AI, and the end of ‘I Don’t Know’

I sometimes wonder how much of another human being we can ever really know. We watch what people do, listen to what they say, and fill in the unseen parts for ourselves. Most of the time, we probably don’t even realise we’re doing it.

For instance, we may decide that someone is upset, jealous or hiding something. We presume that we understand why they behaved as they did, although we may have very little evidence beyond the fact that their behaviour affected us.

My interest in courtroom trials has made me particularly aware of this. A witness can appear calm and convincing whilst another struggles to remember events in a neat order. We may instinctively believe the person who tells the better story - yet confidence isn’t evidence of honesty, just as visible distress isn’t evidence that someone is unreliable. Human beings don’t always look the way we expect truthful people to look.

The same assumptions follow us into the workplace. A colleague who speaks with confidence is often considered more capable than somebody who pauses to think. The person willing to say ‘I know’ may command more attention than the person honest enough to say ‘I’m not sure’. If a dispute develops, the person who explains their version calmly may be believed over someone whose frustration has accumulated across months of smaller incidents.

A colleague may appear emotionally detached for reasons we know nothing about. They may communicate differently or cope with pressure in a way that doesn’t fit our expectations. We can do considerable harm when we decide we know what another person feels, particularly when our conclusion is that they feel nothing.

Neither should we ignore the evidence of a person’s actions because we cannot know what’s happening inside them. If somebody repeatedly shifts blame, damages colleagues or changes their account depending on the audience, those behaviours matter. We don’t have to decide what kind of person they are before acknowledging what their actions are doing to others.

Perhaps this is where human judgement becomes most valuable. We have to resist inventing motives and remain willing to examine patterns. That requires more effort than attaching a label. It also requires the possibility that our first interpretation may be wrong.

The Dunning–Kruger effect is often explained in simple terms: the less we know about something, the harder it can be to recognise how poor our understanding is. Someone with limited knowledge doesn’t necessarily have enough knowledge to see the gaps. As we learn more, the subject often becomes less certain rather than more straightforward. We begin to notice the exceptions and the parts we hadn’t thought to ask about.

I wonder what happens to that process when AI can fill every gap almost immediately.

We no longer have to sit with the discomfort of not knowing. We can describe a colleague’s behaviour to an AI tool and receive a plausible explanation in seconds. If we word the prompt from the position that the colleague is manipulative, the response may help us build that case. If we approach it believing that we are the victim, it may produce language that confirms the part we’ve already assigned ourselves.

The result can look researched and psychologically informed. It may contain terminology we hadn’t previously known and explain the situation with a confidence we couldn’t have produced alone. But the tool hasn’t met the colleague. It hasn’t seen our own behaviour or heard the conversations that we forgot to mention. It’s working from the evidence we chose to provide, including all the assumptions contained within it.

That doesn’t make AI useless. It does mean that its fluency can make borrowed certainty feel like personal understanding.

We can already see a version of this on social media. People are able to form fixed opinions about strangers from a short recording or somebody else’s description of an event. They become angry with anyone who interprets the same fragments differently. AI now gives those opinions structure and an appearance of authority. It can make an untested belief sound like a considered judgement.

My concern isn’t simply that this will produce more keyboard warriors. I’m more troubled by what it might do to curiosity. If an answer is always available, will we continue to look beneath it? If our opinions can be justified on demand, what will make us wonder whether the original opinion deserved to survive?

This is also why I’ve begun to question what self-development means in 2026. For years, people have been encouraged to become more confident, to trust themselves and stop seeking approval. There was good reason for much of that, particularly for those whose voices had been dismissed. Yet I’m no longer certain that greater self-belief is always the development we need.

Perhaps we now need to become better at recognising where our knowledge ends. That isn’t the same as becoming timid or allowing other people to think for us. It means caring enough about the truth to accept that we may not possess it yet.

Without that ability, we become easier to divide. We can be kept arguing with one another, each side armed with convincing explanations generated from its existing beliefs. At the same time, we may become less willing to examine the information presented by institutions, technological companies or those with political power. We will feel as though we’re thinking independently because the opinion sounds like our own, without noticing how much of the thinking has been supplied for us.

The future of our species may not depend on everyone becoming more knowledgeable. It may depend on enough people retaining the humility to know that information and understanding aren’t the same thing.

When we encounter the confident colleague, the distressed witness or the person we’re tempted to call a psychopath, we may never know exactly what’s happening inside them. We can still pay attention to their actions and the effect those actions have. We can listen without treating performance as proof. Most importantly, we can leave enough room in our judgement for evidence we haven’t encountered yet.

I don’t know whether AI will weaken that human capacity or force us to value it more. I do know that a society in which everybody can produce an answer, but fewer people remain curious about whether that answer is true, feels like a worrying place to be heading.