Marvin Minsky

Why don't computers have common sense? Why can't they learn from experience? If we want our machines to catch up with us, we must teach them the things that "everyone knows." Only then can they share and promote our goals.

Today, we have programs that outperform people in quite a large range of useful domains. Some can solve complicated equations. Others can recognize voices and faces, assemble cars in factories, and even pilot ships and planes. But none of them can make a bed, or clean a house, or baby-sit, or do most of the things that our two-year-olds do. Why can't we make our computers do so many things that most people can do?

We shouldn't expect much resourcefulness from a program that was designed to work only within some particular realm, by using only a single technique. If a program works in only one way, then it gets stuck when that method fails. But a program with multiple "ways to think" could behave as humans are prone to do: whenever one gets frustrated enough, a person can switch to some other approach—perhaps through a change in emotional state.

So instead of blaming machines for their faults, we must make them more resourceful. We can help them with this by endowing them with large collections of general knowledge, and with multiple methods for using these—in short, by giving them more "ways to think."