What could it actually learn?

During the next decade, researchers built Perceptrons, ran experiments, and asked a harder question: what could this design actually learn?

In 1969, published a book called Perceptrons.

They asked a narrower question than the newspaper headlines:

What could these machines actually compute?

Their book used mathematics to map the abilities and limits of particular Perceptron designs.

The limits were real, but the book did not prove that every learning machine was doomed. Its authors also wrote that further progress was possible.

First, we need to understand the limit they found.