研究者たちのケンカを原文で再現します。
--Minsky and Papert(1969)
The perceptron has shown itself worthy of study despite (and even because of!) its severe limitations. It has many features that attract attention: its linearity; its intriguing learning theorem; its clear paradigmatic simplicity as a kind of parallel computation. There is no reason to suppose that any of these virtues carry over to the many-layered version. Nevertheless, we consider it to be an important research problem to elucidate (or reject) our intuitive judgement that the extension is sterile. Perhaps some powerful convergence theorem will be discoverd, or some profound reason for the faiure to produce an interesting ``learning theorem'' for the multilayered machine will be found. (pp. 231-232)
--Rumelhart and McClelland(1986) p.361
Although our learning results do not quarantee that can find a solution for all solvable problems, our analyses and results have shown that as a practical matter, the error propagation scheme leads to solutions in vitually every case. In short, we beleive that we have answered Minsky and Papert's challenge and have found a learning results sufficiently powerfull to demonstrate that their pessimism about learning in multilayer machines was misplaced.
--Rumelhart and McClelland(1986) p.65
It was the limitations on what perceptrons could possibly learn that led to Minsky and Papert's (1969) pessimistic evaluation of the perceptron. Unfortunately that evaluation has incorrectly tainted more interesting and powerfull networks of linear thereshold and other nolinear units. We have now developed a version of the delta rule -- the generalized delta rule -- which is capable of learning arbitrary mappings.
--Minsky and Papert(1988) p.248
... But we also reiterate that the discipline can grow only when it makes a parallel effort to critically evaluate its apparent accomplishments. Our own work in Perceptron is based on the interaction between an enthusiastic pursuit of models of new phenomena and a rigorous search for ways to understand the limitations of these models.