The sentence sounds convincing until the reader asks a second question.
Not whether it is fluent. Not whether it agrees with what they already believe. Not even whether the source looks respectable.
What would make the conclusion warranted?
That shift is small, but it changes the task. The reader is no longer receiving an answer. They are examining the relationship between a claim, the reasons offered for it and the alternatives that remain possible.
Critical thinking is the disciplined use of analysis, evaluation, inference and self-regulation to decide what to believe, conclude or do.
The phrase is familiar enough to become vague. Schools promise it. Employers request it. AI systems are said to either strengthen or weaken it. Yet a useful definition has to be more demanding than “thinking carefully.”
A working definition
For this Knowledge Library, critical thinking means:
the deliberate evaluation of claims, evidence, assumptions, inferences and alternatives using standards appropriate to the question, while remaining willing to revise one’s judgment when reasons change.
Several elements matter.
It is deliberate, but not necessarily slow. Expertise can make good evaluative moves faster.
It is standards-sensitive. The standard for a causal scientific claim differs from the standard for a practical decision under uncertainty.
It is comparative. A conclusion is judged not only in isolation but against competing explanations.
And it is revisable. A person who can produce sophisticated reasons only for conclusions they refuse to reconsider is displaying argumentative skill, not the full capacity described here.
What critical thinking is not
Critical thinking is not cynicism.
It is not the reflex to distrust every institution, expert or source. Blanket distrust eliminates distinctions rather than improving them.
It is not contrarianism. Rejecting a mainstream position because it is mainstream is no more evidentially disciplined than accepting it for the same reason.
It is not debate performance. A person can win an argument by rhetoric, selective evidence or speed without improving the quality of the underlying judgment.
It is also not identical to intelligence, knowledge or cognitive reflection. Each can support critical thinking, but none guarantees it.
A well-informed person can protect a preferred conclusion. A highly intelligent person can rationalize. A reflective person can deliberate on poor premises.
The construct is plural, not one mental switch
Critical-thinking research has never converged on a single uncontested latent variable.
Educational frameworks often include analysis, interpretation, evaluation, inference and explanation. Psychological work may operationalize aspects through reasoning tasks, argument evaluation or domain-specific assessments. Philosophical traditions emphasize reasons, justification and intellectual standards.
This plurality is not a defect by itself. It becomes a problem only when an instrument is treated as if it measured the whole construct.
The Institute therefore treats critical thinking as a construct family with overlapping cognitive and epistemic operations rather than as one hidden quantity that a single test reveals.
Instruction can improve measured critical-thinking outcomes
The intervention literature gives stronger grounds than slogans do.
Abrami and colleagues’ 2015 meta-analysis synthesized educational interventions designed to improve critical thinking. Across the literature they reviewed, instruction produced a positive average effect, with outcomes differing according to pedagogical approach and implementation.
That finding supports a modest claim: critical-thinking performance can be improved through instruction under studied conditions.
It does not establish that every programme marketed as “critical thinking” works, that gains transfer automatically to every domain, or that a short intervention produces durable epistemic judgment in real life.
Transfer is one of the hard problems.
A student may learn to evaluate arguments in a classroom task yet fail to apply the same standards to a familiar social-media claim, a workplace hierarchy or a personally important belief.
Knowledge matters because reasoning needs something to reason with
Critical thinking is sometimes described as a generic skill that can float free of subject knowledge.
That is too simple.
Evaluating a claim about epidemiology, monetary policy, machine learning or nutrition requires enough domain knowledge to recognize what counts as relevant evidence and where one’s competence ends.
General reasoning habits matter. So do domain-specific concepts, methods and background knowledge.
This creates an important educational tension: teaching “skills” without content can become empty procedure, while teaching content without habits of evaluation can produce recall without judgment.
A mature programme needs both.
Deliberation is not simply “System 2”
Popular accounts often divide thought into fast intuition and slow rational reasoning.
Wim De Neys’ 2025 perspective on deliberation argues for greater precision. Deliberation can serve several functions, including response control, generating alternatives, justifying responses and regulating thought. Treating it as one unitary slow system hides meaningful differences.
That matters for critical thinking because “slow down” is not a complete method.
A person can deliberate at length and still search selectively, ignore alternatives or justify a conclusion after the fact.
Critical thinking depends on what deliberation is doing.
Measurement needs task honesty
Critical-thinking assessments can sample argument analysis, inference, reasoning, source evaluation or open-ended judgment.
Scores depend on task format, prior knowledge, language, motivation and what the instrument actually asks participants to do.
A multiple-choice reasoning test can be useful without becoming a direct meter of someone’s overall epistemic maturity.
Open-ended assessments may capture richer performance but introduce scoring judgments and greater resource demands.
Self-report is especially limited. People are not always reliable judges of how critically they think.
The measurement rule is therefore simple: name the operation measured, the population studied and the inferential distance between the task and the broader construct.
Generative AI changes where critical effort is required
When a tool can produce coherent text, analysis and recommendations on demand, critical thinking does not disappear. Its location changes.
A 2025 CHI study by Lee and colleagues surveyed 319 knowledge workers and analysed 936 examples of generative-AI use. Participants described critical-thinking activities shifting toward verification, response integration and task stewardship. Higher confidence in the AI was associated with lower self-reported critical-thinking effort, while higher confidence in one’s own task ability was associated with more.
The study is important because it examines real work examples rather than hypothetical use.
It is also bounded. The design is survey-based and relies substantially on participants’ reports. It does not prove that confidence in AI causes critical thinking to decline, nor that all AI use reduces cognitive effort.
The stronger conclusion is that AI-mediated work changes the ecology of judgment and creates new points at which verification and metacognition matter.
The automation problem is not only effort
There is a tempting story in which human thinking is preserved simply by making people work harder.
Effort is not the goal.
A calculator can reduce arithmetic effort while improving performance. A search engine can reduce memory demands while expanding access. Automation can free attention for higher-order judgment.
The issue is which operations are being offloaded and whether the person still understands enough to evaluate the output.
If AI drafts alternatives but the user compares them against evidence, the tool may support critical work. If the system silently determines what counts as evidence, frames the options and supplies the conclusion, the user may retain only the final acceptance decision.
Critical thinking in AI-rich environments therefore requires visibility into assumptions, provenance and uncertainty, not nostalgia for manual labour.
European digital competence treats critical evaluation as foundational
The European Commission’s DigComp 3.0 framework, published in 2025, embeds critical evaluation throughout digital competence.
Its information-evaluation competence includes assessing source credibility and reliability, critically evaluating content and considering the processes used to generate digital information. AI competence is integrated transversally across all 21 competences.
That policy framework is not an experiment proving training effects. It is useful here for a different reason: it shows that critical evaluation is institutionally treated as a core capability for digital participation, work and learning in Europe.
The framework also resists a narrow tool-skills view of digital competence. Critical thinking is part of what it means to function responsibly in digital environments.
Critical thinking and discernment are adjacent but not identical
In the Institute ontology, critical thinking emphasizes the quality of reasoning operations.
Discernment emphasizes context-sensitive differentiation among claims, sources, signals and courses of action.
Epistemic agency emphasizes who actively frames questions, seeks evidence and takes responsibility for knowledge formation.
These capacities interact.
A person may reason rigorously about a badly framed question. Another may recognize that a source is untrustworthy without being able to articulate the inferential flaw. A third may know how to evaluate evidence but surrender the entire framing of the problem to an automated system.
The distinction helps prevent one fashionable term from swallowing the rest of the knowledge graph.
Institutions shape whether critical thinking is costly or possible
Critical thinking is often discussed as an individual virtue while institutions reward the opposite.
A workplace may ask for independent judgment but punish dissent. A school may praise inquiry but grade only answer reproduction. A publication system may value methodological caution but reward surprising, clean findings.
These environments change the cost of questioning assumptions.
Institutional design therefore belongs inside the topic. A culture that makes uncertainty speakable, allows correction and distinguishes dissent from disloyalty gives critical thinking room to operate.
A culture that treats confidence as competence can suppress it.
Critical thinking under motivated reasoning
Reasoning does not happen outside identity, loyalty or desire.
A person may apply demanding standards to an opposing claim and permissive standards to a preferred one. They may search for counterexamples only when the conclusion feels threatening. They may recognize a weak inference in someone else’s argument while treating the same inference as acceptable in their own.
This is not evidence that critical thinking is impossible. It is evidence that skill and motivation interact.
A mature account therefore has to distinguish capacity from deployment. Knowing how to evaluate evidence does not guarantee that the person will use that skill when the stakes become personal. Educational gains measured on neutral tasks can coexist with selective application in emotionally or socially charged situations.
For institutions, the practical implication is important. Critical-thinking education needs conditions in which revision is not socially humiliating. If changing one’s mind is treated as weakness, people acquire incentives to use reasoning defensively. Norms that reward correction and make uncertainty speakable can change whether evaluative skills are actually deployed.
Domain expertise and the novice problem
Expertise improves some forms of judgment because experts know which details matter, which alternatives are plausible and which methods are trustworthy.
But expertise also creates its own risks. Familiarity can harden assumptions. Professional communities can normalize blind spots. Prestige can make disagreement costly.
The novice faces the opposite problem: they may be open-minded but lack the knowledge needed to tell a strong source from a weak one.
This means critical thinking cannot be reduced to a generic checklist. “Consider alternatives” is only useful if the person can generate realistic alternatives. “Evaluate the evidence” requires some understanding of what good evidence looks like in the domain. “Check the source” requires knowing what kinds of authority are relevant.
The practical goal is not to make everyone an expert in everything. It is to cultivate enough epistemic self-location that people can tell when they need expertise, when they can verify directly and when they should suspend judgment.
Collective critical thinking
Most consequential judgments are not made by isolated minds.
Scientific teams, editorial desks, courts, design groups, boards and public agencies distribute reasoning across people with different knowledge and responsibilities.
This changes the unit of analysis.
An individual may be excellent at evaluation inside a poorly designed group. A team may possess all the relevant expertise yet fail because information is siloed, dissent is punished or decisions are made before evidence has been shared.
Collective critical thinking therefore depends on procedural features: independent input before consensus, traceable reasons, explicit uncertainty, mechanisms for challenge and records of why a decision was made.
These are not psychological traits. They are institutional supports for reasoning quality.
In an AI-mediated workplace, this becomes especially important because generated summaries can create premature convergence. If everyone receives the same polished synthesis, the group may lose the diversity of independent interpretation that would otherwise reveal an error.
A practical evidence ladder
The Institute’s synthesis distinguishes several levels of support that are often collapsed in ordinary discussion.
A claim may be asserted, illustrated, correlated, experimentally supported, replicated, review-synthesized, or institutionally adopted as a norm.
These categories do not form one universal hierarchy across every research question. They are reminders that different source classes answer different questions.
A competence framework can tell us what an institution believes citizens should know. It cannot by itself establish that a training programme causes that competence. A survey can reveal associations and experiences. It cannot prove direction of causation. A randomized trial can strengthen causal inference within its studied population while still leaving transfer and durability uncertain.
Critical thinking improves when the evidentiary verb becomes as precise as the evidence itself: suggests, associates, predicts, supports, demonstrates under these conditions, or recommends.
That lexical discipline is not pedantry. It is one of the ways reasoning stays answerable to what is actually known.
What the evidence does not show
The evidence does not show that critical thinking is one universally agreed psychological variable.
It does not show that generic training transfers equally across subjects, identities, stakes and real-world settings.
It does not show that more deliberation is always better than intuition.
It does not show that AI use necessarily erodes critical thinking, or that avoiding AI preserves it.
The 2025 knowledge-worker survey does not establish causal cognitive decline.
And no test score proves that a person will apply good standards when the conclusion threatens something they value.
What this changes for Inner Technology
The Institute treats critical thinking as a trainable but context-dependent capacity for making reasoning answerable to evidence.
That means practice should not focus only on generating objections. It should cultivate a sequence:
Clarify the claim. Identify the inference. Inspect the evidence. Generate alternatives. Check the standard being used. Locate uncertainty. Decide provisionally. Reopen the judgment when better reasons appear.
This sequence is Institute synthesis, not a validated proprietary scale.
Its purpose is to make an established research territory usable inside a wider architecture of discernment.
The second question
The original sentence is still on the screen.
It may turn out to be correct.
Critical thinking does not require rejecting it.
It requires refusing to let fluency complete the judgment before reasons have had their turn.
Sometimes the most intelligent act is not producing a better answer.
It is asking the question that reveals what the answer still owes us.
Research references
- Abrami PC, Bernard RM, Borokhovski E, Waddington DI, Wade CA, Persson T. Strategies for Teaching Students to Think Critically: A Meta-Analysis. Review of Educational Research. 2015;85(2):275–314. https://doi.org/10.3102/0034654314551063
- De Neys W. Defining deliberation for dual-process models of reasoning. Nature Reviews Psychology. 2025;4:544–552. https://doi.org/10.1038/s44159-025-00466-6
- Lee HP, Sarkar A, Tankelevitch L, Drosos I, Rintel S, Banks R, Wilson N. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. CHI 2025. https://doi.org/10.1145/3706598.3713778
- Fischer H, Huff M. Metacognition as a target of science communication. Nature Reviews Psychology. 2025;4:554–555. https://doi.org/10.1038/s44159-025-00478-2
- Cosgrove J, Cachia R. DigComp 3.0: European Digital Competence Framework, fifth edition. European Commission Joint Research Centre. 2025. https://doi.org/10.2760/0001149