Every AI system that has ever existed — every large language model, every analytical platform, every recommendation engine — was trained on a corpus of documented human expression. Text, transactions, measurements, recordings. The intelligence in these systems is a function of the intelligence in the documents that trained them. That is a genuinely extraordinary capability. It is also a hard limit.
The limit is this: AI cannot access what has never been expressed. Everything it can process was first written, recorded, or transacted. The understanding that exists in people but has not yet been expressed — the reasoning behind the decision, the tacit knowledge that shapes the intuition, the genuine disagreement that gets smoothed over in the meeting, the peripheral insight that no one thinks is relevant enough to say — does not exist in any training corpus. It exists in people. And it is often the most important intelligence in the organisation.
The documentation gap
Most strategic decisions are made on the basis of documented knowledge — research, reports, data, precedent. The analytical tools organisations use to support decision-making are sophisticated at working with this layer. They find patterns in the data. They synthesise the reports. They surface the precedents.
What they cannot do is surface the undocumented understanding that precedes the documentation. The frontline worker's tacit knowledge of how the customer actually behaves. The experienced practitioner's sense that the strategic direction is heading for a failure mode she has seen before, expressed as a mild hesitation in a meeting rather than a documented objection. The philosophical challenge to a key assumption that was treated as obstruction rather than analysis and never made it into the record.
Every AI system is a sophisticated archive of what humans have already said. The decisions that matter most are shaped by what humans understand but have not yet expressed.
What gets lost in synthesis
There is a second limitation, related but distinct. AI synthesis tools can produce fluent summaries of large bodies of text. They are good at this. What they cannot do is preserve the epistemic diversity of the sources they synthesise. A summary collapses the difference between a claim grounded in expert research and a claim grounded in lived experience and a claim grounded in institutional assumption. They become equivalent sentences in a paragraph.
The diversity of ways of knowing — the range of epistemic grounds from which a community's understanding is formed — is not preserved by synthesis. It is eliminated by it. And that diversity is often the most important thing to know: not what the community concluded, but how different parts of it are thinking, what kinds of knowing are shaping the understanding, and where the genuine uncertainty is.
What collective sensemaking does that AI cannot
Collective sensemaking does not compete with AI. It addresses a different deficit: the deficit of undocumented understanding. A SparkMap creates the conditions in which people can build and express the understanding that has never been documented — with its epistemic ground visible, its connections to other perspectives preserved, and its genuine uncertainty acknowledged rather than synthesised away.
The output is not a better input for an AI system, though it can be that too. The output is genuine collective understanding: structured, characterised, preserved so that it can be worked with over time. Understanding that exists nowhere in any training corpus. Understanding that the most sophisticated AI system in the world cannot generate, because it has never been expressed — until now.
The decision it changes
The most consequential decisions an organisation makes are typically the ones made on the thinnest understanding: strategic redirections, major change programmes, decisions that will affect thousands of people in ways that no analytics platform can fully model. These are precisely the decisions where the undocumented understanding matters most — where the questions that have not yet been asked are as important as the answers to the ones that have. Collective sensemaking infrastructure is how that understanding gets built, expressed, and made available to the people who need it.