Recommendations are responses to requests that take the form “What should I do?” Recommendations have the general form: “You should do X .” Recommendations, that is, suggest actions.
Often, explanations are embedded in recommendations. Here, in separating investigation from recommendation, we have slowed the process, taking things one step at a time, in order to clarify our thinking.
But not all recommendation requests are motivated by specific problems. Some are simply requests for guidance. Some are requests for prediction. Here we will distinguish these as particular kinds of recommendation requests, giving each a subtly different structure.
As always, we’ll start with a distinction, in this case, three varieties of request: Requests for Solution, Requests for Guidance, and Request for Prediction. There are no solutions without problems, there is no need for guidance without a lack of information, and no need for prediction if we already know what will happen.
Requests for Solution (RFS) are motivated by explanations, in particular what to do about some anomalous behaviour.
Requests for Guidance (RFG) are motivated by lack of information, and driven by preference.
Requests for Prediction (RFP) are motivated by uncertainty about an outcome.
In general, Recommendations have four parts: Recommendation Request, Contextual Considerations, Supporting Resources, Rival Recommendations.
Requests for Solution, Requests for Guidance, and Requests for Prediction differ in their Contextual Considerations. In RFSs, context divides into Ordinary Conditions and Distinguishing Conditions. In RFGs, context divides into Aspirations and Aversions. In RFPs, context divides into Expectations and Uncertainties.
Let’s first develop a formalisation of these three recommendation categories. Then, we’ll consider nuances of these characterisations through a variety of practical examples.
A Recommendation Request is usually motivated by a quandary. There is a problem we want to solve; we don’t know how to proceed under some given circumstances; we might want to mitigate risks, given some prediction of how things might go.
We need a systematic way to express the structure of a recommendation so to evaluate rival recommendations and pick out a best, preferred one. Call this the Best Recommendation
Best Recommendations fit consistently with the events, explanations, and conditions under which the recommendation request was made.
Unreliable Recommendations fail to account for some or all of the events, explanations, or conditions under which the recommendation request was made.
Here it is useful to compare against the standards we set out at the beginning of this text, in particular, consistency and comprehensiveness. As we expected of explanations, we expect that recommendations will take into account as much of, if not all of, the context and resources we have included. Best Recommendations, that is, meet the standards we expressed at the beginning of this text.
Note that calling a recommendation “unreliable” doesn’t mean that it won’t have some effect in some given circumstances. Your least reliable friend might sometimes be right on time to pick you up from work. The oldest and shoddiest of cars often do the job just fine. “Unreliable” simply means there’s another option that would, you imagine, accomplish the task better or more reliably. It is this latter option we’re chasing here, and we will call that the Best Recommendation.