When a company's shared space for turning know-how into knowledge includes an AI, what actually changes?

This explores ba, the Japanese word Ikujiro Nonaka uses in his theory of the knowledge-creating company for the shared space where knowledge is actually made. It is not a database or a document. It is the meeting, the apprenticeship, the long conversation, where people with different hunches build a common sense of what matters until some of that sense can be written down. Nonaka's key point is that tacit knowledge, the know-how people can't fully put into words, only turns into organizational knowledge inside such a shared context. The corpus takes up the obvious next question: what happens to ba when one of the participants is a model?

The sharpest answer is in Can machine tacit knowledge spread across organizations unlike human tacit knowledge?. Human tacit knowledge stays local. It lives in particular people and spreads slowly, through working alongside them. Machine tacit knowledge, the patterns a model has absorbed but can't explain, can be copied to every company that uses the model. That sounds like it makes ba unnecessary, but the note argues the opposite. Portable machine know-how only becomes knowledge an organization can use through human-machine feedback loops that build shared context and meaning. In other words, it needs a ba. The model brings scale. The shared context is what makes the scale mean anything.

That is why the corpus keeps coming back to what AI output lacks: context. Is AI returning knowledge to flow-based economies? argues that AI brings back a kind of knowledge that flows instead of sitting still in books, as it did in oral cultures, but without the speaker or giver who used to anchor it. Does AI-generated knowledge have the same structure as hearsay? goes further: AI output looks like hearsay, passed along at a distance with no traceable origin. And Does AI separate intellectual form from the thinking behind it? shows AI producing the finished form of intellectual work without the reasoning behind it. Read through Nonaka, these are all descriptions of knowledge cut loose from its ba: output that arrives without the shared situation that would let you judge it.

The practical notes show what rebuilding ba can look like. In Can codified expertise let non-experts match specialist output?, experts wrote their hard-won rules into an agent's setup, and non-experts then produced expert-rated work, a 206% quality jump that came from the scaffolding, not from a bigger model. That is Nonaka's 'externalization' step (turning tacit know-how into explicit rules) done through software. Does refusing explicit knowledge harm AI system performance? shows the cost of skipping that step: models that learn only from data, with no explicit knowledge put in, are harder to interpret and more brittle. Even Do language models know what they don't know about users? fits here. Giving a model a structured list of what it doesn't yet know about the user, a basic piece of shared context, cut sycophancy and hallucination roughly in half.

What you might not expect is that ba may matter more now, not less. Can AI generate knowledge faster than humans can evaluate it? warns that AI produces claims faster than people can judge them, and judging is exactly what a shared context exists to do. Do humans and LLMs differ fundamentally or just superficially? suggests why a human-machine ba is possible at all. Seen from outside, people and LLMs are utterly different, but inside a conversation they draw on the same language and meanings. So whether AI adds to an organization's knowledge or just floods it may depend less on the model than on whether anyone builds the shared space where its output gets interpreted.