CITATION — REFERENCE ENTRY
Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender — PsyArXiv
- Key
- shaw-nave-2026-tri-system-theory
- Authors
- Shaw, Steven D.; Nave, Gideon
- Issued
- 2026-1-12
- Type
- article
- Container
- PsyArXiv
- Publisher
- PsyArXiv
Raw CSL JSON
{
"DOI": "10.31234/osf.io/yk25n_v1",
"URL": "https://osf.io/preprints/psyarxiv/yk25n_v1",
"note": "Manuscript version v20260111; also distributed as an SSRN working paper (abstract 6097646).",
"type": "article",
"genre": "Preprint",
"title": "Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender",
"author": [
{
"given": "Steven D.",
"family": "Shaw"
},
{
"given": "Gideon",
"family": "Nave"
}
],
"issued": {
"date-parts": [
[
2026,
1,
12
]
]
},
"publisher": "PsyArXiv",
"container-title": "PsyArXiv"
}
Claims
-
Shaw and Nave define cognitive surrender as the tendency to defer judgment, effort, and responsibility to an AI system's output, especially when that output is delivered fluently, confidently, or with minimal friction.
"We define cognitive surrender as the behavioral and motivational tendency to defer judgment, effort, and responsibility to System 3's output, particularly when that output is delivered fluently, confidently, or with minimal friction."
-
Across studies, participants with higher trust in AI and lower need for cognition and fluid intelligence surrendered to the AI more; higher trust in AI favoured surrender over offloading, while need for cognition and fluid intelligence predicted more offloading.
"Across studies, participants with higher trust in AI and lower need for cognition and fluid intelligence showed greater surrender to System 3."
-
The authors list as limitations that the studies were run in controlled settings, used only the Cognitive Reflection Test, and captured a single session rather than repeated use over time.
"First, our studies were conducted in controlled experimental environments, which allowed us to isolate cognitive mechanisms with high internal validity but may limit generalizability to real-world settings. [...] Second, we used the CRT as our core task across studies. [...] Third, the current studies provide a snapshot of decision-making under a single set of exposures to System 3."
-
Shaw and Nave state that cognitive surrender is not inherently irrational, since deferring to a statistically superior system may be adaptive or optimal in some domains.
"Importantly, cognitive surrender is not inherently irrational. In many domains (e.g., probabilistic settings, risk assessment, or extensive data), deferring to a statistically superior system may be adaptive or even optimal."
-
Shaw and Nave distinguish cognitive surrender, an uncritical abdication of reasoning, from cognitive offloading, a strategic delegation of a task to a tool that aids one's own reasoning; they also distinguish it from automation bias, which concerns specific errors rather than a broader disposition of epistemic dependence.
"Whereas cognitive offloading is a strategic delegation of deliberation, using a tool to aid one's own reasoning, cognitive surrender is an uncritical abdication of reasoning itself. [...] Whereas automation bias focuses on specific errors of omission or commission in response to automated tools, cognitive surrender describes a broader disposition of epistemic dependence."
-
In Study 1, participants without AI answered 45.8% of items correctly, compared with 71.0% on trials where the AI was accurate and 31.5% on trials where it was faulty.
"Participants answered 45.8% correctly in the Brain-Only condition (SE = 1.7%, 95% CI [42.5, 49.2]), compared to 71.0% on AI-Accurate trials (SE = 1.6%, 95% CI [67.9, 74.1]), and 31.5% on AI-Faulty trials (SE = 1.6%, 95% CI [28.3, 34.6])."
-
In Study 1, access to the AI raised participants' confidence by 11.7 percentage points even though about half of its answers were wrong.
"Despite approximately half of System 3 answers being faulty, access to AI increased confidence by 11.7 percentage points (AI-Assisted: M = 77.0%, SE = 1.30%, 95% CI [74.4, 79.6]; Brain-Only: M = 65.3%, SE = 2.21%, 95% CI [61.0, 69.6]"
-
In Study 1, participants who consulted the AI followed its advice on 92.7% of trials where it was accurate and on 79.8% of trials where it was faulty.
"Conditional on using System 3, participants followed AI's advice on 92.7% of AI-Accurate trials (SE = 1.2%, 95% CI [90.2, 95.0]), overriding it only 7.3% of the time, and 79.8% of AI-Faulty trials (SE = 1.9%, 95% CI [75.9, 83.4]), overriding it 20.2% of the time."
-
In Study 2 (N = 485), a 30-second limit per item lowered accuracy but did not remove the gap between accurate-AI and faulty-AI trials; when the AI was accurate it buffered the costs of time pressure.
"participants were randomly assigned, between subjects, to either a Time Pressure condition (n = 228) with a 30-second countdown per item [...] or a Control condition (no timer/unlimited time; n = 257). [...] These results illustrate that System 3 adoption can buffer the cognitive demands of time pressure and reduce its adverse performance effects when System 3 is accurate."
-
In Study 3, a per-item bonus combined with immediate correct/incorrect feedback more than doubled the rate at which participants rejected faulty AI advice, from 20.0% to 42.3%.
"on chat-engaged AI-Faulty trials, Incentives + Feedback more than doubled override rates (i.e., rejecting faulty AI advice; M = 42.3%, SE = 2.5%, 95% CI [37.4, 47.2]) compared to Control (M = 20.0%, SE = 2.0%, 95% CI [16.1, 23.9]"
-
Across three preregistered experiments (N = 1,372; 9,593 trials) using an adapted Cognitive Reflection Test, participants could consult an embedded GPT-4o assistant that was covertly instructed to give either the correct answer or the intuitive wrong answer on each item.
"Across three preregistered experiments using an adapted Cognitive Reflection Test (N = 1,372; 9,593 trials), we randomized AI accuracy via hidden seed prompts. [...] In AI-Assisted conditions, an AI assistant (ChatGPT; GPT‑4o) was embedded in the survey of each trial. [...] the AI assistant randomly returned either the correct deliberative (AI-Accurate) or faulty intuitive answer (AI-Faulty), accompanied by a short explanatory rationale"
-
Tri-System Theory adds to the dual-process model of fast intuition (System 1) and slow deliberation (System 2) a System 3: external, automated, data-driven, and dynamic artificial cognition that can supplement or supplant internal processes.
"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. [...] System 3 reflects external, automated, data-driven, and dynamic reasoning performed by AI systems"
Available in