Request for Prediction
for Prediction are commonly found in the sciences, in engineering, in
risk management, and so forth. RFPs are fundamental in Predictive
Troubleshooting, as described in previous chapters.
In RFPs, there is no problem to be solved, and there are no normative standards against which we compare recommendations. Instead, we make analogies between series of events we’ve seen in the past to hypothetical series of events we expect in the future.
The series of events we’ve observed in the past are the patterns that result from some kinds of investigations. Underlying all of this (and perhaps all of science) is the expectation that the future will be like the past. That much we needn’t specify. However, we need to specify how potential, future cases resemble past cases, so to strengthen the analogies we make.
Formalisation of Request for Prediction
Recommendation Request (RFS) is a well-framed question that asks for a likely outcome given some circumstances and expectations.
Expectations (EXP) resemble Ordinary Conditions, as described in Requests for Solution. They provide the context in which uncertainties occur.
Uncertainties (UNC) are descriptions of possible unexpected effects or results in the given circumstances. These are expressions of distinctions between the set of circumstances under which a pattern was discovered, and the set of circumstances currently under consideration.
Rival Recommendations (RR) are suggested outcomes in the given circumstances.
Supporting Resources (SR) are descriptions of the circumstances under which the extra-ordinary events are happening, or under which the recommendation is made.
Best Recommendation (BR) is the recommendation we offer as best among the possibilities.
Some Remarks on Patterns & Predictions
Predictions are forward-looking. We often make predictions in scientific or systematic pursuits, and in either we answer questions of the form “What happens when ?” Predictions are especially sensitive to circumstances, and so the clarity and accuracy in our characterisations of circumstances will significantly affect the quality of our predictions.
Many sorts of investigations into systematic behaviour share the form “What happens when some circumstances occur .” For example, predicting weather and race winners requires that we take conditions and patterns into account. This evidence, taken together, gives us reason to think some outcome most likely. In these cases, “most likely” shares enough features of “best explanation” that we can treat them similarly.
Sea Biscuit, the racehorse, might have a history of running poorly in muddy conditions, which is good reason to think he might not win on a rainy day. This one explanatory resource gives us reason to think he will not win in the rain, but also explains the result when he doesn’t. The difference is whether we’re explaining the past or predicting the future.
In general, we can think of Scientific Predictions as we think of deductive arguments: we aim to show that the conclusions are guaranteed to occur. We can think of Conditional Predictions as we think of non-deductive arguments: the conclusions take the form of very likely suggestions. The distinction is, roughly, between “must” in the scientific case, and “should” in the conditional case.
Scientific Predictions offer conclusions we think must occur.
Conditional Predictions offer conclusions we think should occur.
Expectations are patterns. Recall that one result of an Investigation is a pattern. In this case, we’re using the results of an investigation to underwrite our confidence in a certain outcome. However, when we ask for a prediction, we tacitly suggest that we don’t know what’s going to happen, despite the patterns we might have already discovered.
Recall also that Prediction is the future-tense version of Explanation. Rather than asking about what happened, here we’re asking about what will happen. Note that asking about the future depends on analogical reasoning, where we believe the future will be like the past, except for some difference. The two situations are relevantly similar. It is important to express those possible differences, which we turn to now.
If you know what’s going to happen, then there is no motivation to request a prediction. That is, Requests for Prediction are motivated by uncertainty, which means, roughly, you don’t know for sure what might happen in some given circumstances. We work this into our structure.
Usually we have some guess what might happen (expectations and patterns), but we know that our given circumstances might differ relevantly from the established patterns we cite in Expectations. We express the distinctions here as Uncertainties.
For example, we might have previously noticed a pattern: Whenever Freddy competes in a bicycle race, he wins. If there were a bicycle race coming up, and let’s imagine we know Freddy has entered the race, we might ask for a prediction of who will win.
RFP: Who will win the University Bicycle Race?
EXP: Freddy wins every bicycle race in which he competes.
UNC: Freddy has been focusing on his studies rather than racing.
Though we wouldn’t explain why Freddy wins every bicycle race in which he competes, me might want to explain that he wins all the races. That is, we still want to provide some evidence that our Expectation is reasonable. This will have been the result of a Investigation.
LQ: What happens when Freddy races his bicycle?
EE1: He won in Auckland.
EE2: He won in Christchurch.
EE3: He won in Dunedin.
EE4: He won in Wellington.
RA1: Freddy always wins.
RA2: Freddy always wins in New Zealand.
RA3: Freddy always wins in cities.
ER1: These are all the races Freddy entered.
ER2: Freddy’s average speed was the highest in Australasia.
BE: When Freddy races, he always wins.
It is important to describe our discovery of the pattern, for in that description we can find sources of uncertainty. If the only information leading to the pattern, which is the Best Explanation above, does not include specifics about Freddy’s state of mind at the time, or how dedicated he was to racing when we discovered the pattern, then these might be uncertainties.
Notice the similarity to Ordinary Conditions and Distinguishing Conditions, as found in Requests for Solution. The patterns we use in support of our Requests for Prediction are ordinary conditions, in some sense. But the Uncertainties we specify are not Distinguishing Conditions, because the Uncertainties we specify are hypothetical. Distinguishing Conditions are actual.
Uncertainties could distinguish this particular case from the Expected pattern. Our purpose in making all of this explicit in our Request for Prediction structure is to give reasons to think whether our expectation that the pattern will continue is justified.