Marvin Minsky, emeritus professor of media arts and sciences at MIT, was one of the original participants in the Dartmouth Summer Research Project on Artificial Intelligence in 1956. He will co-open the 50th anniversary commemorative conference at Dartmouth today. (Courtesy of Coveney/MIT.)
As artificial intelligence research celebrates its 50th birthday, the MIT icon asks what makes the minds of three-year-olds tick.
Top computer scientists from around the world are meeting today at Dartmouth College in Hanover, NH, to mark the 50th anniversary of "artificial intelligence." Back in 1956, John McCarthy, then a member of Dartmouth's mathematics faculty, invented the term for the field's seminal gathering, the Dartmouth Summer Research Project on Artificial Intelligence. McCarthy and four other participants in the 1956 project, including MIT's Marvin Minsky, are participating in this week's meeting, which focuses on AI's next 50 years.
Mathematical and philosophical breakthroughs by Alan Turing, John von Neumann, Herbert Simon, Allen Newell, and other giants of computer science made the 1950s a time of great optimism about machine intelligence. Researchers believed they would soon be able to program computers to simulate many forms of human reasoning. Expert systems would embody and manipulate knowledge in the form of symbolic logic. Artificial neural networks would be trained to evolve toward correct answers.
This optimism even spilled over into popular culture, where HAL, the intelligent (and profoundly disturbed) computer in Stanley Kubrick's 1968 film 2001: A Space Odyssey, upstaged the human actors.
But by the late 1960s it was clear that approximating even child-like human reasoning in a computer would require vastly complex webs of logical equations or neural connections. So researchers retrenched. They began breaking down problems, focusing on replicating simple human feats such as moving children's blocks (the subject of Stanford computer scientist Terry Winograd's now-famous program SHRDLU, which used natural-language instructions to manipulate a robotic arm).
Minsky, who will open the Dartmouth conference with McCarthy, admired Winograd's work. But he's long eschewed reductionistic demonstrations in favor of exploring the real mechanisms behind human thought. Working with Seymour Papert in the MIT AI Lab, for instance, Minsky began in the 1970s to develop the "Society of Mind" theory, which posits that layers of purposeful yet mindless "agents" work together to generate consciousness.
Technology Review interrupted Minsky on July 11, as he was proofing the galleys for his forthcoming book, The Emotion Machine, which reinterprets the human mind as a "cloud of resources," or mini-machines that turn on and off depending on the situation and give rise to our various emotional and mental states.
Technology Review: Can you believe that it's been 50 years since the first Dartmouth AI meeting? Does it feel like five decades have passed?
Marvin Minsky: I haven't experienced many intervals of 50 years, so it's hard for me to say.
TR: Fair enough. So, what are your thoughts about the state of AI research today, compared to where it was in 1956?
MM: What surprises me is how few people have been working on higher-level theories of how thinking works. That's been a big disappointment. I'm just publishing a big new book on what we should be thinking about: How does a three- or four-year-old do the common-sense reasoning that they're so good at and that no machine seems to be able to do? The main difference being that if you are having trouble understanding something, you usually think, "What's wrong with me?" or "What's wasting my time?" or "Why isn't this way of thinking working? Is there some other way of thinking that might be better?"
But the kinds of AI projects that have been happening for the last 30 or 40 years have had almost no reflective thinking at all. It's all reacting to a situation and collecting statistics. We organized a conference about common-sense thinking about three years ago and we were only able to find about a dozen researchers in the whole world who were interested in that.
A massive AI project called CALO could revolutionize machine learning.
Guest (Steve Rose (Maui))
Congratulations to Dr. Minsky on 50 years of successful promotion of himself and his work.
It's too bad that his book "Perceptrons", a term coined by Frank Rosenblatt, put an end for twenty years to the neural modeling approach to understanding intellgence, and may have contributed to Rosenblatt's solo boating accident on Cayuga Lake, especially since the book was a smear designed to divert funding to Minsky's "black box" approach to AI (and was fabulously successful in doing so).
The book basically discredited Rosenblatt's work by proving that a two layer Perceptron couldn't separate figure from ground, so Perceptrons weren't worthy of further investigation. Rosenblatt was aware of the simplicity of the two layer approach he was using, and did so so that we could begin to understand neural nets without the complexity of hidden layers (which could be explored later, along with other learning algorithms).
(continued)
Re: Congratulations to Dr. Minsky
Steve said:
[The book Perceptrons by Minsky and Papert] basically discredited Rosenblatt's work by proving that a two layer Perceptron couldn't separate figure from ground, so Perceptrons weren't worthy of further investigation. Rosenblatt was aware of the simplicity of the two layer approach he was using, and did so so that we could begin to understand neural nets without the complexity of hidden layers (which could be explored later, along with other learning algorithms).
I presume that you are repeating a rumor, and did not actually read the book. So far as I know, no addition of "hidden" layers will help to enable a loop-free (non-recursive) neural network to recognize or separate connected images on a retina. And because of this limitation of the networks themselves, no improvement in learning procedures will help.
Indeed, loopfree neural networks can do many useful things indeed, but no matter how many layers they have, there is no reason (or evidence) that they can surmount the topological limitations discussed in the book.
I should add that while there have been many statement that try to discredit the book, it is very strange that there has been virtually no further research to show that the same limitations do not apply to networks with multiple layers -- unless one allows unlimited numbers of inputs to each neuron. (Of course, in that case, a 2-layer network can compute any Boolean function by expressing it in disjunctive normal form.
Perhaps we did make a mistake by emphasizing the problem of computing parity, because this particular function can be computed in logarithmically few layers. However, if one cask about the "majority" function instead, I suspect that the problem remains quite difficult, and I'd like to see some of those critics demonstrate how large a network would be required to recognize whether an input contains more zeros than ones.
No such problems have been solve the kinds of invective that we see in this letter.
Guest (Steve Rose (Maui))
However, Roseblatt had naively allowed the popular press to play up his work, which grated on Minsky. From Wikipedia (article: Frank Rosenblatt): "For years Minsky crusaded against Rosenblatt on a very nasty and personal level, including contacting every group who funded Rosenblatt's research to denounce him as a charlatan....". It is a shame that Minsky's work turned up so many dead ends, and ironic that one of the primary researchers who have begun to really understand how the brain works is Jeff Hawkins ("On Intelligence"), for whom brain function was a hobby until recently. Also ironic is Minsky's complaint of an absence of people working on higher level theories. It would have been interesting to see the progress of Rosenblatt's work over the last 40 years had Minsky not hounded him.
Guest (Joe Reuter)
Delayed recognition is often the reward for Genius
http://www.ieee.org/portal/pages/about/awards/sums/rosensum.html
Guest (Steve Rose)
Guest (John LaMuth)
Guest (Len Bullard)
Why Do Children Learn So Fast?
A question that I don't see answered is how can children learn that much that fast? The scaling problem of AI is centered in how situation semantics (for lack of a better term) are related so quickly with the number of observations made.
Guest (Paul)
Its biological. Stimulus causes new neurons to develop in the brain, creating new memory pathways and learning. Computers don't develop biologically, so it is harder for them to "learn" from experience and data input.
common knowledge memories as a data base design issue.
Yes, but as a statistical association matrix.
all data is a pathway / index to other data. Ideas are built up as a result of the path you follow. Each branch in the path builds up parts of the data 'solution'. Think of the most general concepts as being closer to the root of the decision tree, a very specific idea / object a leaf. See the following:
The memory code.
By Joe Z. Tsien
Scientific American, July 2007 (vol 297-1, pg 52)
Researchers are closing in on the rules that the brain uses to lay down memories. Discovery of this memory code could lead to the design of smarter computers and Robots and even to new ways to peer into the mind.
Dr. Tsien is noteworthy for creating a new strain of lab mice ‘Doogie’(s) that have enhanced memory abilities.
Guest (sholliman)
I think that it's more like there's nothing very "common" about common sense. I expected to see expert systems and applications coming from AI expermentation. Something more practical like the space race or the human genome project. What I see are the fear factor driving us away from serious advances in this field, that could have some real help for humans, towards a machine with emotions that people are afraid of. I want a tool that I can use to build me a solar generator and help me plan the future. If you think in terms of modularized units that can work together or separately on smaller units of work, then, hey, it's that alot like us?
Guest (Glen)
I noticed that Mr. Minsky referrs to the OpenMind project as if it is ongoing, but the last several times I've checked www.openmind.org, it was obviously unmaintained and broken. Anyone know whatever became of OpenMind?
Guest (me)
I think the driving force behind the project (push singh) has died. I hope this doesnt mean the end of the whole thing.
Guest (G Roper)
"Common sense" like "intelligence"...
Neither are well-defined. It's always best to define a topic clearly in operational terms before doing research. Good Old-Fashioned AI (GOFAI)never did that. The reason was that, GOFAI was the wrong approach. So-called Nouvelle AI is now finding the answer.
Unfortunately Minsky _built_ the GOFAI road. His legacy will be "A bright guy, with lots of ideas, who led us astray for 20 years."
Guest (G Roper)
Nouvelle AI research, initiated by Hans Moravec and Rodney Brooks, is yielding AI systems that work in the real world.
And the astonishing experimental work of Luc Steels at the Paris Sony Labs demonstrating how languages are created by communicating robots deserves a Nobel prize.
Guest (Melchor batista)
Most of what is being discussed here was discussed in great depth in one book: "Workings of Mind" (neural pathways) and popularized in two other: "Lateral Thinking" (different perspectives)and "Six Thinking Hats" (different modes of thinking, including emotions) by Dr. Edward de Bono back in the late 60's and early 70's.
Manufacturing in the United States is in trouble. That's bad news not just for the country's economy but for the future of innovation.
This document is part of the “How-To Guide for Most Common Measurements” centralized resource portal. This tutorial provides a detailed guide for measurement and device considerations to take temperature measurements using thermocouples. Get an introduction to thermocouples, which are inexpensive sensing devices widely used with PC-based data acquisition systems. Also review some specific thermocouple examples and learn how thermocouples work and ways to integrate them into a data acquisition measurement system.
View full PDF >
Listen to story > 
Our list of the 50 most innovative companies, including the following:
On Twitter
Become a Fan on Facebook
Subscribe to the Feed
Guest (mwhitlock)
Wouldn't it take 3-4 years to start...
Has anybody noticed that it takes 3-4 years of learning and experience for a 3-4 year old child to think like a 3-4 year old. And you act surprised that this process can't be replicated in a shorter time span....hhhmmm.
Reply
cassiuszedaker
1 Comment
Re: Wouldn't it take 3-4 years to start...
AI research itself lacks common sense. Have AI researchers even defined AI? The Turing Test is not a definition, merely a scientific toy.
Reply
santi.ontanon
2 Comments
Re: Wouldn't it take 3-4 years to start...
You could say the same about biology. There is no good definition of "life", and that doesn't stop them. The lack of a definition for AI, is simply the lack of a definition for "intelligence", which is not just a problem of the AI community, but of many others. But anyway, that is not a problem for advancing the research. As long as there are open problems and questions (such as what is intelligence), there can be sound research
Reply
nkryten
2 Comments
Re: Wouldn't it take 3-4 years to start...
This was my first thought at reading the article, why start at a 3 year olds level? Surely if we want to replicate the development process of a human we must begin by looking at what abilities a new born has and trying to replicate that in machine form.
Reply
santi.ontanon
2 Comments
Re: Wouldn't it take 3-4 years to start...
That's like trying to fly like a bird. Imitating the human process of learning is one way. But it does not have to be necessarily the only way...
Reply