Marvin Minsky
Yesterday brought the sad news that Marvin Minsky passed away at age 88. I never met Minsky (I wish I had); I just had one email exchange with him back in 2002, about Stephen Wolfram’s book. But Minsky was my academic great-grandfather (through Manuel Blum and Umesh Vazirani), and he influenced me in many other ways. For example, in his and Papert’s 1968 book Perceptrons—notorious for “killing neural net research for a decade,” because of its mis- or over-interpreted theorems about the representational limitations of single-layer neural nets—the way Minsky and Papert proved those theorems was by translating questions about computation into questions about the existence or nonexistence of low-degree polynomials with various properties, and then answering the latter questions using MATH. Their “polynomial method” is now a mainstay of quantum algorithms research (having been brought to the subject by Beals et al.), and in particular, has been a mainstay of my own career. Hardly Minsky’s best-known contribution to human knowledge, but that even such a relatively minor part of his oeuvre could have legs half a century later is a testament to his impact.
I’m sure readers will have other thoughts to share about Minsky, so please do so in the comments section. Personal reminiscences are especially welcome.
Follow
Comment #1 January 26th, 2016 at 12:04 pm
As an MIT sophomore (“wise fool”) math major I started to take Minsky’s course in AI, cross-listed in math and electrical engineering. It was before there was a computer science major, and before there were AI textbooks, so we used his grad students’ dissertations, later to be published as Semantic Information Processing. (Names you would know like Patrick Winston, Bertram Raphael, Danny Bobrow, and John McCarthy.) There were “no prerequisites,” but I was glad to have had 18.03 Calculus III. The first assignment was “hill climbing.” I could do it. After that I was in over my head. It’s the only course I’ve ever dropped. Since then, I’ve taught AI a dozen times. Minsky was patient, but he did not suffer fools gladly — even “wise fools.”
Comment #2 January 26th, 2016 at 9:05 pm
At EXTRO-3 in 1997, Eric Drexler said he had a nightmare in which he spent centuries trying to explain to the AIs of the future why he had let his thesis supervisor die irreversibly, and then he gave Minsky a cryonics suspension contract. Minsky accepted bashfully, saying that perhaps he had been put off by the paperwork.
Comment #3 January 27th, 2016 at 12:36 am
It’s interesting that Minsky was sceptical of Bayesianism and didn’t believe that statistical methods alone could get to AGI.
Minsky emphasized the importance of a variety of techniques, rather than trying to rely on one grand idea:
“You don’t understand anything until you learn it more than one way”.
Here’s an interesting paper where he talks about different reasoning methods and suggests that different methods apply at different levels of abstraction:
web.media.mit.edu/~push/StThomas-AIMag.pdf
That’s my view of things also. It is just plain foolish to think that probability theory is going to continue to apply at all levels of epistemological abstraction; thinking that ‘probability’ is going to continue to apply in epistemological domains way outside those it’s worked in so far is just as silly as thinking Newtonian physics can continue to work in black holes.
Comment #4 January 27th, 2016 at 1:37 pm
just when there are huge breakthroughs happening in practical AI
http://spectrum.ieee.org/tech-talk/computing/software/monster-machine-defeats-prominent-pro-player?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed:+IeeeSpectrum+%28IEEE+Spectrum%29
Comment #5 January 30th, 2016 at 3:22 pm
[…] Update (2016-01-30): Scott Aaronson: […]
Comment #6 February 1st, 2016 at 1:15 am
Is it possible to start on a new project after 40 and do breakthrough work after 40 if you are a male scientist?
Comment #7 February 1st, 2016 at 7:43 am
Scott: It’s probably not connected directly to Minsky, but can you please comment on the recent advances in computers playing the Go game?
Comment #8 February 1st, 2016 at 8:54 am
jonas #7: It’s an amazing achievement!
My Go knowledge is surely insufficient to get much out of them, but does anyone have a link to the actual games against the European champion that the machine won?
Comment #9 February 1st, 2016 at 8:56 am
eli #6: Yes, it’s possible (also if you’re a female scientist), though surely not easy. Exercise for the reader to give examples.
Comment #10 February 1st, 2016 at 3:27 pm
Scott #8
Comment #11 February 1st, 2016 at 3:34 pm
Scott,
what is opinion on the issue of AI becoming one day too advanced for our own sake?
People like Stephen Hawking and Elon Musk seem pretty concerned about it ( http://observer.com/2015/08/stephen-hawking-elon-musk-and-bill-gates-warn-about-artificial-intelligence/ )
Comment #12 February 1st, 2016 at 4:52 pm
Eli #6 and Scott #9:
As someone who recently turned 40, I too would like to know the answer to whether “old dogs…” (i.e., scientists past 40) “…can learn new tricks” (i.e., embark on new projects and make major contributions).
As a theoretician myself, I’m happy to have merely a plausible intuition upon which to build and then some sort of existence proof, so here goes:
Plausible Intuition: I imagine if one (A) has a solid foundation of knowledge in the field, (B) has developed a good nose for new fields with low lying fruit, and then (C) is lucky to live in a time where there are new academic fields and/or practical applications that arise in her or his 40’s or 50’s that have lots of low lying fruit, then the answer is “yes, old scientific dogs can most certainly successfully learn new tricks”
Existence Proof: Max Born. For his PhD, he did work on good ol’ boring 19th Century physics — namely continuum mechanics. But it wasn’t ho-hum work. Born did it with such mathematical rigor and craft that he made good friends that’d greatly help his career, namely mathematicians of the epochal-making caliber of Hilbert, Klein, and Minkowski. Through Minkowski, he quickly developed the nose for the new and was an early worker in Special Relativity. Then, Born was indeed in his early 40s when he worked with Werner Heisenberg and brought his straight-form-the-source Hilbert Space skills to the creation of Matrix Mechanics. And then, of course, once quantum mechanics was formulated, there was tons of low lying fruit, and Born did excellent work in its applications to solid state physics as well as doing magisterial work in the perhaps-boring-but-ever-more-technologically-useful 19th Century physics of his schoolboy youth (e.g., his famous _Principles of Optics_ textbook/reference with Wolf).
Comment #13 February 1st, 2016 at 5:29 pm
fred #11: See my post The Singularity Is Far. Yes, if civilization lasts long enough, unfriendly superintelligent AI taking over might very well eventually become the biggest problem we have to face. But there’s a lot of other problems we’ll need to solve first even in order to survive for long enough to have such a problem.
Comment #14 February 1st, 2016 at 7:47 pm
Here are the full game records, with also some links to commentary: https://www.reddit.com/r/baduk/comments/42yt5i/game_records_of_alphago_vs_fan_hui/
(I don’t play Go myself so don’t ask me about this.)
Comment #15 February 1st, 2016 at 8:04 pm
Hmmmm … here are four transformational STEM advances by workers over 40:
Thermonuclear weapons Stanislaw Ulam and Edward Teller, invented circa 1951, at ages 42 and 43 respectively.
For details, see Randall Munroe’s wonderfully clear Thing Explainer essay “Machine For Burning Cities.”
Magnetic resonance imaging Raymond Damadian in 1972, Paul Lauterbur in 1973, Richard Ernst in 1975, Peter Mansfield in 1978, at ages 36, 44, 42, and 45.
For details, see Randall Munroe’s Thing Explainer essay “Tiny Bags of Water You’re Made Of.”
Single-molecule biomicroscopy Stefen Hell in 1962, Eric Betzig in 2006, and William E. Moerner in 2006, at ages 38, 46, and 53.
For details, see Randall Munroe’s Thing Explainer essays “Picture Taker” and “Colors of Light”
Scanning microscopy Gerd Binning and Heinrich Rohrer, in 1981, at ages 34 and 48.
Hmmmm, wouldn’t scanning microscopes have made a TERRIFIC Thing Explainer essay? 🙂
—-
Conclusion Transformational advances in systems engineering commonly have been simple enough to explain in the plainest of words, yet these “obvious” advances commonly have required decades of experience to conceive in integrated form; perhaps this explains the preponderance of STEM inventors who have been over the age of 40?
Comment #16 February 2nd, 2016 at 4:43 pm
Yitang Zhang was over 50 when he solved the prime pairs problem.
Comment #17 February 2nd, 2016 at 10:33 pm
Hmmm … on the mathematical end, no one has mentioned two very prominent recent claims by over-40 mathematicians:
abc conjecture Shinichi Mochizuki, in 2012, at age 43.
Graph isomorphism in quasi-P László Babai, in 2015, at age 65.
Peer review is still pending in both cases.
Conclusion Just admiration for sustained commitment.
Comment #18 February 3rd, 2016 at 12:36 am
I believe Weierstrass did all his work post 60 or so. Before that he was a school teacher. Legend has it that he later made Ph.D. students stand at the board and rewrite proofs until they had it exactly right.
I think his approach (integral formulas + power series) is what transformed complex analysis into an easy standard tool that underlies almost all of advanced mathematics.
Comment #19 February 3rd, 2016 at 5:33 pm
Another notable triumph for over-40 STEM workers in 2015 was the release by Paul Horowitz and Winfield Hill of the much-expanded and eagerly-awaited third edition of their The Art of Electronics. If quantum supremacy is ever demonstrated experimentally, it almost certainly will be at a lab that has one or more tattered copies of Horowitz and Hill on its bookshelf.
Pretty good for two guys of age 73 and (can this be right?) 89.
The Art of Electronics is a paradigmatic example of a STEM book that, by virtue of its immense scope and integral clarity, can itself be appreciated as a deliberate work of (cognitive) systems engineering. Perhaps in crafting such books, the more decades of experience, the better!
Another such artful book (as it seems to me) is Alexander Grothendieck’s and Jean Dieudonné’s Éléments de Géométrie Algébrique (universally known as EGA). This work was published between 1960 and 1967, when Grothendieck was aged 32-39, and Dieudonné was aged 54-61.
It’s easy to undervalue Dieudonné’s contribution to Grothendieck’s seminal work; survey articles like Dieudonné’s “The Historical Development of Algebraic Geometry” (1972) expose the broad-and-solid foundations of Dieudonné’s cognitive perspective.
Conclusion Perhaps young Shinichi Mochizuki’s great misfortune is that he has not found (yet) his mature Jean Dieudonné, and in consequence Mochizuki’s young ideas do not rest (yet) upon mature foundations of explanation and understanding.
Comment #20 February 4th, 2016 at 8:47 am
Scott #13
Right, it’s possible that humanity will be wiped out before AI is powerful enough to be a threat.
But, on the other hand, we have to recognize that the pace of AI improvement is exponential and we’re finally seeing undeniable breakthroughs.
Those breakthroughs are actually motivated by immediate practical applications ($$$) and aren’t about building AI as a theoretical/academia toy (beating humans at Go is to make a point that there is something new going on here).
http://preview.tinyurl.com/jskm7lx
And this sort of “applied” AI will probably become an essential tool to help us solve all those other issues that threaten us, driving more and more investment, which itself would accelerate AI development even more.
Comment #21 February 4th, 2016 at 12:49 pm
Mathematician (and Fields Medalist) Michael Harris two very recent weblog essays “Beschleunigung, perfectoid or otherwise” (February 3, 2016) and “More thoughts on acceleration” (February 4, 2016) are meditations upon the topic of cognitive acceleration in general.
A survey of recent work motivating Harris’ essays is provided by a recent Scientific American weblog essay by mathematician Evelyn Lamb, titled “Thinking about How and Why We Prove” (January 21, 2016).
Caveat Michael Harris’ weblog Mathematics Without Apologies (and his 2015 book of the same title) asks plenty of tough questions, and presents plenty of challenging observations, without offering any easy answers or reassuring roadmaps. Harris relies upon the comments to his weblog for that. 🙂
Working conclusion The ongoing acceleration of global mathematical culture has produced a surplus of Grothendiecks and a paucity of Dieudonnés; in consequence “accelerated” intelligences already are walking among us; it is natural for STEM workers (young and old alike) to be simultaneously exhilarated and frightened by this reality.
Comment #22 February 4th, 2016 at 3:49 pm
eli #6:
refer to
http://mathoverflow.net/questions/25630/major-mathematical-advances-past-age-fifty
Comment #23 February 4th, 2016 at 3:58 pm
Hilbert began working on Physics in his 50s.
I’ve heard that the reason it appears that people stop doing work after age x is not that they are old, but its that their field has dried up— all of the problems that can be solved have been, leaving only hard ones or ones that aren’t that interesting. So they key may be to change fields.
Comment #24 February 5th, 2016 at 5:48 am
Given that this is Scott’s blog, I’m surprised no one has yet mentioned Feynman and his seminal work on quantum computing, which came after he had turned 60.
Comment #25 February 5th, 2016 at 3:18 pm
what is the polynomial method?
Comment #26 February 5th, 2016 at 4:00 pm
eli #25: Here are PowerPoint slides from a tutorial talk about it that I gave at FOCS’2008.
Comment #27 February 6th, 2016 at 1:00 am
The eminent knot-theorist Joan Birman got her Ph.D after age 40, though she must have started working in knot theory a little before she was 40.
Comment #28 February 6th, 2016 at 10:21 am
[…] Sidles writes on Scott Aaronson’s […]
Comment #29 February 6th, 2016 at 2:09 pm
The following passage augments in bold the Wikipedia article on the mathematician Emma “Emmy” Noether:
How unfortunate it is, that Noether’s tragically early death (at age 53) deprived humanity of a fourth “Emmy Epoch”. As Hermann Weyl said at her funeral “Her heart knew no malice; she did not believe in evil.” … oy … 🙁
Jean Dieudonné’s survey “The Historical Development of Algebraic Geometry” (1972, see comment #19) provides further details, and an extended mathematical context, regarding the transformational impact(s) of Emma Noether’s work (and that of her mathematician-father Max Noether too).
Comment #30 February 9th, 2016 at 7:09 am
John Sidles writes in his fourth (but not last) post “Mathematician (and Fields Medalist) Michael Harris …”. It is definitely false that Michael Harris is a Fields medalist. I suggest posters check these new gadgets, the Internet and Wikipedia, before writing misleading comments.
Comment #31 February 10th, 2016 at 9:15 am
That was an embarrassing goof: I referred to a Clay Research Awardee (Michael Harris, 2007) as a “Fields Medalist.”
For details, interested Shtetl Optimized readers are referred to Chapter 2 “How I Acquired Charisma” of Michael Harris Mathematics Without Apologies (2015).
Its notable (even surprising) that Harris’ book refers nowhere to his own Clay Research Award; Harris’ narrative rather is mainly concerned with (what Harris calls) “the relaxed field” that remains when peripheral considerations of personal honors, national interests, state secrets, and intellectual property are removed from centrality to mathematical practice.
Needless to say, not everyone agrees on the merits of Harris’ view of mathematics as a practice that (ideally) is “not subject to the pressures of material gain and productivity.”
Thank you, Attila Smith, for helping to inspire these corrections, reflections and citations.
Comment #32 February 11th, 2016 at 11:30 pm
Let’s not let these fine Shtetl Optimized comments end without a tribute to yet another remarkable group of over-forty (in fact, over-sixty!) STEAM-workers.
The
Turing TestDEVO Test From ordered foundations in rationality, self-construct/self-deconstruct self-realizing general intelligences. Because we’ll know, for sure, that strong AI has arrived, when its avatars start wearing Energy Domes. 🙂Comment #33 March 23rd, 2016 at 10:10 am
I attended the New England Complex Systems Conferences on complex systems in 2006 (ICCS 2006) where Marvin Minsky was a plenary speaker. I was self-conscious because I requested an overhead projector for my session talk, while other participants (much younger than I) were skilled PowerPoint presenters.
When Minsky kept the audience waiting while an overhead projector was wheeled into place and adjusted, he drily quipped: “I just work with computers; I don’t like them.” A forever endearing comment.