AI supports judgement, it does not replace it
AI is becoming genuinely useful in capital projects. It can assemble context from scattered documents, compare perspectives across disciplines, retrieve the reasoning behind earlier decisions, identify gaps in an argument, generate counter-arguments and monitor drift from the approved decision basis. Work that once took days of reading and reconciling can now take minutes.
That is real value. But it is worth being clear about what AI produces.
It still produces a representation. Bernardo Kastrup puts the distinction bluntly: “I can simulate in a computer all the chemical reactions that take place in human kidneys. Yet, this is no reason to believe that the computer will start peeing on my desk. A simulation of the phenomenon isn't the phenomenon.”
The same holds for every tool a project team uses. A model can calculate exposure without carrying the risk. A schedule can sequence work without performing it. A cost report can describe overrun without paying for it. And AI can summarise a conversation without owning the decision that comes out of it.
Responsibility does not transfer to the representation. When a project goes wrong, it is people who answer for it: the sponsor who approved the case, the team who shaped the scope, the engineers who signed off the design. They carry the consequences, so they must carry the judgement. A tool that cannot be held accountable should not be the one deciding.
Consentropy therefore uses AI as partner, memory and mirror.
As a partner, it widens the inquiry. It surfaces perspectives that might otherwise be missed, asks the awkward question nobody in the room wants to raise, and offers counter-arguments to test a position before it hardens.
As memory, it holds the reasoning behind decisions, not just the decisions themselves. When conditions change months later, the team can see what was assumed, what was agreed and why, rather than reconstructing it from fragments.
As a mirror, it reflects the team’s thinking back so they can examine it. It helps people challenge their map: the shared picture of the project they are working from, with all its gaps and hidden assumptions.
Its own output must remain open to challenge. This matters more than it might seem. Human–AI research shows that incorrect automated advice can pull judgement towards error, including after people initially reach a sound view.¹¹ A confident, well-written answer carries authority it has not earned. The risk is not only that AI gets something wrong. It is that people defer to it when their own first instinct was right.
So the discipline runs both ways. People should use AI to test their thinking, and they should test what AI tells them with the same rigour. Every summary, comparison or recommendation is an input to the conversation, not a conclusion to it.
The temptation with any powerful tool is to use it for speed: faster answers, faster sign-off, faster closure. But in capital projects, premature closure is often where the trouble starts. Decisions made before the problem is properly understood tend to come back later, at greater cost.
Use AI to strengthen inquiry, not to create faster closure.
