Talk:Frank Rosenblatt
From Wikipedia, the free encyclopedia
 |
This article is within the scope of WikiProject Biography. For more information, visit the project page. | ||
| Start | This article has been rated as Start-Class on the project's quality scale. [FAQ] | ||
More biographical information needed, if available.
This article contains a lot of errors. Minsky is at MIT, not Carnegie-Mellon University.
(This error checked and Minsky's location corrected to MIT on 5/31/06)
And the descriptions of Minsky's work on perceptrons are highly inaccurate.
(See: "History of the Perceptron" http://www.csulb.edu/~cwallis/artificialn/History.htm which briefly documents the Rosenblatt-Minsky dispute over perceptrons and the effect of Minsky and Papert's 1969 book "Perceptrons". The 1969 first edition of "Perceptrons" made the false claim that the severe computational limitations Minsky found in simple 3-layer perceptrons would also apply to any arbitrary higher-layer-count perceptron (i.e. essentially to any neural net) -- a claim Minsky later retracted in the 1980's, after his 1969 original mistake had killed essentially all research into neural nets for a decade.)
This makes me doubt the accuracy of the rest of the article.
I'm glad that you corrected CMU to MIT. But whoever is writing this article, you should read the Minsky book on Perceptrons, instead of quoting some web page which is ridiculously wrong. What are your sources for the rest of the information in this article? How can I believe any of it now?
This article is an absurd and inappropriate character attack on Minsky, which should in my opinion be removed. I agree with those above - it contains a lot of errors.
[edit] A couple of additional observations
I took Rosenblatt's course in 1965-66 (or maybe 1966-67--who can remember such things?). Rosenblatt may never have responded publicly to Minsky, but he had some colorful things to say in class.
He told us that he and Minsky were high school classmates (Bronx Science, IIRC). This could presumably be verified. He also said that they remained personal friends although they bitterly disagreed about the best way to model the brain. Rosenblatt sought to build crude analogs of brain circuits. He was a student of brain anatomy and ran a lab at Cornell that conducted numerous experiments on spinal cats (cats whose spinal cords were severed below the neck). A lot of our course readings were in neuroanatomy. He was dismissive of Minsky's black box approach to artificial intelligence (as long as the results of some deterministic computer program somehow resembled human behavior, the program was therefore a good model of the way the brain works) and really didn't think there was any long-term value to Minsky's work. He considered Minsky a lesser intellect and chalked his behavior up to jealousy.
Rosenblatt, as the article says, was a colorful character. But he was not well-liked by faculty. Part of the reason was his extraordinary brilliance (the first person who ever made me feel like a plodding dolt in comparison). For example, he had worked out and patented some device that is used on the end of a telescope to do some sort of filtering (details are lost in the fog bank of my memory, but, again, this is something that could be researched at USPTO). Astronomers had been trying to develop such a device for some time and thus the astronomy department were alleged to be angry that this guy from another field entirely had beaten them to it.
It should be mentioned that Rosenblatt's partner in the Perceptron (the cover article in the August 1960(?) Scientific American, by the way), was Ernest Bloch (not sure I have that right, but I'm close--check the Scientific American archive to get the name right), of Cornell's electrical engineering faculty, who taught a complementary course to Rosenblatt's.
Dmargulis 15:53, 29 December 2006 (UTC)
I can't comment on the level of personal animosity between Minsky and Rosenblatt (though remarks in the literature suggest there was some degree of animosity) but for the impact of the Minsky-Papert work see e.g. "No harm intended" http://www.demo.cs.brandeis.edu/papers/perceptron.pdf The upshot is that Minsky did cause harm, and noting the fact in the article is quite compatible with NPOV.
- What is this review? Was this published, or is this just something a certain researcher wrote and passed along by the web? Where was it published, who refereed it? I can make a review myself too, with a beautiful references section and nice roman fonts, accusing this reviewer of making just "whimsical remarks", of having bad intentions and of appending an empty document to an already finished discussion... By the way, who is this anonymous person who posted this? -- NIC1138 (talk) 03:34, 25 March 2008 (UTC)
[edit] Even more additional observations
I wold like to dwell a little bit more on this confusing and controversial subject. First of all, I believe that even if what the article ending says is true, the tone is completely inadequate. And second, I believe it is not true.
We can't keep blaming a certain book for something of that magnitude. It is a grave offense to the hearts and minds of all persons involved in the research on AI in the 60s and 70s to say that a single thought in a book could have such an impact. That someone could single-handedly diverge everything for a decade. Anyone must agree that thing are never that simple. I could believe if you tell me for example something like “there was a trend growing, a feeling for concentrating into another direction, and the book was the excuse used”.
I liked very much to hear about all those facts regarding Rosenblatt. Looks like he was a great person. But how can someone believe the idea of the animosity between the researchers when the book Perceptrons is (if I am not very much mistaken) open-heartedly dedicated to Rosenblatt? More then that, I had the opportunity to have a read on his book, and on Perceptrons. I think they are two great books, and I feel that they can like perfectly side-by-side.
And please, someone tell me something: Where are all those marvelous multi-layered networks? To my knowledge, the usual structures, with an "input" layer, a "hidden" layer and an "output" layer, is still the most usual and useful. More layers then that are frequently an extravagance. Of course there are many justified cases, but the possibility to have many layers was never a breakthrough. And the most important thing: this 3-layers structure shows up in Rosenblatt's and Minsky's book. Some people imply that the middle layer was the 1980s revolution, this is plain wrong...
I intend to make some modifications to this ending of this article later this week, unless anyone vehemently opposes. -- NIC1138 (talk) 07:25, 22 March 2008 (UTC)
- I agree that the article is very badly written, and that the conclusion in particular needs help. But — and this is a very serious "but" — it did seem to me at the time that the book by Minsky & Papert did indeed have a very negative effect on the direction taken by artificial intelligence. I was a student in Rosenblatt's class in 1968-69 at Cornell, though my major was mathematics and my primary interests lay elsewhere. The pattern recognition problems that Minsky & Papert proved the perceptron could not solve were nontrivial topological problems. It was my opinion that they (Minsky & Papert) were correct, but that this in no way detracted from the solid abilities of the basic perceptron. It was asking too much of a simple feedforward network to expect that it should do well on nontrivial topological patterns, and everyone should have realized this. The brain itself is certainly not a feedforward mechanism, though it contains some primitive circuits of that nature. What really happened here, in my opinion, was that Rosenblatt and others expected too much from the simplest perceptrons, and then, when Minsky & Papert proved as much, everyone expected too little. As is so often the case, the truth was somewhere in between. Modern artificial neural network discriminators are the direct descendents of the perceptron, and Rosenblatt's reputation should rest on that solid foundation. — Aetheling (talk) 02:28, 26 March 2008 (UTC).
