Archive for the ‘deduction’ Category

Dealing with ideas as squishy as pattern recognition, learning, and analogy making

December 25, 2007

In a previous post I described the differences between induction and deduction: Humans are relatively good at induction and relatively poor at deduction; computers are just the opposite

The discussion of induction vs deduction is quite interesting and relevant to most everyone.  Here are further ideas from the book The Origin of Wealth by Eric Beinhocker:

Deduction only works on very well-defined problems such as chess moves; for deduction to work, the problem cannot have any information missing or ambiguity.  Deduction is thus a powerful method of reasoning, but inherently brittle.

While induction is more error prone, it is also more flexible and better suited for incomplete and ambiguous information that the world throws at us.  It thus makes evolutionary sense that we would be built this way.

Through induction, humans are able to deal with ideas as squishy as pattern recognition, learning, and analogy making.

Note that models of induction featuring pattern recognition and learning have become a staple of computer science research, and many of these models are used in practical applications that range from recognizing the faces of terrorists at airports, to recognizing fraudulent charge patterns on credit cards.

Humans are relatively good at induction and relatively poor at deduction; computers are just the opposite

December 21, 2007

Induction is reasoning from a limited number of observations toward a general conclusion. A classic example: After observing that 2 or 10 or 1,000 ravens are black, you may decide that all ravens are black.

Another way of thinking about induction is that it is reasoning by pattern recognition – we fill in the gaps of missing information.

With deduction you start with a set of possibilities and reduce it until a smaller subset remains. For example, a murder mystery is an exercise in deduction. Typically, the detective begins with a set of possible suspects — the butler, the maid, the business partner, and the widow. By the end of the story he has reduced this set to only one person: “The victim died in the bathtub but was moved to the bed. Neither woman could have lifted the body, nor could the butler with his war wound. Therefore, the business partner must have committed the crime.”

Humans are relatively good at induction and relatively poor at deduction. Any of us is capable of instantly recognizing a face (an inductive task), yet most of us would have a tough time quickly doing the deductive calculation:

(239.46 x 0.48 + 6.03) / 120.9708

Computers are relatively poor at induction and relatively good at deduction. A simple pocket calculator can quickly and perfectly do the calculation, while it is a very hard programming challenge to get even a powerful computer to accurately recognize a face.

The Origin of Wealth by Eric D. Beinhocker and Logic for Dummies by Mark Zegarelli