This chapter discusses the scientific controversies that have shaped neural network research from a sociological point of view.

It looks at the controversy that surrounded Frank Rosenblatt’s perceptron machine in the late 1950s and early 1960s. Rosenblatt was well aware of the main problems of his machine, and that he even insisted on them in his books and papers. Emphasis is given on one of the main problems of early neural network research, namely the issue of training multilayer systems.

The chapter analyzes the main results of that project, and shows that Minsky and Papert, and neural network researchers interpreted those results rather differently. It discusses the processes through which this interpretative flexibility was closed and the effects that the crisis of early neural network research had upon the 3 most important neural network groups of the time, namely Widrow’s group, Rosenblatt’s group, and the group at SRI.

The chapter also looks at the influence that factors like the emergence of symbolic artificial intelligence (AI) and computer technology had on the closure of the neural network controversy. After the closure of the perceptron controversy, symbol-processing remained the dominant approach to AI over the years, until the early 1980s. Some of the most important aspects of that changing context are reviewed and the history of backpropagation is discussed.

Introduction: A Sociological View of Scientific Controversies
The Controversy of the Perceptron
The Problems of Early Neural Networks
Training Multilayer Networks: A “Reverse Salient” of Neural Network Research
Interpretative Flexibility
Closure of the Controversy 1: Widrow’s Group
Closure of the Controversy 2: The SRI Group
Closure of the Controversy 3: Rosenblatt
The 1980s: A Changing Context
History of Back-Propagation
Back-Propagation: Learning in Multilayer Perceptrons
The Neural Network Explosion
The Current Debate: Conclusions
1. Debate Continues
2. Conclusions
Appendix 1: List of Those Interviewed
Appendix 2: List of Personal Communications by Letter

The author Mikel Olazaran spent a long time in the early 1990s interviewing what looks have been almost all the surviving connectionists & Minsky etc.

Olazaran argues that all the connectionists were perfectly aware of the Perceptrons headline conclusion about single-layer perceptrons being hopelessly linear, which drafts had been circulating for like 4 years beforehand as well, and most regarded it as unimportant (pointing out that humans can’t solve the parity of a grid of dots either without painfully counting them out one by one) and having an obvious solution (multiple layers) that they all, Rosenblatt especially, had put a lot of work into trying.

The problem was, none of the multi-layer things worked, and people had run out of ideas. So, most of the connectionist researchers got sucked away by things that were working at the time (eg. the Stanford group was having huge success with adaptive antennas & telephone filters which accidentally come out of their NN work), and funding dried up (for both exogenous political reasons related to military R&D being cut, and just the lack of results compared to alternative research programs like the symbolic approaches which were enjoying their initial flush of success in theorem proving and Samuel’s checkers player etc, and had not run headlong into the wall of Moravec’s paradox).

So when, years later, Perceptrons came out with all of its i’s dotted & t’s-crossed, it didn’t “kill connectionism” because that had already died. What Perceptrons really did was it served as a kind of excuse or Schelling point to make the death ‘official’ & cement the dominance of the symbolic approaches. Rosenblatt never gave up, but he had already been left high and dry with no more funding and no research community.

Olazaran directly asks several of them whether more funding or work would have helped, and it seems everyone agrees that it would’ve been useless. The computers just weren’t there in the ’60s. (One notes that it might have worked in the ’70s if anyone had paid attention to the invention of backpropagation, pointing out that Rumelhart et al doing the PDP studies on backprop were using the equivalent of PCs for those studies in the late ’80s, so if you were patient & dedicated, you could hypothetically have done them on minicomputers/mainframes in the ’70s. But not the ’60s.)]