Top 15 Marvin Minsky Quotes

It would seem that Perceptrons has much the same role as The Necronomicon -- that is, often cited but never read.

I cannot articulate enough to express my dislike to people who think that understanding spoils your experience… How would they know?

Every system that we build will surprise us with new kinds of flaws until those machines become clever enough to conceal their faults from us.

We still remain prone to doctrines, philosophies, faiths, and beliefs that spread through the populations of entire civilizations. It is hard to imagine any foolproof ways to protect ourselves from such infections. ...the best we can do is to try to educate our children to learn more skills of critical thinking and methods of scientific verification.

Most of our future attempts to build large, growing Artificial Intelligences will be subject to all sorts of mental disorders.

I suspect our human "thinking processes" often "break down," but you rarely notice anything's wrong, because your systems so quickly switch you to think in different ways, while the systems that failed are repaired or replaced.

Perhaps it is no accident that one meaning of the word express is "to squeeze"—for when you try to "express yourself," your language resources will have to pick and choose among the descriptions your other resources construct—and then attempt to squeeze a few of these through your tiny channels of phrases and gestures.

We could extend them either by scaling up small connection­ist models or by combining small-scale networks into some larger organization. In the first case, we would expect to encounter theoretical obstacles to maintaining [generalized delta rule]’s effectiveness on larger, deeper nets. And despite the reputed efficacy of other alleged remedies for the deficiencies of hill-climbing, such as “annealing,” we stay with our research conjecture that no such procedures will work very well on large-scale nets, except in the case of problems that turn out to be of low order in some appropriate sense. The second alternative is to employ a variety of smaller networks rather than try to scale up a single one... No single-method learning scheme can operate efficiently for every possible task; we cannot expect any one type of machine to account for any large portion of human psychology.

More concretely, we would call the student's atten­tion to the following considerations: 1. Multilayer machines with loops clearly open all the questions of the general theory of automata. 2. A system with no loops but with an order restriction at each layer can compute only predicates of finite order. 3. On the other hand, if there is no restriction except for the absence of loops, the monster of vacuous generality once more raises its head. The perceptron has shown itself worthy of study despite (and even because of!) its severe limitations. It has many features to 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 judgment that the extension is sterile. Per­haps some powerful convergence theorem will be discovered, or some profound reason for the failure to produce an interesting “learning theorem” for the multilayered machine will be found.

Have you considered perceptrons with many layers? ... We have not found (by thinking or by studying the literature) any other really interesting class of multilayered machine, at least none whose principles seem to have a significant relation to those of the perceptron. To see the force of this qualification it is worth pondering the fact, trivial in itself, that a universal computer could be built entirely out of linear threshold modules. This does not in any sense reduce the theory of computation and programming to the theory of per­ceptrons.

[We] became involved with a somewhat therapeutic compulsion: to dispel what we feared to be the first shadows of a “holistic” or “Gestalt” misconception that would threaten to haunt the fields of engineering and artificial intelligence as it had earlier haunted biology and psychology. For this, and for a variety of more practical and theoretical goals, we set out to find something about the range and limitations of perceptrons.

We do not see that any good can come of experiments which pay no attention to limiting factors that will assert themselves as soon as the small model is scaled up to a usable size.

Perceptrons have been widely publicized as “pattern recognition” or “learning” machines and as such have been discussed in a large number of books, journal articles, and voluminous “reports.” Most of this writing...is without scientific value.

One popular version is that the publication of our book so discouraged research on learning in network machines that a promising line of research was interrupted. Our version is that progress had already come to a virtual halt because of the lack of adequate basic theories... Most theorists had tried to focus only on the mathematical structure of what was common to all learning, and the theories to which this had led were too general and too weak to explain which patterns perceptrons could learn to recognize... The trouble appeared when perceptrons had no way to represent the knowledge required for solving certain problems. The moral was that one simply cannot learn enough by studying learning by itself; one also has to understand the nature of what one wants to learn.

What magical trick makes us intelligent? The trick is that there is no trick. The power of intelligence stems from our vast diversity, not from any single, perfect principle.