# Why opposite sides of a debate share the same mistake > Source: > Published: 2026-07-16 14:34:59+00:00 In past weeks, I've been thinking a lot about agentic AI and its impact on human work in the long term. While doing so, I came across a wide spectrum of opinions from very pessimistic to naively optimistic, although I've rarely seen something in between. The sentiment itself stayed remarkably consistent across different contexts in these discussions. It seemed that a mental model wasn't built around specific topics, but rather around a *category umbrella*, in this case agentic AI. The real surprise came when I tried to turn this observation onto myself. I wanted to consciously build a genuinely balanced view of the whole thing, but the more effort I put in, and the deeper I dug to find the ultimate answer, the more uncertainty there was, and the more I *unconsciously* wanted to be on one side of the debate. That pushed me towards an uncomfortable question. Why is polarisation the human default, despite its flaws and a massive social cost? Constructing an opinion properly requires significant cognitive effort. First, we have to really *understand* the problem. Second, we need to find *precise* words to describe it. Third, we must *articulate* it, which validates the first two steps. None of this is trivial, so cutting corners is, in fact, a way to conserve effort [1]. This is why we follow people we trust and agree with: they don't just hand us an off-the-shelf explanation, they give us a safety net that lets us feel comfortable holding *any*opinion, as long as it's roughly in line with ours. AI itself is a good example of a category so broad and imprecise that it makes the public debate more engaging than it should be. Does someone mean LLMs, or the tooling built around them? Or a general artificial intelligence concept, which unsurprisingly is far from exact, since as human beings we still don't fully agree on what intelligence really is. What makes this worse is that people arguing under the same label often think they’re arguing about the *same* thing, when in reality the category is wide enough that each side is really talking past the other. Two people can spend an hour disputing AI, convinced they’re having the same conversation, while one means large language models and the other means some distant vision of general intelligence. And it’s precisely this capaciousness, the fact that almost anything can be poured into the same word, that makes the category so easy to fill with whatever we already believe. The category is wide enough to fit almost anyone's existing identity into it. When an opinion becomes part of who we are, it stops being just an opinion and becomes emotional [2]. We hold onto it at any cost, fighting for the truth as if it were the only one, ignoring valid arguments to the contrary . The more resistance we meet, the more that identity hardens. It's a reinforcing loop. [[3]](#fn3)This ambiguity is further exploited by engagement algorithms on social media, which favour vivid, polarising discussions over correctness and balance [4]. Creators capable of making an accurate point feel pressured to oversimplify it, so that it's understood by an average person unfamiliar with the topic, just to avoid being penalised by the algorithm. And the more feedback they get confirming this is the right approach, based on the sheer number of reactions, the more they stick to it, if only to protect their income. Similarly, the companies driving this transformation have little incentive to change that behaviour, since it keeps investors satisfied. Again, we see a reinforcing loop, this time at the sociological level. Being wrong and biased is relatively low-risk when it only affects an individual sharing their view within a closed circle. But it becomes a multiplier inside organisations, where decisions carry real stakes. It's easy to imagine the blast radius of a wrong decision made by a senior leader, with consequences that might play out over years and affect hundreds of people. The longer a leader holds a public opinion, and the more senior they are, the higher the political and psychological cost of changing their mind. Publicly admitting to being wrong is widely perceived as a loss. In fact, it's the opposite [5]. What's more, once a decision is made by a senior leader, it quickly becomes part of the organisation's identity, framed as *our*strategy. Changing course at that point has consequences not just for that person, but for the whole organisation the strategy was built around, and it can produce resistance from teams, investors, or the market, regardless of whether the decision was right. Since we can't really avoid being biased, given how difficult it is to deep-dive into every single topic, the first thing we can do is simply *acknowledge* that this bias exists (though it's still worth trying, for the topics that matter most, to rely on our own judgement). Being familiar with the pattern makes it easier to catch ourselves in the act of unconscious thinking. Divide-and-conquer helps too: narrowing a problem down into smaller pieces gives our minds some breathing room and helps avoid brain fog. Finally, systemic solutions can act as guardrails at the organisational scale. Making decisions collaboratively lowers the impact of any single person's bias, and building a culture where that's the norm helps. Just as powerful is an attitude that stays open to changing course and feels comfortable not being right. I remember a few years back when I really wanted to be right in every discussion. It was a way for me to prove my value, since I believed a leader had to be someone who knows. Then I switched to asking the right questions instead of having ready answers on the spot, a value the company held at the time. That shift was a real game changer, and it still holds true today. *Need for cognitive closure*(Kruglanski, A.W. & Webster, D.M., 1996): introduces the construct describing the motivation to reach a firm answer and avoid ambiguity, with two consequences termed seizing (grabbing the first available answer) and freezing (holding onto it regardless of new information).[↩︎](#fnref1)*Affective polarization*(Iyengar, S. & Westwood, S.J., 2015): shows that partisans increasingly evaluate their own side positively and the opposing side negatively regardless of actual policy agreement, with implicit partisan bias exceeding implicit racial bias.[↩︎](#fnref2)*Identity-protective cognition*(Kahan, D.M., 2012): experimental study finding that subjects who scored highest in cognitive reflection were the most likely, not the least, to display ideologically motivated reasoning.[↩︎](#fnref3)*Algorithmic amplification of division*(Piccardi, T. et al., published in Science, 2024): field experiment on X/Twitter showing that reducing exposure to hostile, partisan content in the feed measurably lowered affective polarization, causal evidence that engagement-based ranking amplifies division rather than merely reflecting it.[↩︎](#fnref4)*Wrongness admission and trust*(Fetterman, A. et al., University of Houston, 2026): scientists who admitted a prior research finding was wrong were rated as more trustworthy, competent and effective than those who did not.[↩︎](#fnref5)