1
Kinds of learning in the broadest sense of ways a mind may change from experience.
Remember: Simply the retainment of any belief beyond an instant. A common computer language tends to lose every computation's result because the language has no way to know the conditions of a result's truth. Pure functional languages cheat by contriving that all results are unconditional.
Mutate: Sense new beliefs, not limited to the belief and suppression of a few fixed ideas.
Habituate: A measure of one idea. Examples: overlook the useless flux of a belief, track the general value of a means.
Associate: A measure between ideas. Associations may run both ways, not distinguishing cause from effect.
Are other kinds of learning possible?
2
Cause vs. effect. Empirical causes are unprovable. All a mind can see are associations. The intended effect of an act, one amongst infinite effects, has no status outside the acting mind.
Empirical vs. logical causality. Empirical cause: a mind's belief in what must exist for another thing to exist. Logical cause: a thing's imagined parts.
Every thing has infinite causes but every mind is finite. A mind can only afford to know the causes that are likely to need the mind's action and that are within the mind's power.
What caused x? What precisely might we then mean by this question? If we wanted to end x, then the answer would be a state that's disbelief coincides with disbelief in x.
Is there an alternative to cause and effect?
3
Habitual blindness. The activity of a mind's senses could easily exceed the mind's power to process. Pursuing every inference from a sensation, and every inference from what's inferred, costs a mind energy or time. In a mind simulated sequentially, finite sensation queues habituate, losing sensations that threaten to bury the mind.
4
Contemporary intelligence research overrates learning. Minds that can't learn remain immensely useful and non-trivial to build well.
5
Why sleep? A need so large and dangerous must be important. Empirically, sleep seems defined by the isolation of the brain. The body is paralyzed and the senses closed while the brain works. Without new sensations, the brain could only process its memories—learn, model, simulate experiments.
A human brain in sleep becomes a passive mind preparing to better act when woken. Why can't this be done when awake? Is this a universal limitation of mind or a technical limitation of brain minds?
6
Present machine minds are largely either trivial, like speed governors, incapable of learning associations, or passive, like spam filters. An active mind that learns associations will not merely learn to infer plain facts but to infer action causing goals. This combination introduces interesting problems. Example: the validity of inferring from sensations likely caused by a mind's own acts? And reveals the priority of strength over learning: the odd acts of a mind that learns associations are even more dangerously unpredictable.
7
You shouldn't generalize. What might this mean? Is this advice well thought and well intended? What might be the alternative to generalizing? To applying memories of past things to those similar in the present.
A mind in all but the dullest universes must generalize to see associations. Without ignoring details, any association would be so specific that it could never reoccur.
A mind must not overgeneralize, missing important exceptions. Neither can it fail to generalize, never learning, forever repeating the same mistakes. How to know when to do which? How to know how long to spend deciding? How to know how long to spend deciding how to decide?
Take the most sensitive object: humans. TV and self-interest cause humans to tell each other to judge each person alone, as though a man is an atom, unchanging and indivisible. Is a single man qualitatively more real than a set of men? Isn't it a horrible prejudice to judge individually, to presume a man's behavior based on how another with the same name and a similar face behaved yesterday?
If a dog bites me, can't I strike more than its fangs?
Why not judge a man again every time you meet? As though he were a stranger. Or every minute? Wouldn't this justly recognize the fact that a man can change at any time?
Reductio ad absurdum. A mind balances between prejudice and judgment. Any absolute statement about what level to prejudge at will inevitably in some cases be mistaken, be a prejudice. Every non-trivial idea is a divisible bundle of impressions over space, time or both.
Overlooking differences, emphasizing similarities, has its political use but don't overlook the cost of pretending to be stupid.
8
Theory vs. action. Only action is real. Theory is prediction and merely improves the order of experiments, of acts.
9
Optimize vs. anticipate. Instead of laboring to improve one method from N2 to N steps, build a mind that learns to anticipate the need for any method's results. What matter if an algorithm and input takes an hour to finish when you know of the need more than an hour before it occurs? A mind that anticipates needs is the universal optimization.
10
Tabula rasa. Impossible in an opaque mind. For a blank mind to learn, it must have senses, which presume the knowledge to build them and the forms they impose on their input.
11
Certain effects tend to follow certain actions under certain conditions. Science is merely the formalization and institutionalization of the associative learning method in the unconscious human mind.
12
Can a parrot learn to reach ends? Is mind better imagined as a case of learning, not learning as a case of mind?
13
Learning: form vs. method. First define how a mind keeps learned knowledge. Prove that the form can hold the desired behavior. Last define the learning method that fills the form.
14
Recall feeling sad. You wished to feel otherwise. In that state, you had feelings not felt when happy. Your mind leapt to the belief that the odd feelings caused your sadness. Was it right? If you changed them, would you become happy? Or were those sad facts conscious because you were sad? Are they causes or effects?
15
Human minds leap to judgment. Flip a fair coin. It can easily show five heads in a row. But show a human five trials of anything unfamiliar and he will judge it. Even a scientific trial, of statistically significant length and difference, is uncertain. Your experience can always be a fluke. The human quickness to judgment is individually understandable when you can't divide the cost of an experiment across a scientific study's million readers.
16
Cause vs. coincidence. What divides a cause from a correlation? Action. If to a mind, acting to cause x leads to sensation of y then x causes y. It is only tricky to divide the two when you try to do so from passive theory.
17
Not only how best to learn, but when to learn? Learning isn't free. Animal brains limit some kinds of learning to childhood. Ideally, learn when to learn. Or make learning a means.
18
Methods of making and unmaking associations: cooccurrence, theory, action, pain.
19
How a mind can gain experiences to learn from.
Passively learn: learn without acting.
Actively learn: learn from acts with other intended effects.
Experiment: learn from acts taken with no intent but to learn for future use.
20
Teaching: a mind improving a mind in terms of the student mind's ends and indirectly, not injecting goals and inferences. The student mind needn't be conscious of what it is learning, as in the case of physical training, or even know that it is learning at all.
Only a mind that learns, in the sense of changing from experience, can be taught. A teacher's method is determined by the kinds of change that the student mind can learn. To train, the teacher must have a means to cause the acts that he wants to reinforce: injection, the power and desire to imitate.
The perfect teacher. Its ends are your ends. For it, teaching you is merely a means to your common ends. Teaching teleologically defined:
Goal: to ensure that a mind knows a, b, c.
First means: test the student to know what it knows.
Present state: the student knows a, b.
Other means of seeing that the student knows c: tell, train, …
For one mind to learn from another, both minds must sense and act on levels of the subject that are at least analogous to each other. Example: for a mind in a computer to learn from you, it must see the display and sense use of the keyboard and mouse.
Order of subjects. First teach a mind to master its immediate environment, the most urgent conditions of its survival. For a mind in a computer, don't start its schooling with chess, stock picking, or speech recognition, but freeing drive space and runaway process termination.
How easily can a mind master its environment? A computer mind's handicap: from nothing it must master a complex product of evolution and culture.
Comments