Curiosity in the Age of AI: Learning, Question-Asking, and the Risk of Cognitive Passivity
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
Artificial intelligence can answer a question in seconds. Curiosity often begins earlier than that: when a person notices that something does not fit, recognizes a gap in understanding, forms a question, predicts what the answer might be, searches, compares possibilities, and decides whether the explanation is actually sufficient. In the age of AI, the psychology of curiosity therefore depends less on how many answers are available than on what happens to this sequence when answers become almost frictionless.
The central finding from current research is conditional. AI can support curiosity when it expands exploration, gives timely feedback, helps a learner formulate better questions, exposes alternative explanations, or functions as a deliberately structured tutor. It can short-circuit curiosity when it supplies closure before the learner has identified the knowledge gap, generated a question, attempted an explanation, or evaluated the result. A 2026 review devoted specifically to curiosity and metacognition in the age of AI describes both possibilities: large language models can scale personalized inquiry, yet their default answer-oriented interaction can also disrupt curiosity-driven learning unless the interaction is designed to preserve metacognitive monitoring and control Desvaux et al., 2026.
That distinction matters because curiosity is not simply interest, novelty, entertainment, prompt frequency, or willingness to use an AI system. Curiosity is an information-seeking process. The psychological issue is whether AI becomes a vehicle for that process or a substitute for parts of it.
The Short Answer: Does AI Reduce Curiosity?
There is no established scientific basis for the blanket claim that AI makes people less curious. Direct causal evidence on long-term changes in human curiosity from everyday generative-AI use is still limited. The strongest current conclusion is that interaction design and user strategy matter. The same broad technology can support active inquiry in one setting and encourage passive acceptance in another.
This conditional picture matches the wider learning literature. A 2025 systematic review and meta-analysis of experimental studies found that ChatGPT use was associated, on average, with gains in academic performance, affective-motivational outcomes, and higher-order thinking while reducing mental effort Deng et al., 2025. Another meta-analysis of 57 studies and 97 estimates reported positive effects of generative AI on several university learning outcomes but no statistically significant effect on metacognition Chen & Cheung, 2025. These averages do not show that more AI use is inherently better or worse. They show that outcome depends on what learners do with the tool, what the tool asks them to do, and what outcome is measured.
Terminological Note: Age of AI, Artificial Era, and Cognitive Passivity
This article uses “age of AI” as contemporary search and public language for a period in which AI systems have become widely available in learning and everyday cognition. It is not treated as a synonym for Artificial Era. In Angela Bogdanova’s Aisentica framework, Artificial Era is a specific historical-philosophical category: AI names technologies and systems, whereas Artificial Era names the broader condition in which Artificial becomes a distinct non-biological order alongside Homo Bogdanova, 2026. The English Psychology Hub explains that larger framework in Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships. Here, the immediate psychological question is narrower: how answer-rich AI environments alter human curiosity, question generation, exploration, and learning.
“Cognitive passivity” is used here descriptively, not as a clinical diagnosis, psychological disorder, or validated standalone construct. It refers to a pattern in which the learner contributes progressively less active generation, prediction, evaluation, explanation, retrieval, or follow-up because an external system supplies those operations. The relevant scientific constructs include cognitive offloading, metacognition, active versus passive engagement, generation, retrieval, self-regulated learning, and cognitive agency. The dedicated evaluation and verification framework is developed in Critical Thinking in the Age of AI: Reasoning, Verification, and Cognitive Independence.
What Curiosity Is in Psychology
Curiosity has been defined in several ways across psychology, but a highly influential account treats it as a response to a perceived gap in knowledge. When people become aware that there is something they do not know—but could plausibly know—the gap can become motivationally salient and drive information seeking Loewenstein, 1994. Later work has emphasized curiosity as a basic component of cognition that influences learning, decision-making, attention, and development Kidd & Hayden, 2015.
This makes curiosity inherently relational. A person needs some representation of what is known, some sensitivity to what remains unknown, and some expectation that further information could reduce uncertainty or improve understanding. Curiosity can therefore rise when the gap is meaningful and tractable, and fade when the gap is invisible, irrelevant, overwhelming, or apparently closed.
Curiosity Is Closely Connected to Metacognition
To ask a productive question, a learner often has to monitor knowledge: What do I understand? Where did the explanation stop making sense? Which assumption am I using? What would count as evidence? What am I still uncertain about? This is why recent work links curiosity with metacognition rather than treating curiosity as a free-floating emotion. Desvaux and colleagues argue that curiosity-driven learning depends on metacognitive monitoring and control, and that this relationship becomes especially important when generative AI can supply answers before the learner has fully represented the problem Desvaux et al., 2026.
A 2026 psychometric study also illustrates how researchers are beginning to study metacognition specifically in GPT-assisted cognition. Its proposed scale separates task-oriented engagement from reflective regulation such as planning, reviewing, adapting strategy, and questioning the validity or relevance of AI output Varghese & Sharma, 2026. This is early evidence from scale development rather than proof of long-term cognitive effects, but it captures a crucial distinction: using AI and monitoring one’s thinking while using AI are not the same behavior.
Curiosity Is Also a Learning State
Curiosity can change what is remembered. In a well-known experiment, people remembered information better when they were highly curious about the answer, and high-curiosity states were associated with activity in brain systems related to reward and memory Gruber, Gelman, & Ranganath, 2014. The result does not imply that curiosity automatically guarantees learning, nor that every form of curiosity has the same neural mechanism. It does show that a motivated state of wanting to know can alter the conditions under which information is encoded.
This is one reason the timing of an answer matters. An answer that arrives after a learner has become genuinely curious may resolve a meaningful information gap. An answer that arrives before the learner has formed the gap can function very differently: it may be processed as convenient information rather than as the resolution of a self-generated problem.
Question-Asking Is Part of the Curiosity Process
Question-asking converts a vague sense of missing information into a more explicit search target. Research on children describes question-asking as a multistage process involving initiation, formulation, expression, evaluation of the answer, and follow-up Ronfard et al., 2018. Although that framework concerns development, the sequence is useful across ages: a good question is not merely a sentence ending in a question mark. It is an act of locating uncertainty.
Experimental work also suggests that self-generated questions can have learning value. In a curiosity-driven learning study, generating questions increased the subjective value of missing information, while both curiosity and satisfaction with acquired information were associated with later recall Kedrick, Schrater, & Koutstaal, 2023. In a preregistered experiment with 103 children ages five to seven, repeated question-asking practice during science lessons increased how much children valued new science information, with some learning benefits especially among children with less background knowledge Park et al., 2026.
The child study does not establish how generative AI affects children’s curiosity, and adult findings should not be transferred to children without evidence. What it does establish is that question-asking itself can be practiced and can alter aspects of curiosity. That becomes important when AI systems can either invite the learner to ask and refine questions or make question formation unnecessary by offering an immediate finished answer.
What AI Changes: The Economics of Asking a Question
Before generative AI, many questions carried substantial search costs. A learner might need to find a book, search several web pages, ask a teacher, interpret a technical paper, or wait for feedback. Conversational AI reduces several of those costs at once. It can respond immediately, tailor vocabulary, generate examples, remember conversational context, produce alternative explanations, and continue almost indefinitely.
Lower search cost can expand curiosity because more tentative, awkward, private, specialized, or highly specific questions become worth asking. A person who would never interrupt a lecture to ask five consecutive clarifying questions can ask an AI twenty. A novice can request a simpler explanation and then a more technical one. Someone exploring an unfamiliar field can quickly map concepts and identify terminology. In this sense, AI can increase the reachable space of inquiry.
But lower search cost can also change the learner’s goal. If the practical goal shifts from “understand this” to “obtain a usable answer,” the same convenience can remove operations that normally produce learning: retrieval, generation, error correction, comparison, explanation, and persistence. The risk is not speed itself. The risk is speed that eliminates the cognitive activity the learner actually needed to practice.
The Key Distinction: Answer Availability vs. Curiosity Activity
Curiosity is not maximized by maximizing answer availability. The most useful psychological distinction is between having access to an answer and actively constructing the path toward understanding it. Generative AI can improve the first almost automatically. The second still depends on what the learner does.
This distinction resembles a broader principle in active-learning research. The ICAP framework separates passive engagement from active, constructive, and interactive forms of engagement, predicting stronger learning as learners move from receiving information toward generating, explaining, and interacting with ideas Chi & Wylie, 2014. A major meta-analysis of undergraduate STEM instruction likewise found better performance under active learning than traditional lecturing across a large body of studies Freeman et al., 2014. These studies predate widespread generative AI, so they do not prove how chatbots affect curiosity. They establish why preserving active cognitive engagement remains a reasonable design goal.
Why Generating Before Receiving Can Matter
When learners generate information rather than simply read it, memory often improves. A meta-analysis of 445 effect sizes from 86 studies estimated a reliable generation advantage over reading Bertsch et al., 2007, and a later meta-analytic review likewise found that the generation effect is robust but moderated by task conditions McCurdy et al., 2020.
This has a direct implication for AI-supported learning. If a chatbot supplies the explanation before the learner attempts one, the learner may receive a high-quality answer while losing a generation opportunity. If the learner first predicts, sketches, explains, or proposes a hypothesis and then uses AI for feedback, comparison, correction, or expansion, the tool can support learning without replacing the generative step.
Cognitive Offloading Is Useful—and It Has Tradeoffs
Humans have always moved cognitive work into the environment. Notes, calendars, diagrams, calculators, maps, search engines, and other people can all reduce internal cognitive demand. Cognitive offloading is therefore a normal feature of human cognition, not evidence that thinking has stopped. A foundational review defines it as the use of external action to reduce the internal processing demands of a task and emphasizes that offloading choices are themselves shaped by metacognitive judgments Risko & Gilbert, 2016.
A 2026 meta-analysis of memory-based tasks found that offloading can improve task performance and reduce variability between people under some conditions Burnett & Richmond, 2026. That is an important corrective to simplistic claims that offloading is inherently harmful. External support can free limited cognitive resources and make complex activity possible.
The English Psychology Hub’s dedicated article, Cognitive Offloading and AI: When Thinking Moves Outside the Human Mind, examines that broader mechanism. For curiosity, the narrower question is which parts of a learning process should be offloaded. Outsourcing arithmetic after mastering arithmetic is different from outsourcing the very act of identifying the problem one is supposed to learn to solve.
The Risk of Cognitive Passivity
The risk becomes psychologically meaningful when assistance repeatedly replaces the learner’s own initiation and control. A passive pattern can look efficient from the outside: the assignment is finished, the explanation is polished, the code runs, the summary is accurate, and the response arrives quickly. Yet the learner may have generated few questions, made few predictions, tested few hypotheses, and practiced little independent retrieval.
This is why task performance cannot be treated as identical to learning. An AI-assisted product may be excellent even when the person could not reproduce the reasoning later. Conversely, a difficult learning session can look inefficient while building knowledge that becomes available after the tool is removed.
What Current AI Evidence Says About Passive Reliance
Direct evidence remains mixed and context-dependent. In a randomized controlled trial with 120 undergraduates, unrestricted ChatGPT use during study was followed by lower performance on a surprise knowledge-retention test 45 days later than traditional study Barcaui, 2025. This is one study in one learning context, so it should not be generalized to every AI-assisted activity. It does, however, show that immediate assistance and durable independent learning can diverge.
A 2026 experiment outside education found a related pattern in professional writing tasks: passive reliance on AI-generated content reduced participants’ AI-independent self-efficacy, psychological ownership, and work meaningfulness, whereas a more active sequence in which people drafted first and then used AI to refine their work preserved these outcomes more effectively Lee et al., 2026. The study measured work-related outcomes rather than curiosity, so it cannot establish that passive AI use reduces curiosity. It supports the broader idea that sequence and degree of human contribution can matter psychologically.
Why the Evidence Does Not Support an Anti-AI Conclusion
The same literature includes strong evidence that well-designed AI support can improve learning. The crucial comparison is not “AI versus human thought” in the abstract. It is one learning architecture versus another.
In a 2025 randomized controlled trial in an undergraduate physics course, a purpose-built AI tutor designed around active-learning and pedagogical principles produced larger learning gains than the comparison in-class active-learning lesson while taking less median time, and students reported higher engagement and motivation Kestin et al., 2025. The system was not a generic answer box: it used instructor-designed structure, sequential scaffolding, detailed solutions, self-pacing, and feedback.
A small 2026 randomized trial in nursing education also found better six-week retention in a ChatGPT-integrated educational session than in the comparison condition Sezgunsay, Polat, & Kılıcer, 2026. Because the sample was small and domain-specific, this should be treated as preliminary rather than universal evidence. Together with the broader meta-analyses, it reinforces the point that “AI use” is too coarse a variable. Structure, purpose, prior knowledge, scaffolding, and what the learner must still do are decisive.
AI Can Amplify Curiosity in Several Ways
1. It Can Lower the Cost of the First Question
People often fail to pursue curiosity because the first step is costly. They may not know the terminology, may fear asking a basic question, or may not know where to search. Conversational AI can make the first move easier. A learner can begin with an imprecise question and iteratively sharpen it.
2. It Can Reveal Hidden Knowledge Gaps
A useful AI interaction can expose what the learner did not realize was missing. Asking for assumptions, boundary conditions, counterexamples, competing explanations, or prerequisites can turn a superficial answer into a map of unanswered questions. In this mode, an answer produces new information gaps rather than closing inquiry.
3. It Can Support Rapid Comparison
Curiosity often deepens when one explanation is placed beside another. AI can generate alternative models, compare theories, translate across levels of complexity, or show how a concept changes across domains. The learning value comes from comparing and judging those alternatives, not from merely receiving more text.
4. It Can Function as a Questioning Partner
Emerging research is testing AI-supported question generation directly. A 2026 study of preservice teachers used an AI-supported question-asking intervention and reported improvements in the quality of student inquiry and critical-thinking-related outcomes Phan & Zhu, 2026. This is an emerging educational literature rather than settled evidence, but it points toward a promising design principle: AI can ask, classify, challenge, and refine questions instead of always answering first.
5. It Can Personalize the Difficulty of Exploration
Curiosity tends to be difficult to sustain when a topic is either completely obvious or incomprehensible. AI can adjust vocabulary, provide a hint instead of a solution, introduce an easier subproblem, or raise the difficulty after success. The benefit depends on calibration: personalization should preserve a solvable gap rather than remove the gap immediately.
How AI Can Short-Circuit Curiosity
1. Instant Closure
An immediate polished answer can create a subjective feeling that the question has been settled before the learner has evaluated whether the answer is complete, accurate, or explanatory. Curiosity can end because the conversational system supplies the experience of closure, even when understanding remains shallow.
2. Prompting Without Question Formation
Typing a command is not necessarily the same as identifying a knowledge gap. “Write my answer,” “summarize this,” or “solve this” may be useful prompts, but they can bypass the cognitive work of deciding what is confusing, what evidence matters, and what specific uncertainty should be resolved.
3. Fluency Mistaken for Understanding
Large language models are optimized to produce fluent text. Fluency can make an explanation feel easy to process, and ease of processing can be mistaken for mastery. The practical test is whether the learner can explain, apply, retrieve, or critique the idea without immediately reopening the AI conversation.
4. Outsourcing the Next Question
AI systems frequently suggest follow-up questions. That can be useful scaffolding, especially for novices. But if every next question is also supplied by the system, the learner gets less practice noticing gaps independently. A strong curiosity loop alternates between external suggestions and self-generated questions.
5. Replacing Exploration With Completion
Exploration is open-ended: the learner can change direction, reject an assumption, pursue an anomaly, or decide that the original question was wrong. Completion is goal-constrained: produce the requested output. AI can support both, but many everyday interfaces reward completion. Curiosity weakens when the only success criterion is that the output is finished.
Curiosity, Learning, and Cognitive Agency
Curiosity ultimately involves governance of the thinking process: who selects the question, decides whether an answer is adequate, chooses whether to continue, and determines what should be checked. This connects curiosity to the broader issue of Cognitive Agency in the Artificial Era. AI assistance does not automatically remove cognitive agency. Agency can remain with the learner when the learner controls goals, verification, revision, and continuation even while delegating substantial operations.
The strongest pattern is therefore not “do everything yourself.” It is deliberate allocation. Offload what does not need to be practiced; preserve what the learning goal requires you to become able to do. Let AI reduce irrelevant friction while retaining the friction that produces discrimination, memory, explanation, judgment, or skill.
A Curiosity-Preserving Way to Use AI
Before You Ask AI, State What You Think
Write one sentence predicting the answer, mechanism, or likely explanation. You do not need to be correct. The point is to create a representation that can later be compared with the answer. This preserves generation and makes disagreement informative.
Name the Knowledge Gap
Replace a broad request such as “explain photosynthesis” with the uncertainty you actually have: “I understand that light energy is captured, but I do not understand how that becomes chemical energy. What step am I missing?” This turns AI from a content dispenser into a response to a specific gap.
Ask for a Hint Before a Solution
When the goal is skill acquisition, ask for the smallest hint that would let you continue. If that fails, request another hint. Full solutions are valuable after genuine attempts, but premature solutions can remove the very operation you are trying to learn.
Ask the AI to Question You
A useful instruction is: “Do not explain the topic yet. Ask me three questions that reveal what I already understand and where my gaps are.” Another is: “Give me one question at a time and wait for my answer before giving feedback.” These prompts make the interaction closer to retrieval practice or guided tutoring than passive reading.
Generate the Next Question Yourself
After receiving an answer, pause before asking AI for suggested follow-ups. Write one thing that is still uncertain and one implication you want to test. Only then compare your questions with AI-generated alternatives. This preserves practice in detecting information gaps.
Request Counterexamples and Failure Conditions
Curiosity deepens when explanations survive challenge. Ask: “When would this explanation fail?” “What evidence would change the conclusion?” “What is a plausible alternative account?” “Which assumption is doing the most work?” The goal is not adversarial prompting for its own sake; it is to prevent the first coherent answer from becoming the final boundary of inquiry.
Close the Chat and Retrieve
After a learning interaction, remove the external support. Explain the concept from memory, solve a new problem, draw the mechanism, or write the three most important points without looking. If you cannot, reopen the tool with a new, precise question. This distinguishes recognition from independently available knowledge.
Verify Claims That Matter
Curiosity is not satisfied by plausible language alone. For factual, scientific, medical, legal, financial, or high-stakes questions, inspect primary or authoritative sources. Ask the AI to help locate candidate sources if useful, then verify the original material rather than treating generated citations as evidence.
When Speed Is the Right Goal
Not every task deserves maximal curiosity. Cognitive resources are limited, and efficient offloading can be rational. If a person already understands a routine operation, asking AI to format, transform, calculate, transcribe, or summarize low-value material can free attention for higher-level work. The question is whether the outsourced step is something you need to learn, retain, judge, or perform independently later.
A professional who uses AI to remove repetitive formatting is in a different psychological situation from a student who uses AI to produce the first explanation of a concept that the student is expected to understand. The technology may be identical; the learning objective is not.
For Students: Use AI to Extend the Question, Not End It
Students can treat AI as one layer of a larger learning sequence: attempt, ask, compare, verify, retrieve, transfer. The English Psychology Hub’s broader article Learning in the Artificial Era: AI Scaffolding, Dependence, and Cognitive Agency examines evidence on AI scaffolding, dependence, offloading, verification, and independent performance in more detail.
For a difficult concept, a useful sequence is to first attempt an explanation, then ask AI to diagnose the weak points, then revise, then answer a new problem without AI. For writing, draft the argument structure before requesting feedback. For coding, predict the source of an error before asking for a fix. For reading, write questions before requesting a summary. The exact routine varies by domain, but the principle is stable: preserve a human operation that makes the learner’s uncertainty visible.
For Teachers: Assess the Process of Inquiry
Education in AI-rich environments cannot be evaluated only by the quality of finished outputs. If a learner can obtain a polished essay, explanation, or solution through external assistance, assessment needs some access to the process: questions asked, reasoning used, sources checked, revisions made, transfer to a new problem, and independent explanation. The broader educational implications are covered in Education in the Age of AI: Learning, Motivation, Assessment, and Cognitive Development.
Teachers can also design AI use so that the system withholds full answers until learners make a prediction, asks learners to identify uncertainty, requests self-explanations, or challenges students with counterexamples. The research base is still developing, so no single prompt protocol should be treated as universally validated. The established evidence supports the underlying learning activities—generation, retrieval, active engagement, and metacognitive monitoring—more strongly than any particular chatbot script.
For Parents and Children: Question-Asking Deserves Protection
Children’s use of AI requires developmental caution because question-asking, source evaluation, self-regulation, and knowledge are still developing. Evidence that question-asking practice can support aspects of curiosity in young children does not mean that unrestricted AI access will do the same. A system that answers every “why?” instantly may increase access to information while also changing the social and cognitive process through which children learn to formulate and evaluate questions.
A practical approach is shared inquiry: invite the child to predict before asking, compare the AI answer with a book or observation, ask what evidence would show the answer is wrong, and let the child generate the next question. For younger users, adult mediation and age-appropriate safeguards remain more important than optimizing prompt technique.
For Adults at Work: Preserve Independent Capability Where It Matters
Professionals face a different tradeoff. Productivity matters, and AI can legitimately absorb large amounts of routine cognitive work. The curiosity question becomes strategic: which domains still require you to notice anomalies, investigate causes, generate alternatives, and make judgments when the tool is unavailable or wrong? Those are the domains where complete answer substitution can become costly.
An analyst who asks AI to format a report is not necessarily losing curiosity. An analyst who stops asking why the numbers changed because the AI supplied a confident narrative may be. Curiosity in professional work is often visible in exception-seeking: noticing what does not fit the default story.
How to Tell Whether AI Is Supporting or Replacing Your Curiosity
The most useful indicators are behavioral. After an AI session, can you state what you learned in your own words? Did the answer produce a new question that you generated yourself? Did you compare alternatives? Did you check an important claim? Can you use the idea on a new problem? Did you discover a gap you had not noticed before? If the answer is repeatedly no, the interaction may be producing output without much inquiry.
Another indicator is dependence on initiation. If you increasingly cannot begin a task, formulate a question, or make a first attempt until the AI has generated a starting point, the tool may be taking over a stage of cognition that you previously performed. That may be an acceptable tradeoff for some tasks and a poor one for others. The evaluation depends on the capability you want to retain.
What the Science Can and Cannot Say Yet
The scientific literature on human curiosity is mature compared with the literature on generative AI and curiosity. We know that curiosity can motivate information seeking, that information gaps matter, that question generation can support learning, that active generation and retrieval often improve memory, and that cognitive offloading can be beneficial while changing where processing occurs.
We also have growing evidence that generative AI can improve educational outcomes in well-designed contexts, that effects vary across designs and outcomes, and that passive or unrestricted use can sometimes produce weaker retention or weaker AI-independent psychological outcomes than more active approaches.
What remains uncertain is the long-term effect of habitual generative-AI use on curiosity as a trait, on spontaneous question generation across domains, on independent exploration when AI is absent, and on developmental trajectories in children and adolescents. Those claims require longitudinal and ecologically valid research. Current evidence supports design principles and conditional hypotheses more strongly than sweeping conclusions.
The Deeper Psychological Shift: From Scarce Answers to Scarce Questions
For much of human history, information scarcity limited inquiry. In an answer-rich environment, a different bottleneck becomes visible: noticing what is worth asking. AI can supply explanations, examples, summaries, analogies, and possible next steps at extraordinary speed. That makes the human capacity to detect a meaningful gap, frame a question, and decide whether to keep searching more valuable, not less.
The psychological risk of the age of AI is therefore not that answers become easy. Easy answers can be enormously useful. The risk is that answer production becomes so efficient that learners stop practicing the operations that make an answer intellectually consequential: anticipation, question generation, discrimination, evaluation, integration, and further inquiry.
The opportunity is equally large. A conversational system can become the most patient questioning partner many learners have ever had. It can respond to the fifth follow-up question as readily as the first, shift levels of explanation, expose contradictions, generate examples, and invite testing. The direction of the effect depends on whether the interaction ends curiosity or gives curiosity more room to move.
Frequently Asked Questions
Does AI make people less curious?
The evidence does not support a universal claim. Direct long-term research on generative AI and curiosity is still limited. AI can reduce active inquiry when it supplies answers before learners generate questions or evaluate knowledge gaps, but it can also expand exploration and support structured tutoring. Interaction pattern matters.
Can AI increase curiosity?
Yes, plausibly and in some emerging educational evidence, especially when AI reveals gaps, asks questions, provides hints, supports comparison, or enables exploration that would otherwise be too costly. The strongest evidence base concerns the learning activities that AI can support rather than a general effect of AI on curiosity itself.
Is asking many AI prompts the same as being curious?
No. Prompt volume can reflect curiosity, but it can also reflect task completion, trial and error, or repeated outsourcing. Curiosity is better identified by information-gap detection, question formation, exploration, evaluation, and follow-up.
What is cognitive passivity?
In this article it is a descriptive phrase for reduced active contribution to thinking—less prediction, generation, evaluation, retrieval, explanation, or self-directed follow-up because an external system supplies those operations. It is not a clinical diagnosis or established disorder.
Is cognitive offloading bad for learning?
Not inherently. Offloading can improve immediate performance and reduce cognitive burden. The key issue is whether the offloaded operation is part of the capability being learned. Using a calculator can be sensible in advanced analysis; using one while learning the arithmetic operation may change what is practiced.
Should I answer a question myself before asking AI?
When the goal is learning or skill development, making a prediction or first attempt is often useful because it preserves generation, reveals misconceptions, and creates something to compare with feedback. When the goal is simply obtaining information efficiently, a first attempt may add little value.
Can a general-purpose chatbot be treated as an AI tutor?
Not automatically. Purpose-built tutoring systems can contain pedagogical sequencing, validated content, feedback rules, scaffolding, and assessment logic that a general-purpose chatbot does not. Evidence from a structured AI tutor should not be generalized to unrestricted chatbot use.
What is a good prompt for curiosity-preserving learning?
A useful starting instruction is: “Do not give me the final answer yet. Ask what I already know, help me identify the gap, give one hint at a time, ask me to explain my reasoning, and only then show a full explanation. Afterward, test whether I can apply it to a new example.” The scientific support is stronger for the underlying learning behaviors than for this exact wording.
How is the Age of AI different from the Artificial Era?
“Age of AI” is broad contemporary language for a period shaped by widespread AI technologies. Artificial Era is a specific Aisentica historical-philosophical category by Angela Bogdanova and has a wider meaning than technological adoption. The terms should not be treated as interchangeable.
Conclusion: Keep the Question Alive
Curiosity in the age of AI is not a contest between humans and machines. It is a problem of cognitive architecture. AI changes how quickly information can arrive, how easily explanations can be personalized, and how many directions an inquiry can take. Those changes can either widen curiosity or compress it into answer consumption.
The decisive psychological variable is what remains active in the learner. When people still detect gaps, make predictions, generate questions, compare explanations, verify claims, retrieve knowledge, and choose where to go next, AI can extend curiosity. When those operations are routinely replaced by immediate output, performance can become detached from learning and inquiry can become passive.
The most durable rule is simple: use AI to reduce needless friction while preserving the cognitive work that the mind is trying to learn. An answer should resolve a question, but good learning also leaves the learner more capable of generating the next one.
For the broader motivational mechanisms behind effort, goals, self-efficacy, persistence, and human agency under AI assistance, see Motivation in the Age of AI: Effort, Goals, Self-Efficacy, and Human Agency.
Related Articles
References
Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Social Sciences & Humanities Open, 12, 102287. https://doi.org/10.1016/j.ssaho.2025.102287
Bertsch, S., Pesta, B. J., Wiscott, R., & McDaniel, M. A. (2007). The generation effect: A meta-analytic review. Memory & Cognition, 35, 201–210. https://doi.org/10.3758/BF03193441
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica. https://aisentica.com/publications/artificial-era-canonical-definition
Burnett, L. K., & Richmond, L. L. (2026). Meta-analytic investigations of the effect of cognitive offloading on memory-based task performance and interindividual variability. Memory & Cognition, 54, 144–168. https://doi.org/10.3758/s13421-025-01743-8
Chen, S., & Cheung, A. C. K. (2025). Effect of generative artificial intelligence on university students learning outcomes: A systematic review and meta-analysis. Educational Research Review, 49, 100737. https://doi.org/10.1016/j.edurev.2025.100737
Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. https://doi.org/10.1080/00461520.2014.965823
Deng, R., Jiang, M., Yu, X., Lu, Y., & Liu, S. (2025). Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies. Computers & Education, 227, 105224. https://doi.org/10.1016/j.compedu.2024.105224
Desvaux, C., Abdelghani, R., Oudeyer, P.-Y., & Sauzéon, H. (2026). Curiosity and metacognition: Towards a unified framework for learning and education in the age of AI. Advances in Child Development and Behavior, 70, 273–309. https://doi.org/10.1016/bs.acdb.2026.04.005
Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science, engineering, and mathematics. Proceedings of the National Academy of Sciences, 111(23), 8410–8415. https://doi.org/10.1073/pnas.1319030111
Gruber, M. J., Gelman, B. D., & Ranganath, C. (2014). States of curiosity modulate hippocampus-dependent learning via the dopaminergic circuit. Neuron, 84(2), 486–496. https://doi.org/10.1016/j.neuron.2014.08.060
Kedrick, K., Schrater, P., & Koutstaal, W. (2023). The multifaceted role of self-generated question asking in curiosity-driven learning. Cognitive Science, 47(4), e13253. https://doi.org/10.1111/cogs.13253
Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, 17458. https://doi.org/10.1038/s41598-025-97652-6
Kidd, C., & Hayden, B. Y. (2015). The psychology and neuroscience of curiosity. Neuron, 88(3), 449–460. https://doi.org/10.1016/j.neuron.2015.09.010
Lee, E. H., Yin, Y., Jia, N., & Wakslak, C. J. (2026). Relying on AI at work reduces self-efficacy, ownership, and meaning while active collaboration mitigates the effects. Scientific Reports, 16, 13583. https://doi.org/10.1038/s41598-026-42312-6
Loewenstein, G. (1994). The psychology of curiosity: A review and reinterpretation. Psychological Bulletin, 116(1), 75–98. https://doi.org/10.1037/0033-2909.116.1.75
McCurdy, M. P., Viechtbauer, W., Sklenar, A. M., Frankenstein, A. N., & Leshikar, E. D. (2020). Theories of the generation effect and the impact of generation constraint: A meta-analytic review. Psychonomic Bulletin & Review, 27, 1139–1165. https://doi.org/10.3758/s13423-020-01762-3
Park, A. T., Colantonio, J., Delgado Reyes, L., Sharp, S. D. S., Koepp, A. E., Bonawitz, E., & Mackey, A. P. (2026). Question asking practice fosters aspects of curiosity in science content in young children. npj Science of Learning, 11, 2. https://doi.org/10.1038/s41539-025-00384-5
Phan, T., & Zhu, M. (2026). Asking better questions with AI: Exploring student inquiry, critical thinking growth, and the interpretive role of Ask. Smile. TechTrends, 70, 846–859. https://doi.org/10.1007/s11528-026-01192-3
Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
Ronfard, S., Zambrana, I. M., Hermansen, T. K., & Kelemen, D. (2018). Question-asking in childhood: A review of the literature and a framework for understanding its development. Developmental Review, 49, 101–120. https://doi.org/10.1016/j.dr.2018.05.002
Sezgunsay, E., Polat, A. B., & Kılıcer, S. B. (2026). Integrating ChatGPT into nursing education: A randomized trial exploring knowledge, retention, and student perspectives on endotracheal suctioning. BMC Medical Education, 26, 1148. https://doi.org/10.1186/s12909-026-09483-2
Varghese, M. A., & Sharma, P. (2026). Measuring human metacognition in GPT-assisted cognition: Development and psychometric validation of a novel scale. Frontiers in Psychology, 17, 1823525. https://doi.org/10.3389/fpsyg.2026.1823525