Instrumental Learning is an early name for what was later reformulated as Operant Conditioning.

The gist is that a person or animal (or other) comes to produce some response which is instrumental, which serves as an instrument, in doing something else - that is, which achieves some goal for it.

It can easily be seen that this description, though very vague, applies very broadly - to most things which we would normally say is 'learned by doing' or 'learned from experience,' like riding a bicycle, deliberately making someone angry, or pressing a lever.

Not to press it too far - well, yes, to do that - but if you believed that 'knowledge that' is (somehow) 'knowledge how,' then you could also make instrumental learning do a lot of (but certainly not all of) the work of learning conceived as 'gaining knowledge.' (You could regard this as a bonus tip on how to deliberately make an epistemologist angry.)

The real difficulty in defining instrumental learning - that is, if you're B.F. Skinner - is that the notion of goal seems teleological, not behavioral. At any rate, there's something difficult about studying it scientifically. That motivated the Skinnerian definition of reinforcement, which turns the definition of instrumental learning as on its head by first defining the behavior involved (the person or animal or whatever begins to do something when doing that has a certain result, and does it less or stops when it is less likely to have that result), then claims that this is just what instrumental learning was all along. If you can buy that, then you are probably at least partly a radical behaviorist.

What this clever definition leaves out is major - any account of how it is that an animal can do such a thing in a way that works. Considered from the animal's point of view when it is 'reinforced,' the question is something like 'To what about what I was just doing - for however long - can I assign the credit or blame for that lovely reinforcement?' This is, for obvious reasons, called the assignment of credit problem, and it stuck in the early Marvin Minsky's craw something awful. How people and animals assign credit can be regarded as one of the biggest remaining problems in the study of natural instrumental learning. How to effectively assign credit, in the context of various learning problems, serves as a summary of many huge and interesting problems in machine learning:reinforcement learning, unsupervised learning...

All this makes instrumental learning a lot more interesting and basic than it usually gets credit for.

Log in or register to write something here or to contact authors.