The future is already here — just not evenly distributed*
The PolymathMind.ai Substack from Adam Riley and I this week makes an argument I want to push one step further.
The data shows that human involvement in AI-assisted decisions remains high – the UK Government’s AI Adoption Research found 84% of businesses using AI report at least some human input or checking, with 67% reporting significant input.
On the face of it, this is reassuring. Humans are in the loop. But “in the loop” is not the same as “meaningfully engaged.”
A person can be technically present at a decision while lacking the time, training, authority or confidence to actually challenge an AI output. A manager can remain accountable for a recommendation without understanding how it was produced. A team can check an AI-generated
answer superficially – reading it for fluency rather than interrogating the assumptions beneath it.
This is the gap that worries me most. Not whether humans are present, but whether their presence is actually doing anything. The risk is what we might call ‘decorative or performative accountability’ … the appearance of human oversight without the substance of human judgement. It satisfies governance frameworks. It survives audits. It is also exactly the condition under which serious errors get waved through, because the people who could/should have caught them were structurally unable to do so.
Closing this gap is not a technology problem. It is a capability problem. It requires teaching people to challenge fluent, confident systems – which is harder than teaching them to use those systems in the first place. Most organisations have invested heavily in the second and barely at all in the first.
If we want human judgement to remain real rather than ornamental, that imbalance has to change.