Why Language Understanding Is Not the Same as Judgement

A comfortable assumption runs through modern institutions: that better information produces better judgement. Give decision-makers more data, more summaries, more analysis, and decisions will improve. Two decades of studying judgement in complex organisations — and a prior career engineering operational systems — have convinced me the assumption is wrong in an instructive way.

Information and judgement are different kinds of things. Information describes a situation. Judgement takes responsibility for acting within it. Between the two sits interpretation: the work of deciding what the situation actually is, whose account of it to trust, and which of its tensions matter. That interpretive work is where institutions quietly succeed or fail.

Language sets the conditions of judgement

Decision-makers rarely encounter situations directly. They encounter language about situations: briefings, assessments, dashboards, recommendations. That language arrives pre-framed — it has already decided what is relevant, what is normal, what is thinkable. By the time a leader 'exercises judgement', much of the judgement has already been exercised by the words in front of them.

This is why understanding language is not the same as judging well. A system — or a person — can parse every sentence of a briefing and still miss what the briefing does: the responsibility it diffuses, the assumption it smuggles in, the contradiction it papers over with confident prose.

Where machines fit, honestly

Machines are now fluent. They summarise, draft and answer with remarkable competence, and it is tempting to read fluency as understanding. But fluency is a property of text; judgement is a property of persons who can be answerable for outcomes. No system should be described as judging, because no system can be held responsible.

What technology can do — and what we are building ImaginAND to do — is make the interpretive work visible. Surface the contradiction between a strategy and its operating model. Flag where responsibility is assigned to no one. Show where language has drifted from stated values. These are aids to judgement: they enlarge what the responsible human can see.

The test of any decision-support system is not whether it sounds intelligent. It is whether the person who remains accountable understands the situation better after using it. That is the standard we hold TheAristotle to.