AI programs are given only one goal.
Even if there appear to be two goals, they can be prioritized and regarded as one goal.
There is only one problem that must be solved.
However, solving one problem may require solving another problem.
I would like to clarify what kind of problem occurs in which case.
To make some inductive reasoning, we need to solve the following sub-problems.
1. Guess what the next hypothesis will be
2. Guessing the value of objective variable and explanatory variable used for hypothesis
3. Inference of bias "unknown value" related to hypothesis
To solve these child problems, we need to recursively solve the grandchild problems.
Hypotheses that are exactly the same as hypotheses that have already been tested are not worth making.
There are countless hypotheses, but in order to reach your goal, you have to guess what kind of hypothesis is good.
For example, surrounding or combining hypotheses with relatively good results.
A good hypothesis can be statistically inferred from previous hypotheses and results for the current problem.
It is possible to statistically infer what kind of hypothesis was effective from examples of other similar problems.
Variables used in hypotheses may be inductively inferred values, including "unknown".
In that case, inductive inference of that value is required.
This applies not only to explanatory variables, but also to objective variables.
The bias of each sample used in the hypothesis is represented as an "unknown value".
Its value must be guessed.
This has to be guessed in every possible way.
For example, suppose you used some data in inference.
If you can guess the probability that the data is false, increase the "unknown value" by that amount.
It is difficult to guess everything about where and what causes of bias occur, but I would like to guess as much as possible.
It also infers statistically.