferred rather than directly measured

Abstract; Study 1 Results, Accuracy
Quote verified
Secondary · Well-supported

Time pressure (Study 2) and per-item incentives and feedback (Study 3) shifted baseline performance but did not eliminate this pattern

the AI-Accurate versus AI-Faulty gap persisted across both randomized moderator studies and the pooled synthesis

Abstract; Studies 2–3 Results; General Discussion, Cognitive Surrender Effect Size
Quote verified
Secondary · Supported

Engaging System 3 also increased confidence, even following errors

AI access increased confidence, though error-specific confidence was isolated more directly in Study 3 than Study 1

Abstract; Study 1 Results, Confidence; Study 3 Results, Confidence
Quote verified
Secondary · Supported

Across studies, participants with higher trust in AI and lower need for cognition and fluid intelligence showed greater surrender to System 3

pooled moderator analyses support the pattern, but individual-difference evidence is correlational and uneven across separate studies

Abstract; General Discussion, Individual Differences in Susceptibility to Cognitive Surrender
Quote verified

Derived from the full evaluation — not a separate score.

Strengths · Limitations

Strengths

#methodological rigour

Randomized evidence isolates AI dependence

Study 1 compares Brain-Only performance with AI-Accurate and AI-Faulty trials, while Studies 2 and 3 reproduce the accuracy gap under time pressure and incentives with feedback.

↳ Study 1, Figure 3; Study 2, Figures 5–6; Study 3, Figures 8–9

#contribution

Core effect replicates across studies

The preregistered trial-level synthesis reports a large pooled association between AI correctness and participant correctness, with similar Cohen's h values across all three studies.

↳ General Discussion, Cognitive Surrender Effect Size; Figure 10

#positioning

Prior literature is substantively engaged

The theoretical and discussion sections distinguish the proposal from cognitive offloading and automation bias while addressing both algorithm aversion and algorithm appreciation.

↳ Tri-System Theory of Cognition; Mechanisms Underlying Cognitive Surrender

Limitations

#methodological rigour

Internal cognition remains indirectly inferred

Cognitive surrender is defined as minimal scrutiny and absent deliberation, but the experiments measure consultation, adoption, override, accuracy, and confidence rather than whether internal reasoning occurred.

↳ Table 2; Common Procedures; Cognitive Surrender vs. Cognitive Offloading

#contribution

Broad theory exceeds task evidence

Claims about a triadic cognitive ecology and altered autonomy extend beyond seven adapted CRT items, short exposures, and one experimentally controlled AI system.

↳ General Discussion; Societal Implications; Limitations and Future Research Directions

#application pathway

Real-world translation remains preliminary

The AI was forced through hidden prompts to provide correct or confidently incorrect answers, and the proposed interface and policy interventions were not tested in naturalistic settings.

↳ Empirical Overview; Societal Implications; Limitations and Future Research Directions

Does this match your read? Add your assessment ↓

Nabu Commentary

The strongest evidence is the repeated randomized contrast showing that participant accuracy tracks whether AI advice is correct or faulty, supported by the trial-level synthesis and consistent per-study effect sizes. Preregistration, Brain-Only comparisons, repeated-trial modeling, and replication across settings support a quality score in the competent-execution band. The main downward pressure comes from the definition of cognitive surrender as absent deliberation, because the experiments observe behavior but do not directly identify the internal process distinguishing surrender from assisted reasoning. The controlled CRT paradigm and forced AI-error manipulation also limit transfer to the broad societal domains discussed.

Evaluate another paper

Free · About 5 minutes · Research preview

Author Note / Field Feedback

Nabu’s assessment, alongside the field’s view.

Are you an author of this paper?

Dimension breakdown

Evidence Score

Out of 5
012345
3.6±0.2

Contribution
3.8

The randomized AI-accuracy paradigm and replication across three experiments meaningfully extend evidence on human reliance on AI advice. The empirical advance is clearer than the claim that Tri-System Theory establishes a distinct cognitive architecture.

Abstract; Empirical Overview; General Discussion

accuracy significantly rose when AI was accurate and fell when it erred

Methodological Rigour
3.6

Randomized trial type, Brain-Only comparisons, preregistration, and mixed-effects models fit the repeated-trial behavioral question. The design does not directly observe the absence of deliberation required by the paper's definition of cognitive surrender.

Common Procedures; Statistical Analysis; Table 2; Cognitive Surrender vs. Cognitive Offloading

Minimal internal engagement.

Reporting
3.8

The paper presents a traceable theory-to-hypothesis-to-study sequence and reports its main contrasts with effect sizes and confidence intervals. Its abstract and conclusion sometimes convert observed consultation and adoption into stronger claims about bypassed internal cognition.

Abstract; Table 2; General Discussion; Conclusion

often surrender to its outputs

Positioning
3.5

The discussion engages dual-process criticism, cognitive offloading, automation bias, algorithm aversion, and algorithm appreciation. Its limitations identify the controlled setting, CRT specificity, and single exposure, although the broader societal claims are not always narrowed accordingly.

Mechanisms Underlying Cognitive Surrender; Societal Implications; Limitations and Future Research Directions

may limit generalizability to real-world settings

Reliability

Caveats4 of 4 checks


Internal CoherenceNoted

The principal numerical results are coherent, but the reported synthesis sample has a minor unresolved arithmetic discrepancy and the interpretation of adoption as minimal-scrutiny surrender is not fully identified by the observed measures.

Research ConductClean

IRB approval, informed consent, debriefing, preregistration, and OSF availability are declared in the Data Collection and Empirical Overview sections. External materials were not supplied for verification, which is an assessability gap rather than a conduct concern.

Flags: 3 declared / 5 total

Reference IntegrityClean

56 of 56 checkable references verified

59 references in manuscript 3 have no canonical index record — counted, but not index-checkable 4 references confirmed by manual review

References not verified automatically are resolved manually within 24 hours.

Post-Publication RecordClean

No retraction or expression of concern found in Crossref (includes Retraction Watch).

Last checked: 17 Oct 2026

Sources: Crossref (incl. Retraction Watch) ✓

Impact Profile

Impact areaSociety and culture · SDG 4
Type of advanceConceptual
Route to useNamed
AdvanceAdvances
CitationsToo early
Policy · patentNot yet traced
ReachedNot reachedNot assessable or unverified
ReadinessControlled
Foundational
Controlled
Piloted
Validated
Implementable
FoundationalImplementable

Three preregistered experiments pinned down the cognitive surrender effect under controlled laboratory and online conditions, but the authors note generalizability to real-world settings is untested.

From the paper
“we outline practical implications for designers and policymakers concerned with AI's ethical and practical integration into society”

The paper names AI designers and policymakers as actors who should apply its findings on cognitive surrender to interface design and digital literacy policy.

Pathway

AI interface designers and digital literacy educators would build uncertainty cues, feedback, and calibrated-trust features into AI tools to reduce uncritical adoption of faulty outputs.

Barrier

The evidence rests on a single lab task (the CRT) with experimentally manipulated AI accuracy, and the authors themselves call for field studies before real-world application.

Evaluated 12 Nov 2026 · GBD 2023 · glm_v1_3_3 · impact_profile_v1_3_3

Report an issue →

AI-generated, human-governed. Something look off? Contact us to request a review.

Keep reading

Related evaluations

Other papers Nabu has evaluated blind, starting with the same field.

  1. The Ethics of Prompt Engineering

    Germani, Spitale · 2026 · Philosophy, ethics and religion

    Primary claim

    prompting should therefore be understood as an epistemic and normative act

    Supported
    Evidence Score
    3.2±0.2
    Reliability
    Caveats
  2. When AI Reviews Train AI Reviewers: Scientific-Judgment Collapse and Mitigation

    Ho, Liu, Tran, et al. · 2026 · Computer and information sciences

    Primary claim

    After one recursive step, increasing exposure from 0% to 100% makes recommendation ratings more concentrated, independently generated reviews of the same paper about 11% more similar, and reviews across different papers about 5% less diverse.

    Supported
    Evidence Score
    3.6±0.2
    Reliability
    No concerns
  3. Neighborly social pressure and collective action: Evidence from a field experiment in Tunisia

    Jöst · 2024 · Political science

    Primary claim

    I do not find evidence to suggest that neighborly social pressure encourages participation in neighborhood cleanups, with low participation rates both for the neighbor and outsider contact conditions.

    Tentative
    Evidence Score
    3.2±0.2
    Reliability
    Caveats

How the rubric works →