Raising Good Thinkers in an AI World
By: Shelby Clark
Reflections from preparing for the Children and Screens "Character Matters" Virtual Research Retreat panel on Youth Intellectual Development in a Digital World.
I recently had the privilege of joining a panel at the Children and Screens "Character Matters" Virtual Research Retreat, where we tackled a question I find myself frequently thinking about in my own research: how do we cultivate intellectual character in young people who are growing up alongside generative AI and infinite information?
As I prepared my notes, I realized the through-line of what I wanted to say came back to my, and my colleagues’, research on intellectual risk-taking — the willingness to ask a question when you're unsure, share a tentative idea, admit confusion, or revise your thinking publicly (Soutter & Clark, 2023). I’m excited about the possibilities that AI presents for encouraging students’ intellectual risk-taking, and, at the same time, the many ways that we are already seeing AI encouraging students to turn inwards.
These are the notes, syntheses, and arguments I worked through in preparation for the presentation I was honored to share with Drs. Tenelle Porter, of Rowan University, and Judith Danovitch, of the University of Louisville. I'm share my thoughts here because I suspect other educators, researchers, and parents are wrestling with the same questions.
What is intellectual character, and why should we care?
Intellectual character encompasses the qualities that promote good and productive learning, wondering, reasoning, contemplating, and deliberating. Philosopher Jason Baehr describes intellectual character strengths as those qualities seen in someone we'd describe as a "good thinker" — someone who is curious, intellectually humble (admits when their ideas might be wrong), intellectually courageous (willing to take intellectual risks), thorough (attentive and careful), and open-minded (takes different perspectives into consideration) (Baehr, 2011, 2013).
At its core, it's a commitment to seeking and engaging with truth.
The biggest tension I see right now is that our educational systems still tend to reward speed, correctness, and grades (see Never Enough by Jennifer Breheny Wallace) — while the actual habits of good thinking require slowness, comfort with being wrong, and a tolerance for productive confusion. Modern technology, and AI in particular, is intensifying this tension rather than resolving it.
The decline we should be paying attention to
Curiosity and motivation tend to decline from elementary through high school — a pattern documented across decades of motivation research (Gottfried, Fleming, & Gottfried, 2001; Lepper, Corpus, & Iyengar, 2005). The question this panel considered is: to what extent are digital affordances accelerating that decline?
In our preliminary research on intellectual risk-taking in computer science classrooms, we're seeing more surface-level engagement among students who have AI assistance constantly available. In our qualitative data, students frequently describe "checking" answers on “Chat” (ChatGPT) quickly rather than working through confusion. The act of sitting with not-knowing or grappling with their mistakes— which is where real learning happens — is getting compressed or skipped entirely.
These findings align with broader concerns about how easy access to answers can short-circuit the productive struggle that builds understanding (Kapur, 2008; Bjork & Bjork, 2011).
AI as a thinking partner, not a thinking replacement
It's not all bad news, though.
In one experiment our team conducted comparing how ChatGPT and human collaborators engaged with ethical dilemmas discussions, we found something interesting: ChatGPT gave thorough, structured, broad responses, but humans engaged more deeply in relational and contextual reasoning. AI was efficient whereas humans were more nuanced, particularly around the relational aspects of the dilemma.
This isn't a competition. It's a clue. We can teach students to interrogate AI output, compare it to their own thinking, and use the comparison itself as a learning tool. A colleague, William Cochran, at Wake Forest University, recently framed this distinction as the difference between “substitutive” and “instrumental” uses of AI. In the first, AI helps us to think; in the second, AI does our thinking for us (see also Sharples, 2023, on AI as a Socratic interlocutor rather than an oracle.)
Some of the ways I'd love to see students use AI:
Generate better questions, not just answers
Follow curiosity threads ("wait — why does that happen?")
Ask AI to challenge their thinking or push back on their ideas
Deliberately compare multiple perspectives or viewpoints
Role-play disagreements or explanations
Revise drafts using AI feedback rather than letting AI write the draft
"Think out loud" with AI as a dialogue partner
Ask AI to point out gaps, weaknesses, or missing considerations
Request hints or guidance instead of full answers ("help me figure it out")
Trace how their thinking changes over time ("what did I think before vs. now?")
AI, used this way, can provide the scaffolding for intellectual character development rather than a way to cognitively offload thinking altogether (Risko & Gilbert, 2016).
The peer judgment problem (and why digital spaces can make it worse)
One of the most consistent findings in our intellectual risk-taking research is that perceived peer judgment is one of the strongest inhibitors of curiosity and risk-taking behaviors (Beghetto, 2009; Soutter & Clark, 2023). Students would rather be privately correct than publicly uncertain.
Digital spaces often heighten this dynamic because peers have access to you in more places, more often, with more permanence. The visibility that social media affords — to peers, not just to the platform — turns every classroom contribution into a potential long-lasting reputational event.
The opposite exists as well, however. The online disinhibition effect (Suler, 2004) means some students are more willing to take risks online than in person due to the anonymity provided. This can be helpful for shy or marginalized students, and, at the same time, dangerous when it leads to patterns of cyber-bullying (Kurek, Jose, & Stuart, 2019; Lapidot-Lefler & Barak, 2015). So digital environments aren't monolithically bad for risk-taking. But the dominant pattern, especially in highly visible synchronous formats, is often suppression.
Question-asking is one of the clearest indicators of intellectual risk-taking. And yet, recently, undergraduate students in our research consistently report:
Preferring to look things up rather than ask
Worrying that questions signal a lack of ability (i.e. imposter syndrome)
Withholding tentative ideas until they're "polished enough" to share
The good news: we know that when norms shift — when teachers explicitly value questions and model not-knowing — question frequency increases dramatically. I currently teach exclusively asynchronous classes, and I like to use thinking routines like “Think, Puzzle, Explore” in my courses (see Ritchhart, Church, & Morrison, 2011 for more), to help immediately make question-asking and curiosity a normal part of class each week.
Why peer norms may matter more than individual traits
Here's where I spent a lot of time thinking before the panel.
Bandura's classic social learning theory established that people learn by observing models, especially when those models are salient or rewarded for their behavior (Bandura, 1977). In classrooms, students watch not only teachers but also classmates to infer what kinds of participation, effort, curiosity, and risk-taking are valued.
Peer-network research goes further: “important peers" matter more than “generic peers”. Veenstra and colleagues' review of adolescent peer-network studies emphasizes that influential, central, or socially visible students often act as norm-setters, and peer-led interventions tend to work by training or mobilizing these students (Veenstra & Laninga-Wijnen, 2022).
Strong classroom evidence comes from Paluck, Shepherd, and Aronow (2016), who conducted a large experiment across 56 middle schools. They trained selected students to promote anti-conflict norms, and disciplinary reports of conflict dropped; there were stronger effects when the selected students included more socially influential "referent" students. That is, it wasn't adult messaging that changed student behavior; it was that influential students changed what other students perceived as normal.
Norms work through two channels (Cialdini, Reno, & Kallgren, 1990):
Descriptive norms: what people like us do
Injunctive norms: what people like us approve of
For descriptive norms: it's not enough to just say we value curiosity. Students need to see peers actually asking questions, revising work, disagreeing respectfully, and helping others.
For injunctive norms: this connects to pluralistic ignorance (Prentice & Miller, 1993). Students may privately value effort, kindness, and intellectual risk-taking but assume others don't. Norm-change interventions often work by making desired private attitudes and publicly visible.
The upward comparison trap
Here's the tension I considered, though, because it complicates the "use influential peers as models" recommendation.
The peer norms literature says: “influential peers can shift norms in positive ways.”
But the upward comparison literature says: comparing yourself to higher-status others can undermine well-being and motivation (Festinger, 1954; Wills, 1981; Buunk & Gibbons, 2007).
At first, this seemed like a bit of a paradox to me. But, these aren't in fact contradictory. They describe different psychological pathways.
Upward comparison can inspire us when the person we’re comparing ourselves to feels attainable — "I could become like that" (Lockwood & Kunda, 1997). In contrast, it can harm us when the model feels unreachable — "I'll never be like that."
So what determines whether we feel motivated or not? A few elements stand out:
1.Perceived attainability. Inspiration requires a sense that the model's success is reachable (Han et al., 2017)
2. Identification and similarity. "They are like me" motivates; "they're a different kind of person" discourages (Han et al., 2022).
3. Visibility of process vs. outcome. Seeing only success is harmful; seeing effort, mistakes, and revision is constructive (Dweck, 2006).
These elements are why some of our, and other’s research has found parents to be a top role model for student (Clark et al, 2025): parents are salient, attainable, and similar people that students get to see struggle, fail, and get back up again day in and day out. The take-away for peer role models and norm setting is that peer behavior can’t signal unattainable superiority; instead, peer norm setters should be oriented around accessible, shared practices that other students will be able to identify with.
Designing digital environments for intellectual character
If we wanted to intentionally design digital learning environments to strengthen intellectual character, what would we prioritize? Here's my working list, drawing from the evidence above and our intellectual risk-taking work (Soutter & Clark, 2023):
Belonging first. Small, stable discussion groups instead of whole-class posting into a void. Identity-sharing spaces that humanize participants beyond usernames. Norm-setting built into the platform itself. For example, Matias's large-scale online experiment found that simply displaying community rules in online science discussions reduced harassment concerns and shaped participation (Matias, 2019).
Normalize confusion and mistakes. Prompt students to ask a question before giving an answer. Organize threads around questions rather than responses. Give visible recognition to good questions, not just correct answers. Separate buttons for "Ask a Question" vs. "Share an Answer." Make thinking visible (Miro boards, posted reasoning, draft-stage sharing).
Low-stakes entry points. Allow for private or semi-private thinking before public sharing. Use multiple modalities — voice, text, small group. "Draft mode" before publishing. Quick-response tools (emoji, sliders, short reactions) before deeper engagement.
Inquiry cycles. Engage in inquiry cycles (Question → Explore → Discuss → Reflect → Revise). Slow thinking down instead of optimizing for speed.
Seed norms through credible peers. Have respected students model the target behavior: thoughtful questions, public revision, generous disagreement, crediting others.
Make positive norms more visible than negative ones. Pin examples of strong peer comments, thoughtful questions, "changed my mind because…" moments. Online communities are vulnerable when harmful behavior becomes visible and unchallenged (Kiesler et al., 2012).
Use descriptive norm messages carefully. Not "many students aren't participating," but "students are already using this space to test ideas, ask questions, and support each other."
Pair rules with identity. Online rules work better when they signal "this is who we are here."
Create peer-visible rituals. Weekly "best question," "bravest revision," "kindest challenge," "most generative disagreement," or "I used to think / now I think" posts.
If we designed online spaces for intellectual risk-taking, they would look less like social media feeds and more like structured communities of inquiry — where it's safe to be wrong, revisions are expected, and it’s valuable to show your thinking.
Where this leaves us
It’s true that many of the affordances of current digital environments — the visibility, the speed, the easy answers, the algorithmic optimization for engagement over reflection — align almost perfectly with what we'd predict to decrease intellectual risk-taking and undermine intellectual character development.
But, what I hope this post leaves you with is this: culture and context can matter, even when we’re thinking about digital tools. Norms are malleable. AI can be a thinking partner or a thinking replacement, depending on how we teach students to use it. Influential peers can be threats or models, depending on what kind of behavior we make visible.
The work for educators, parents, designers, and researchers right now is to actively mediate our students’ digital environments rather than assume the tools will do the developmental work for us. That means professional development, not just tool adoption. Norm-building, not just rule-posting. Designing for slowness, not just engagement. Teaching students to use AI to deepen their thinking, not offload it.
The good news is that more and more people are working on this. The bad news is that this work requires intention, and intention is exactly what most digital environments are designed to bypass. We’re going to need to exercise our own intellectual character – intellectual attentiveness, intellectual thoroughness, intellectual humility – in order to create a digital world where our children’s intellectual character can thrive.
References
Baehr, J. (2011). The inquiring mind: On intellectual virtues and virtue epistemology. Oxford University Press.
Baehr, J. (2013). Educating for intellectual virtues: From theory to practice. Journal of Philosophy of Education, 47(2), 248–262.
Bandura, A. (1977). Social learning theory. Prentice Hall.
Beghetto, R. A. (2009). Correlates of intellectual risk taking in elementary school science. Journal of Research in Science Teaching, 46(2), 210–223.
Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher, R. W. Pew, L. M. Hough, & J. R. Pomerantz (Eds.), Psychology and the real world: Essays illustrating fundamental contributions to society (pp. 56–64). Worth Publishers. https://bjorklab.psych.ucla.edu/wp-content/uploads/sites/13/2016/04/EBjork_RBjork_2011.pdf
Buunk, A. P., & Gibbons, F. X. (2007). Social comparison: The end of a theory and the emergence of a field. Organizational Behavior and Human Decision Processes, 102(1), 3–21.
Cialdini, R. B., Reno, R. R., & Kallgren, C. A. (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58(6), 1015–1026.
Clark, S., Mucinskas, D., Barendsen, L., McHugh, K., & Cantor, A. (2025). Global youth perspectives on “good work” role models: a qualitative study of character strengths and behaviors. Journal of Character Education, 21(1-2), 5-29.
Dweck, C. S. (2006). Mindset: The new psychology of success. Random House.
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2), 117–140.
Gottfried, A. E., Fleming, J. S., & Gottfried, A. W. (2001). Continuity of academic intrinsic motivation from childhood through late adolescence: A longitudinal study. Journal of Educational Psychology, 93(1), 3–13.
Han, H., Kim, J., Jeong, C., & Cohen, G. L. (2017). Attainable and relevant moral exemplars are more effective than extraordinary exemplars in promoting voluntary service engagement. Frontiers in Psychology, 8, 283.
Han, H., Workman, C. I., May, J., Scholtens, P., Dawson, K. J., Glenn, A. L., & Meindl, P. (2022). Which moral exemplars inspire prosociality? Philosophical Psychology, 35(7), 943–970.
Kapur, M. (2008). Productive failure. Cognition and Instruction, 26(3), 379–424.
Kiesler, S., Kraut, R., Resnick, P., & Kittur, A. (2012). Regulating behavior in online communities. In R. E. Kraut & P. Resnick (Eds.), Building successful online communities: Evidence-based social design (pp. 125–178). MIT Press.
Kurek, A., Jose, P. E., & Stuart, J. (2019). 'I did it for the LULZ': How the dark personality predicts online disinhibition and aggressive online behavior in adolescence. Computers in Human Behavior, 98, 31–40.
Lapidot-Lefler, N., & Barak, A. (2015). The benign online disinhibition effect: Could situational factors induce self-disclosure and prosocial behaviors? Cyberpsychology: Journal of Psychosocial Research on Cyberspace, 9(2), Article 3.
Lepper, M. R., Corpus, J. H., & Iyengar, S. S. (2005). Intrinsic and extrinsic motivational orientations in the classroom: Age differences and academic correlates. Journal of Educational Psychology, 97(2), 184–196.
Lockwood, P., & Kunda, Z. (1997). Superstars and me: Predicting the impact of role models on the self. Journal of Personality and Social Psychology, 73(1), 91–103.
Matias, J. N. (2019). Preventing harassment and increasing group participation through social norms in 2,190 online science discussions. Proceedings of the National Academy of Sciences, 116(20), 9785–9789.
Paluck, E. L., Shepherd, H., & Aronow, P. M. (2016). Changing climates of conflict: A social network experiment in 56 schools. Proceedings of the National Academy of Sciences, 113(3), 566–571.
Prentice, D. A., & Miller, D. T. (1993). Pluralistic ignorance and alcohol use on campus: Some consequences of misperceiving the social norm. Journal of Personality and Social Psychology, 64(2), 243–256.
Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.
Ritchhart, R., Church, M., & Morrison, K. (2011). Making thinking visible: How to promote engagement, understanding, and independence for all learners. John Wiley & Sons.
Sharples, M. (2023). Towards social generative AI for education: Theory, practices and ethics. Learning: Research and Practice, 9(2), 159–167.
Soutter, M., & Clark, S. (2023). Building a culture of intellectual risk-taking: Isolating the pedagogical elements of the Harkness method. Journal of Education, 203(3), 508-519.
Suler, J. (2004). The online disinhibition effect. CyberPsychology & Behavior, 7(3), 321–326.
Veenstra, R., & Laninga-Wijnen, L. (2022). Peer network studies and interventions in adolescence. Current Opinion in Psychology, 44, 157–163.
Wallace, J. B. (2023). Never enough: When achievement culture becomes toxic — and what we can do about it. Penguin.
Wills, T. A. (1981). Downward comparison principles in social psychology. Psychological Bulletin, 90(2), 245–271.