Explanation and Prediction ​

Now we can say what prediction is, how it is constructed, and how it is related to explanation.

Prediction is a statement about the future world (possible world) with some epistemic status.

Example:

I say "it will rain tomorrow" — this is a prediction, it means that I believe/think it will rain tomorrow, but I don't communicate my level of certainty, and I don't say anything that could indicate it, and others can only guess about it from my tone.

Example:

I say "I am almost certain that at least half of the students will be able to defend the final project for the course" — here is my degree of certainty expressed by the words "I am almost certain". It could also be expressed quantitatively, "I am 80% sure that at least half of the students...".

Note that no explanatory model has been given in any of these predictions so far.

In a naive understanding of the scientific method, the following opinion is widespread: we create all explanations to generate predictions. This is not entirely correct. We create explanations for their explanatory power, which includes generating predictions, but also includes other things.

The point is that predictions are related to direct observations. And the same observations can correspond to completely different explanations and, consequently, to different larger parts of the world view.

By observation, we will understand not in the sense of the experience that the agent receives, and not in the sense of the process, but in the sense of the type of statements that concern the surrounding world and to which one could substitute "I observe that".

Example:

[I observe that] "Vasya is talking to Masha, "it is raining outside".

But not: "you are sad" (we do not observe how another person feels); "you are incompetent" (we do not observe the level of competence directly, only consequences from — presumably — this level).

Some parts of reality are available for observation (often including our inner reality); some others are available with distortions because our mediating means introduce distortions; other parts are not available for observation for technical reasons; and some are completely unobservable.

From whether we can observe some part of reality, we can conclude whether we can make a prediction about it.

Of course, we would like the explanation to help better predict some observable parts of the world.

But if we compare explanations only by predictive power, then paradoxes arise.

Let's assume that in the case of a camera, we have (1) an explanation about the genie; (2) an explanation about the photo-printing mechanism. We observe how the button is pressed and the photo is printed, but we can explain this in different ways.

Note that both of our explanations have the same input data and provide the same predictions: if we press the button, we will get a printed photo. If we were only seriously interested in predictions, we should say that the explanations are the same — they predict the same thing. But in fact, they lead to different, more general pictures of the world because they use different assumptions and are generally integrated into the rest of reality in completely different ways.

Assumptions are statements about the world that have to be considered true without the possibility of verification in order to formulate an explanation. If it turns out that to formulate an explanation, it is necessary to consider assumptions true that are not present in the agent's view of the world or that the agent considers false (or doubts too much about them), then the explanation turns out to be difficult or impossible to formulate.