Training pigeons to solve problems is like training a neural network. In this respect informs University of Iowa.
The researchers gave the pigeons complex categorization tests where logic or reasoning would not help. Instead, by trial and error, the pigeons were able to remember the answer correctly enough to start sorting objects with 70% accuracy.
Ed Wasserman and his colleagues have developed a test they call “diabolically difficult.” Each experimental pigeon was shown a visual stimulus (pattern) and had to peck a button on the right or left to decide which category the stimulus belonged to. To determine the category, it was necessary to “measure” the width of the line, its angle, and the features of the round patterns. The correct answer gave the bird a delicious cake; wrong answer did nothing. According to Wasserman, the test was difficult because of its arbitrariness: no rules or logic will help solve the problem, just too much experience.
“These incentives are special. The scientist says that they are not alike and never repeat. “To complete the task, you must memorize the individual images or the principle behind them.”
Each of the four pigeons started with the correct answer about half the time. But after hundreds of tests, the birds increased their score to 68%.
“Pigeons are like artificial intelligence,” says Wasserman. “They use a biological algorithm given to them by nature, whereas a computer uses an artificial algorithm given to them by humans.”
Strictly speaking, both pigeons and artificial intelligence use associative learning (rather than logical learning, which is the human way of sorting things). The main difference between artificial algorithms and natural ones is the huge computer memory and the ability to “remember” tens of thousands of cases.