"Practice is an important part of learning, not studying. Studying is a complete waste of time. No one ever remembers the stuff they cram into their heads the night before the exam, so why do it? Practice, on the other hand, makes perfect. But, you have to be practicing a skill that you actually want to know how to perform."
Roger C. Schank
At present, human and machine learning are very different.
One of the big problems with both knowledge and learning is that of Monotonicity. Monotonic Learning is where new facts or knowledge do not conflict with previously known facts or knowledge. Monotonic learning implies that everything learned in the past is always true. Humans seem to cope well with non-monotonic learning, learning the truth about ‘Father Christmas’ for instance, but it is much more difficult to get computer algorithms to function in this way.
Remembering and forgetting are important issues in human learning and things that are not remembered may also not be forgotten. Computers on the other hand do not forget anything unless specifically instructed to do so.
One of the goals of Artificial Intelligence is to get computers to learn rather than have to be programmed to do everything. Although some of what humans do results from their genetic programming, much of what they can eventually achieve is as a result of learning.
As we try to develop better ways for people to learn and take advantage of new opportunities and new technology, we should also think about what machines should learn and the best ways for them to go about learning.
- Can everything be learned from scratch, or are there some things that must be pre-programmed in order for learning to take place?
- Will it be best for machines to learn in the same way humans do or should we devise more appropriate ways for machines to learn given their different construction and function?
- Will mutual machine - human learning (cooperative learning) be possible in the future?
There is an excellent web based publication about learning is ‘Engine for Education’ by Roger Schank. Roger has interests in both human and machine learning. This work concentrates on human learning.
There are also many good works on machine learning for instance, ‘Machine Learning, paradigms and methods’ edited by Jaime Carbonell. MIT Press. 1990.