They were right

Through the 1970s and into the 1980s, a small group kept working on exactly that problem.

Little funding. Little recognition. The mainstream had moved on. But these researchers were focused on one question: how do you train a network with multiple layers? How do you take a mistake at the end and work backwards to fix the right weights deep inside the network?

They found the answer. It was called backpropagation — a method for tracing errors back through every layer and adjusting each weight accordingly.

With that, everything the single Perceptron couldn't do suddenly became possible.

They were right. The next chapter picks up where they left off.