By Faye Miller.
What makes it so challenging for research teams to be truly receptive to being wrong? And what can teams do to make doubt expressible and useful?
Being aware that knowledge is always situational, incomplete, and prone to error, as well as the willingness to hold opinions tentatively, be receptive to change, and recognise the boundaries of understanding, are all components of epistemic humility. Humility about what one knows and can know is an intellectual quality in individual researchers. Yet epistemic humility also has a structural dimension: which doubts get expressed, whose knowledge is heard, and how teams handle what they don’t yet know, are challenges that go beyond the individual researcher to shape how research teams function.
This i2Insights contribution is an attempt to highlight two challenges that need to be addressed:
- the confidence trap, built on the pressure to exude certainty, and
- the silence trap, arising from the social dynamics that can suppress productive doubt.
Neither represents a failure of individual character or commitment. Both are shaped by forces larger than the individual. The confidence trap shapes what crosses the boundary between teams and the outside world, while the silence trap shapes what can be expressed within teams.
The confidence trap
In the context of a research team, the confidence trap tends to mean only expressing conviction and keeping silent about areas of doubt. The trap functions through publication rules, funder expectations, and genre traditions that originate outside the team, ie., in journal conventions, institutional definitions of what counts as a finding, and the expectations of those who commission research. These are so embedded that they can be hard to recognise. Addressing the trap inside the team is necessary but not sufficient. The norms and expectations the team operates within shape what is possible, and deserve explicit attention, even when they are harder to change.
This is not an argument against confident communication and some simplification is necessary for that. When a team practises epistemic humility, they can be honest about which conclusions are still tentative and which are confident enough to act on the findings.
For example, one research team working on urban water policy introduced a two-column reporting format: one column for confident findings, and one for live questions, giving funders a clearer picture of where further investment would be most valuable, while preserving intellectual honesty.
The silence trap
In teams where individuals privately hold doubts, these doubts can go unvoiced. These dynamics can be self-sustaining. For example, early career researchers may find it costly to challenge a direction that senior colleagues have publicly committed to, and—for those senior colleagues—reputational investment can make it harder to stay open to complicating evidence. Close-knit teams can experience challenge as a threat to trust even when it is intellectually valuable. Cultural context shapes all of these. Those whose perspectives least fit within the dominant disciplinary framework, such as community partners, practitioners, and those from underrepresented research traditions, are often the first to hold back. Their quietness deserves attention, not assumption. Creating conditions for doubt means building practices that make it expressible across the full range of cultural and institutional positions in the room.
Some practical approaches to reduce the social cost of expressing doubt include:
- anonymising early-stage feedback,
- assigning a “critical friend” role within the group, with the specific task of voicing doubts, and rotating this role around the group, and
- creating space for concerns to be raised before group discussion occurs and then including the concerns in the discussion.
For example, one health research team tried closing meetings with every member submitting one anonymous question or concern via a shared form, with the lead researcher addressing these at the next meeting. Over time this seemed to normalise uncertainty as a shared resource, with several of the project’s most productive pivots coming from concerns that might otherwise never have been raised.
Identifying where the traps may be operating
At the start of a project or when making important decisions during a project, the following diagnostic questions might begin to help a team identify where the traps may be operating. Gathering answers from team members individually before sharing them with the group may produce more honest responses.
Confidence trap
- What three things are most uncertain in the direction the team is going?
By identifying specific uncertainties instead of vague ones, it is more likely that the issues will be discussed openly and monitored as the project progresses.
- Has the team identified and recorded the specific conditions or evidence that would require its primary conclusion to be changed?
This may help make clear what sort of evidence could contradict the main conclusion, identify any assumptions being made, and support a process of accountability as the research process continues.
Silence trap
- Who would find it hard to disagree openly, and how could they contribute safely?
Naming the people who face the highest social or cultural cost for voicing doubt, and designing a channel for them, can be more useful than a general invitation to speak up.
- When was the last time someone changed the team’s direction? What allowed that to happen?
This looks for evidence that expressed doubt is actually influencing decisions. If no one can answer these questions, that may be a signal worth paying attention to.
A note on artificial intelligence
Artificial intelligence tools can contribute to both traps if used without care. They generate outputs that sound assured regardless of the evidence, feeding the confidence trap. They can also deepen the silence trap: when an artificial intelligence output exuding certainty enters a discussion early, it “anchors” group thinking, so that the social cost of departing from it is raised, compounding the pressures that already make doubt costly to voice. My previous i2Insights contribution, A framework for navigating the impact of using artificial intelligence on collaborative research communication, offers a five-stage framework for navigating how artificial intelligence affects collaborative communication. That framework works best in teams that have already built the habit of naming uncertainty and voicing doubt.
Conclusion
The practical approaches described above can help teams make epistemic humility a working practice. One thing teams that practice epistemic humility tend to have in common is that they see not knowing as a valuable contribution, not a sign of weakness. Doubt that is voiced safely can become a source of change, and uncertainty that is named early can become a resource.
What are your experiences with cultivating epistemic humility? Do you see the confidence and/or silence traps in your work? What other challenges to epistemic humility in teams might exist? What other kinds of practices or conditions might help your team feel free to express doubt and stay open?
Use of Artificial Intelligence (AI) Statement: Artificial intelligence was not used in the development of this i2Insights contribution or the work on which the contribution is based. (For i2Insights policy on artificial intelligence please see https://i2insights.org/contributing-to-i2insights/guidelines-for-authors/#artificial-intelligence.)
Biography: Faye Miller PhD is a research director, knowledge broker, and career educator. As Founder of Human Constellation Consulting, she collaborates globally with technology companies, universities, and research organisations on social and ethical aspects of science and technology, digital and information literacy, transdisciplinary knowledge integration and shared understanding, all of which are research areas in which she has published. She is currently based in Canberra, Australia.
Faye Miller’s contribution is thoughtful and aptly identifies two social forces that militate against the expression of intellectual humility and therefore against its manifestation in research teams. Both the confidence and silence traps can be reinforced by fears of reputational damage if one expresses doubts, uncertainties, or ignorance. The silence trap often is a special kind of pluralistic ignorance. If no team member expresses any doubts, then each team member with doubts believes that they are the only one who has any doubts, and a team leader will believe that no-one on the team has any doubts. Dr Miller’s suggestions for managing and avoiding both traps are helpful and feasible.
I have what I hope are two useful points to add to Dr Miller’s framework. First, there is a genuine need for us as individuals to find ways to encourage our own intellectual humility. There is considerable evidence that we tend to be over-confident about how much we know, and some of us will even claim to know about things we’ve never heard of, including things that don’t exist. Experts fall prey to this as well as novices (Atir, Rosenzweig, & Dunning, 2024). Moreover, we tend to think we know more than most other people do (Zell et al., 2020).
Some of the social forces that Dr Miller highlighted motivate us to be overconfident about our knowledge, because they reward us for appearing to know a lot rather than for careful analyses of what we don’t know. Additionally, an important epistemic constraint also hinders our intellectual humility. We can detect false beliefs and meta-ignorance (not knowing what one doesn’t know) in others but not in our here-and-now selves. Thus, two kinds of ignorance that we can see in other people are denied to our self-awareness (Smithson, 2026: 50-51). Unless we reflect on how our past self has had false beliefs or meta-ignorance, we may be pulled into perceiving ourselves as less ignorant than other people.
The second point I’ll raise relates to Dr Miller’s useful note on artificial intelligence. Large language models (LLMs) not only exhibit overconfidence, but they also have not been trained to express uncertainty or ignorance. They are trained only to give answers no matter what, whence the occurrences of made-up answers (so-called “hallucinations”). Fortunately, there are a few initiatives toward training LLMs to express uncertainty (e.g., Yona, Geva, & Matias, 2026). Meanwhile, we can leverage a few indicators of LLM uncertainty by using prompts that ask LLMs to assess degrees of confidence and by detecting heterogeneity or disagreement among multiple LLMs’ responses to the same prompt.
References:
Atir, S., Rosenzweig, E. and Dunning, D., 2024. Does expertise protect against overclaiming false knowledge? Organizational Behavior and Human Decision Processes, 184, p.104354.
Smithson, M., 2026. The Psychology of Ignorance. Oxford, UK: Oxford University Press.
Yona, G., Geva, M. and Matias, Y., 2026. Hallucinations Undermine Trust; Metacognition is a Way Forward. arXiv preprint arXiv:2605.01428.
Zell, E., Strickhouser, J. E., Sedikides, C. and Alicke, M. D., 2020. The better-than-average effect in comparative self-evaluation: A comprehensive review and meta-analysis. Psychological Bulletin, 146(2), pp.118.
Thanks very much, Michael – your additions have opened several new dimensions to the framework.
Your framing of the silence trap as pluralistic ignorance, and the picture you described of how each team member might assume they’re the only one with doubts, while the leader thinks everyone’s confident, captures a shared misunderstanding, which is what the practical suggestions aim to address. This could help us reconsider how leaders should interpret such silences in practice.
The point you make about reputational fear running across both traps adds another layer. I initially saw them as shaped by separate forces, and while I think that distinction still holds, your insight shows there’s a hidden link between them that would be worth exploring further.
I find the point about missing our blind spots in the moment important. Meta-ignorance under these social conditions might call for greater emphasis on individual epistemic humility, in addition to the structural. The idea that looking back at past mistakes might help us catch what we currently miss opens up a new angle that relates to Hussein’s earlier points on our slow inner work of self-formation. You both seem to converge on the same point that maintaining humility requires continuous personal effort, even if structures offer support, they cannot replace it.
On AI, your observation adds another dimension to the discussion. I was thinking about how generative AI outputs can anchor group discussion, but it is important to understand how these tools shape the very foundations of the problems we talk about later. This appears to reflect Hussein’s observation mentioned earlier about how responsibility has shifted from the person to the technical systems. I think your suggestions about getting AI to evaluate its confidence and compare responses across different models are some of the most honest practical mitigation steps we currently have, along with practising information literacy in AI contexts.
I would welcome your thoughts on the parallel between AI overconfidence and human expert overclaiming. Both seem to involve appearing more certain than the knowledge warrants, and perhaps your thinking goes further on this than I have yet been able to explore. Thank you for sharing the references. I look forward to exploring them further, particularly your own work on ignorance, Atir et al. on expert overclaiming, and Yona et al. on metacognition.
Thanks, Faye, for your considered response and the linkages with Hussein’s insights. I’m no expert on AI, but I have been following the growing literature on overconfidence in LLMs. There are consistent findings that when LLMs are asked, they overestimate the probability that their answer to a question is accurate, often by more than 20% and to a greater degree than human overconfidence (e.g., Bentegeac et al., 2025). One study (Wen et al., 2024) also reported that large LLMs mirror human confidence miscalibration, i.e., overconfidence for challenging tasks and underconfidence for easy ones, whereas small models are always overconfident.
An unfortunate side-effect of getting LLMs to report their confidence levels is that it may influence human team members’ confidence. A recent preprint (Sun et al., 2025) reported that getting LLMs to provide both their answer and confidence rating boosted human confidence more than just providing the LLM’s answer, especially for tasks or problems where initial human confidence is low. Although this line of research has yet to be properly critiqued and replicated, it suggests that teams using LLMs should do so with some caution and a dose of healthy skepticism.
References:
Bentegeac, R., Le Guellec, B., Kuchcinski, G., Amouyel, P. and Hamroun, A., 2025. Token probabilities to mitigate large language models overconfidence in answering medical questions: quantitative study. Journal of medical Internet research, 27, p.e64348.
Sun, F., Li, N., Wang, K., & Goette, L. 2025. Large language models are overconfident and amplify human bias. arXiv preprint arXiv:2505.02151.
Wen, B., Xu, C., Wolfe, R., Wang, L.L. and Howe, B., 2024, October. Mitigating overconfidence in large language models: A behavioral lens on confidence estimation and calibration. In NeurIPS 2024 Workshop on Behavioral Machine Learning.
Extremely interesting thread. Following and learning from the exchange.
Thank you for these additional references, Michael. It is interesting that the early findings from Wen et al. suggest that more challenging tasks are where LLMs are most likely to be overconfident, and Sun et al. suggest that it is where initial human confidence is low that displaying an LLM’s confidence level most inflates human confidence in turn.
This leads me to think that rather than being discussed alongside any LLM output, the diagnostic questions I suggested – in particular asking what the three most uncertain things are in the direction the team is going (the question I think is most vulnerable to being displaced when an LLM response is presented alongside its confidence level) may need to be worked out beforehand. If a team identifies its uncertainties or doubts initially, those are already on the table and harder to displace. It is worth noting that the diagnostic questions were designed to encourage moments of private reflection before group discussion opens – whether that connects to anything deeper might depend on the personal formation Hussein described.
Your description of the silence trap as a form of pluralistic ignorance raises a question of whether the same dynamic of pluralistic ignorance could be at work in the confidence trap, when team members act confidently while privately doubting a conclusion, each believing the others are truly certain. If this is the case, the two traps might have a similar cognitive dynamic as well as an emotional driver in reputational fear, which would imply that they are mutually reinforcing in ways the framework has not yet fully addressed. This would also imply that some of the practical strategies I suggested for the silence trap might also be used for the confidence trap, including creating space for concerns before group discussion opens. I would be very interested in whether there is anything from your work that would support this.
I agree that it would be preferable to pose the diagnostic questions you’ve suggested before introducing LLM-generated materials, although there may be practical difficulties in guaranteeing that no team member has consulted a LLM before group discussion gets underway. Despite this, getting expressions of uncertainty or doubt on the table early should help to avoid the silence trap. It also could be useful for team members to “update” their uncertainty assessments after LLM outputs have been obtained, thereby providing a record of how the LLM materials have influenced the team’s appraisals of doubts and uncertainties.
And you’re right—pluralistic ignorance could contribute to or generate confidence traps. I raised this possibility in my “psychology of ignorance” book, but I noted that pluralistic ignorance hasn’t been investigated as a possible contributor to knowledge overclaiming, overconfidence, or over-placement (where we think we know more than other people do). It’s unfortunate that this research gap exists, but it may be due to lack of communication between researchers studying overconfidence (usually cognitive psychologists) and those studying pluralistic ignorance (social psychologists).
Moderator: the book Mike refers to is hot off the press: Smithson, M 2026 The psychology of ignorance. Oxford.
Thank you, Michael. The idea of updating uncertainty assessments after LLM outputs arrive is a practical extension I hadn’t considered. Thanks for highlighting that research gap – if the connection between pluralistic ignorance and the confidence trap can be established, it might be possible to use some of the same interventions for both traps. I look forward to exploring the references you’ve shared, including your new book. I’m grateful for the conversation
Beautifully articulated work. It’s a powerful reminder of what we need to hold onto whenever we practice transdisciplinarity with integrity. I especially appreciated how you highlighted the cultural pattern we’ve normalised: rewarding overconfidence (because that’s what is allowed for publication) at the expense of reflective, transparent engagement with our own ideas. Your point about self‑sustaining discourses is spot on; much of the literature we read easily falls into that trap: “I once said something, I don’t have the humility to revisit it, so I’ll keep publishing without addressing its shortcomings”.
Anil’s addition was excellent: “Some people may take great offence if someone challenges or questions the central dogma, while they might be perfectly fine with someone stoking their ill‑conceived notion, perspective and pompous ego.” It captures an unhealthy double standard in how we filter comments on our work and how selectively we engage with ideas.
In the Arabic scholarly tradition, humility is not just a virtue but a method, an epistemic stance. Scholars cultivated a long practice of engaging with knowledge while constantly reflecting on their own limitations. Omar ibn Abdulaziz’s reminder, “may God have mercy on the one who understands their own limits and capabilities”, is a call to continuous self‑examination. Another well‑known principle attributed to him is that saying “I don’t know” is a valid and even honourable answer.
This is the kind of humility we urgently need to mainstream in our own spaces. It requires holding one another accountable, but it is also a deeply personal, ongoing practice of reflection and self‑work.
Chapeau bas!
Hussein
Truly humbled by your kindness and magnanimity! Anil Wali
Thanks Hussein, for your insightful contribution to this conversation. Many of the points you’ve raised have resonated, in particular, your understanding of epistemic humility as a deeply personal and collective ongoing practice. The inner work, the ongoing self-examination you describe, might actually be the most challenging and least visible aspect. “I don’t know” as an honourable response is a much-needed reminder that different cultural traditions have epistemic stances that contemporary research contexts have long struggled to respect and develop. We have also collectively normalised rewarding overconfidence because publication and funding cultures have incentivised it, at the expense of the kind of reflective engagement with our own ideas necessary for knowledge-building with a sense of integrity. I think you make an important point here that the confidence trap goes beyond the team dynamic, that it is also a part of one’s scholarly identity, shaped by the systems in which we are operating.
Your remark regarding self-sustaining discourses is one of the most insightful expansions of the confidence trap I’ve encountered, and one which I think operates quietly within one’s scholarly identity, making it more difficult to reconsider previous work’s shortcomings. I also think this expansion ties in with the structural conditions suggested in this post – even the most thoughtful researcher is up against strong currents when journal standards and funder expectations actively promote consistency over openly expressed uncertainty. This is why I think both layers, the personal and structural, need to be addressed at the same time. The kind of inner work you describe becomes much more difficult to practise and sustain when it is not supported by the institutional framework.
In light of your work on transdisciplinarity in research and educational settings, I’m curious: are there any circumstances or frameworks you’ve observed that actually help scholars and students in carrying out that kind of inner work without facing too high social or professional costs? Are there institutional designs that allow for the kind of humility you describe to be both personally and structurally sustainable?
Thank you for contributing these traditions and perspectives to the discussion – it is genuinely appreciated, and also for your generosity in responding to Anil’s contribution. It’s wonderful to see the thread evolving into something greater than the sum of its parts. It is a conversation that needs to be kept alive in these times.
That is a brilliant way of connecting all the different points. Your question goes straight to the heart of what motivates most of my work and fuels my curiosity. What I keep observing, in conversations around transdisciplinarity or in Science and Technology Studies, is that we often become overly fixated on finding the right tool or technique: the right reflective method, the right framework for “responsible science,” the right checklist. These explorations usually begin from the assumption that if we just identify the correct technique, regardless of who is using it, we will achieve better practice.
In doing so, responsibility subtly shifts away from the person and onto the tool, framework, or method. And this is where my own exploration has pushed me to revisit how we understand ourselves as beings. Ronald Barnett’s work has been a great source of comfort for me here. He challenges the idea of students as consumers or bundles of competencies, and instead insists on understanding them in their full complexity as beings.
This resonates deeply with the work of al‑Ghazālī and, more recently, Syed Muhammad Naquib al‑Attas, who revisit the human as a composite of faculties: ʿaql (intellect), nutq (reason/speech), nafs (self), qalb (heart), and rūḥ (spirit). All of these must be engaged throughout the learning journey. Starting from this leads to a very different understanding of how we cultivate humility, reflection, and moral discernment. It becomes clear that self‑work happens across multiple layers of the person, shaped by home, social life, schooling, and the wider moral horizon.
This is why classical scholars distinguish between taʿlīm (the structuring of knowledge and education) and ta’dīb (the disciplining of the self through right conduct). Ta’dīb expresses the goal of recognising the “proper place of things” and cultivating the good person. It brings together taʿlīm with tazkiyah (the purification of the soul), the slow transformation that removes vices such as arrogance, greed, and anger, and cultivates patience, humility, and compassion.
This brings me to the second part. Your blog and the question you opened with: “What makes it so challenging for research teams to be truly receptive to being wrong.” Today we are almost allergic to the word wrong, because it carries a normative weight. But if we are serious about humility and reflection, we cannot avoid the normative and moral dimension.
And this is precisely what becomes difficult when we operate entirely within what Charles Taylor calls the immanent frame. Trying to account for “wrongness” without appealing to anything beyond the human circle pushes us toward technocratic or utilitarian answers. It becomes a matter of better tools, better procedures, better checklists, rather than deeper moral formation.
This is why your question is so important. It forces us to confront the limits of an exclusively immanent approach and to recognise that humility, reflection, and moral discernment require a horizon that transcends technique, a horizon that confronts moral questions!
Once again, I really appreciate the blog and the rich conversation it’s opening up!
Thank you
Hussein
Thanks Hussein for continuing this conversation. The difference between structured knowledge and self-discipline, as well as your observation that it is the tool or framework that now bears responsibility rather than the person, are issues I know are present in my work but which I have never really articulated as such. The concept of moral reasoning within the immanent frame is something I will reflect on as I feel it guides us to places transcending the pragmatic approaches raised here. I am familiar with the question in my own work wondering whether the lived experience of collaborative research builds the whole person or simply their capabilities. Thank you for opening up this new horizon of thinking for all of us. It seems vital for developing a genuine practice of epistemic humility. Is anyone in this discussion familiar with approaches that address some of these points? Or perhaps you have experienced this tension within your own research tradition?
Definitely an agenda item for the first meeting of every newly formed research team, or existing team starting a new project. Faye your paper provides an excellent focus to open the discussion. We all need to accept that the traps are a characteristic of the ‘human team dynamic’. Perceiving the presence of the ‘traps’ as a normal team behaviour that needs to be understood and openly addressed, is a mark of a mature and stable team in any area of life’s endeavours..
Thank you so much Bruce, what a practical and generous suggestion! I think it’s a great idea to make this a regular topic on the agenda from the first meeting – straightforward yet has a big impact. A crucial point about human team dynamics – we might change the focus from blame to understanding when we reframe the traps as natural aspects of human team dynamics rather than viewing them as failures. That reframing is itself a mark of team maturity, in research, clinical settings, or as you’ve mentioned, any collaborative endeavour in life. I’m genuinely pleased to hear that the post might be a helpful starting point for having these kinds of open, judgement-free discussions.
Well, it is a tactical balancing act to give due credence to confidence trap versus silencing trap in various situations and circumstances in team group settings among competitive competing priorities and personalities. Some people may take great offence if someone challenges or questions the central dogma, while they might be perfectly fine with someone stoking their ill-conceived notion, perspective and pompous ego. A genuine well-intentioned team member may become reluctant under these situations to provide actual pragmatic nuances and hidden unintended consequences, so s/he may choose the silence route for not bursting the bubble and create unnecessary chaotic situation in group dynamics. Team science among multi- and trans-disciplinary circumstances boils down to the efficiency in achieving the goal with least resistance and friction. So, we have to take all these underlying factors of introvert versus extrovert and well-intentioned constructive feedback with a grain of salt. How do we give our honest feedback when it might be misconstrued as delaying tactics by someone who is blindsided by their unfortunate confidence trap? It is tough balancing act, you let it go and hope good sense will prevail and that the team makes a collective group decision that leads to the fabulous impactful outcome! Kudos to Dr. Faye Miller for bringing this to our attention. Much appreciated! Anil Wali
Thanks very much for your comment, Anil – I really appreciate your response grounded in pragmatism! It’s very interesting to consider the idea of a balance between the confidence trap and the silencing trap that you’ve aptly captured. I believe this is possibly one of the hardest things in connection with multi- and transdisciplinary team science, and yet one of the things that we rarely talk about openly. Being open about providing an honest opinion and at the same time making sure that the team stays united is likely another layer of balancing added to the former through the practical use of emotional intelligence.
This is something that makes me think of the following question: what can team leaders proactively do to encourage psychological safety, to allow team members to decouple the message from the messenger? Could this work towards being able to transform any tensions between keeping the peace and speaking the truth into genuine collective impact?
Thanks Faye, Glad to see level of interest this topic has created as it happens to be close to our hearts on everyday dynamic interactions and engagements with our colleagues and peers. Hussein Zeidan has rightly pointed out the old traditional wisdom, Man’s got to know his limitations echoed by Clint Eastwood. You have raised two important questions in terms of decoupling the message from the messenger and maintaining the peace and tranquility of the workgroup dynamics. You have rightly pointed out the “Emotional Quotient versus Intelligent Quotient”, well EQ always supersedes IQ, like Imagination and connecting dots always supersedes bookish knowledge without any life experience, humility and empathy!
Thank you, Anil. What a wonderful discussion this is turning into!
You made important points about EQ over IQ – this gets straight to the heart of the matter. The bridge you have drawn between Hussein’s intellectual tradition and Clint Eastwood’s maxim serves to remind us that this piece of wisdom has long been an integral part of various cultures, and maybe that universal appeal is its strength. It also seems from the connection between imagination, experience and empathy that epistemic humility depends on the nature of one’s engagement with others and with the world – something which cannot be developed without direct experience. This comes through clearly in Hussein’s remarks on continuous self-examination.