AI Strategy 5 - Knowledge

Knowledge is the explanation of what the information shows happened. The information reports that a figure moved. The explanation states why it moved, and that is something the information does not. For example, three months of sales at 100, 105 and 110 is data. Sales rising by five units a month is information. A new advertising campaign contributing to the rise is the explanation, and that is knowledge. An explanation usually depends on a pattern, which is the same thing observed happening before under conditions somebody judged to be similar. That explains why experience is worth what it is worth.

Knowledge is distinct from information insofar as information is accurate or inaccurate relative to the data it was assembled from - the lineage test in chapter four settles which. Knowledge is sound or unsound relative to the evidence for it, which is a separate test, and one which is much more difficult to ascertain. Unit sales might have increased due to the contribution of a new advertising campaign, the season, a competitor’s stock problem, or a previous price change only now kicking in.

The third of our maps records the explanations the business unit acts on. As usual, we record five components of each:

1. statement - the explanation itself, and which piece of information it explains;
2. evidence - what it depends on, and whether that can be produced on request;
3. status - whether it was arrived at by a defined process or by one person;
4. storage - where it is recorded, and in what form; and
5. accessibility - who can read it, and who can change it.

The distinction between a defined process and a habit, which chapter four applied to information synthesis, applies again here and even more so. A defined process for producing an explanation is written down and produces the same explanation from the same information independently. However, this is a rare occurrence, in our experience. Where there may be a defined process for producing the report, and the conclusion is recorded, the process on how the conclusion was reached is not recorded systematically in such a way that it can be analysed and reproduced. It remains embedded in the prose of the report.

Where this knowledge is stored matters just as much in this third rung as it did for the previous two. An explanation is typically held in a meeting, a chat message, a presentation slide prepared for one occasion, or a cover note in an email, for example.

Chapter one set out the data strategy as a hypothesis. The explanation is what the hypothesis rests on. A business that intends to increase revenue by a quarter is acting on knowledge of what drives its revenue, regardless of whether that knowledge has ever been formally encapsulated. However, when it isn’t, it is more difficult to examine the strategy when it fails. It is difficult to determine whether the explanation was wrong or the execution was, if only the latter was recorded.

Knowledge held in one person’s head is the third of the five habitually unobserved parts we discover in an audit, listed in chapter two. It is what the business loses when that person leaves but isn’t necessarily what you would expect. The conclusions do usually survive - in a spreadsheet, or in an unwritten rule that others continue to adhere to. However, it is the reason behind them that goes. A rule kept after its reason has gone is correct until the circumstances that produced it change, and then it persists if the evidence for it is no longer known, meaning it cannot be revised.

In one of our audits the governance document named one system as the single source of truth and yet the flows defined in the same document did not use it. Where that happens, the business holds two accounts of how it works, and people act on the one that is not written down.

Chapter four asked whether somebody can change the information without changing anything it was derived from. With respect to knowledge, the question is who may change a recorded explanation, and on what evidence. In most businesses there is no rule at all. The test we apply is whether an explanation can be replaced by authority alone or only by evidence. Where seniority is sufficient, what the business has recorded is not knowledge but its most senior view of the moment, and it will be replaced by the next one. Where evidence is required, the requirement has to name what evidence would be sufficient.

This whole chapter is important in the context of AI because it will perform in the same way as a human given the responsibility of knowledge synthesis without formal recording of the process. Give a machine the information and it will faithfully produce an explanation. Act on the explanation and a consequence follows. Where the explanation was written down, with the evidence it rested on and the date it was accepted, the consequence can be set against it and one of the two can be corrected. Where it wasn’t, the business learns what happened without learning why.

Judgement, the subject of the next chapter, determines what decisions the business makes, at what level and by whom, how the relationship between a decision and the knowledge behind it is recorded, and how a decision that was wrong is told apart from one that was made badly. Judgement is applied knowledge, so it inherits whatever this rung produced, including the explanations, whether or not they were recorded.