For a long time, we associated critical thinking with a familiar image: a person facing a problem, gathering information, and constructing an answer. Artificial intelligence changes that scene. The answer appears before the question has fully matured.
That does not make critical thinking unnecessary. It relocates it.
We must now decide what to ask, recognize what we do not know, evaluate the strength of a response, integrate conflicting perspectives, and assume responsibility for the final decision. Production can accelerate; responsibility cannot.
When confidence in the tool reduces vigilance
A study presented at CHI 2025 surveyed 319 knowledge workers and examined 936 real-world examples of generative AI use. Participants described when they applied critical thinking, why they did so, and how much effort it required.
The researchers found an important association: greater confidence in AI was linked with less self-reported critical thinking, while greater confidence in one’s own ability to perform the task was linked with more critical review. The study relies on self-report and does not establish causation. It does not support claims that AI “atrophies the brain.” It does reveal a design and behavioral pattern worth taking seriously: misplaced confidence can reduce vigilance precisely when we need it most.
The study also found that critical thinking did not disappear; it changed form. Effort shifted toward verifying information, integrating responses, and supervising the task. The ability to create from scratch still mattered, but so did the ability to audit, compare, and decide.
Judgment is not permanent distrust
Critical thinking does not mean rejecting every AI response or turning every conversation into an endless investigation. It means matching the depth of verification to the risk of the decision.
A suggestion for reorganizing a paragraph does not require the same review as a medical claim, a legal interpretation, or an educational recommendation that may affect hundreds of students. Judgment includes knowing when a quick check is enough and when we need to return to the primary source.
Four movements can guide us:
- Name: What am I trying to understand or decide?
- Question: What assumptions, omissions, or ambiguities does the response contain?
- Contrast: What independent evidence supports or contradicts it?
- Respond: What decision can I defend as my own?
This cycle turns doubt into method. It does not discredit technology; it defines an adult relationship with it.
Creativity also needs conditions
PISA 2022 offers another perspective. It was the first international assessment of creative thinking across 64 countries and economies. On average across OECD countries, 78% of students reached at least the baseline level of proficiency, and roughly one in two reached Level 4, demonstrating the ability to generate original and diverse ideas in relatively simple tasks.
But revealing tensions appear. Eight in ten students believe creativity is possible in almost any subject, while only about half believe their creativity can change. Between 60% and 70% reported that their teachers valued creativity, encouraged original answers, or gave them opportunities to express their ideas. Those experiences were associated with somewhat higher scores, even after accounting for student and school characteristics.
Creativity and judgment are not fixed talents that a platform activates automatically. They are sensitive to their environment. They grow when a person can propose, make mistakes, explain, and revise without the correct answer closing the conversation too early.
PISA also recorded an average socioeconomic gap of 9.5 points between advantaged and disadvantaged students. Therefore, any discussion of AI as a capacity-expanding technology must also address access, guidance, and pedagogical quality. An available tool is not the same as a fairly distributed opportunity.
A GRAVYA protocol for reading AI responses
Before incorporating a generative response into an assignment, a class, or a decision, ask:
- Which parts are verifiable facts, and which are interpretations?
- What primary source supports the central claim?
- Which relevant perspective is missing?
- What evidence would change my conclusion?
- Can I explain the idea without repeating the tool’s language?
The final question is decisive. If we cannot reconstruct an idea in our own words, we may have incorporated it into the document without incorporating it into our thinking.
From competent users to responsible authors
AI literacy should not be measured only by the ability to obtain strong outputs. It should also be measured by the ability to interrupt automation, detect a fragile response, acknowledge limits, and sustain an independent judgment.
GRAVYA calls the space between the response received and the decision assumed the third space: a field of conversation where speed is subjected to clarity and information acquires weight.
The question does not delay thought. It begins it.
Question for the community: What habit helps you confirm that an AI response has become an idea you genuinely understand?
References
- Lee, H.-P. et al. (2025). The Impact of Generative AI on Critical Thinking. CHI Conference on Human Factors in Computing Systems.
- OECD (2024). PISA 2022 Results, Volume III: Creative Minds, Creative Schools.
- UNESCO (2023, updated 2026). Guidance for generative AI in education and research.