Q&A: Learning from Experience
Learning from Experience
Question
Hello Rabbi,
I listened to the first lecture, and I’d like to suggest an additional explanation for what you call “intuition.”
In my humble opinion, a reasonable explanation is that what you call intuition is exactly what machine learning is based on: a very large collection of observations that ultimately teaches the computer a pattern.
The point is that the computer cannot explain why, on a particular question, it gives one answer rather than another. That is, unlike an ordinary observation, there is no connection between a particular observation and the corresponding “scientific” conclusion.
In a similar way, one could say that human learning is divided into conclusions that we can say we inferred from a concrete observation (empiricism), and conclusions that we simply “know” as a result of the accumulation of observations we have gone through in our lives (perhaps “observation-based rationalism”).
I’d be glad to hear your opinion on the matter.
Thank you,
Asaf
Answer
The comparison to machine learning is problematic. A machine does not learn, because it has no consciousness and no thought. It merely operates. Therefore, machine learning only explains how the machine comes to operate in a given way. But regarding human beings, I am asking how they arrive at their conclusions. This is a question about the thinking that underlies the activity, not about the activity itself. See Column 35.
Discussion on Answer
Obviously intuitions are also a product of accumulated experience. But it cannot be only that. Take, for example, the least-squares technique, which prefers a straight line. The software assumes a preference for a straight line and a least-squares criterion, and that itself is an a priori assumption that it did not learn from experience or training, but was inserted into it by the programmer. Machine learning as well is done on the basis of various assumptions.
If intuition stems only from accumulated experience (which reminds me of the solution to Mill’s problem about the source of deduction), then what about the non-empirical part? Also, it’s difficult for me, because if so, how do we know that the intuition of choice is real? Maybe it is really a whole set of prior causes, and it only feels like choice because it is not a present cause?
I didn’t write, nor do I think, that intuition is only accumulated experience. On the contrary, I explicitly wrote that it is not.
That was my question: what is the part that isn’t? And how do you distinguish it?
I don’t know how to give a general criterion. As Hume wrote, the principle of causality, for example, is not learned from experience.
I’m using machine learning as inspiration, not as a comparison.
The claim is that just as machine learning is based on many observations (and its result is giving a certain answer without a direct logical inference from each concrete observation, but rather a kind of pattern inference from the totality of observations), so too human learning may take place מתוך an aggregate of observations that a person experiences over the course of life, from which he learns patterns that he will later call “intuition.”