AI & ML

How Accurate Was Marvin Minsky in His AI Predictions?

Executive Summary

  • Marvin Minsky is the most well known AI researcher and proponent.
  • How accurate was Minsky in his predictions?

Introduction

Marvin Minsky’s footprint in AI is quite large.

  • Marvin Minsky wrote eight books on AI or AI related topics.
  • He also started the AI lab at MIT.
  • He was one of the primary consultants for Stanley Kubrick’s movie 2001: A Space Odyssey, which more than any other film informed our understanding of what the future of AI would or could be. 

Marvin Minsky is very well known (he is now deceased), there is no doubt about it. He is probably the first person people think of when they think of AI. 

Let us review a sampling of some of the things written about him.

Polymathic in capacity, protean in vision, Minksy spent his life figuring out how to make machines that are intelligent, “whatever that means,” as he liked to say. He developed AI’s two principle schools of thought: the “symbolic school” of abstract manipulations and the “connectionist school” of unstructured self-organization. He built the first learning machine based on neural networks, simulating how the brain works through practice and trial and error, a progenitor of today’s “deep learning.” Smart people said Marvin was the “smartest person I ever met.” Those closest to Marvin said he was “childlike.” This, they meant as a supreme compliment, because he was ever eager to explore new ideas and never too proud to jump into the middle of things. Minsky won the A.M. Turing Award, the most prestigious award in computer science. – Space.com

And from the MIT AI/Media Lab

Marvin Minsky has made many contributions to AI, cognitive psychology, mathematics, computational linguistics, robotics, and optics. In recent years he has worked chiefly on imparting to machines the human capacity for commonsense reasoning. His conception of human intellectual structure and function is presented in two books: The Emotion Machine and The Society of Mind (which is also the title of the course he teaches at MIT).

And from MIT Technology Review

Inspired by mathematical work on logic and computation, Minsky believed that the human mind was fundamentally no different than a computer, and he chose to focus on engineering intelligent machines, first at Lincoln Lab, and then later as a professor at MIT, where he cofounded the Artificial Intelligence Lab in 1959 with another pioneer of the field, John McCarthy.

Minsky’s early achievements include building robotic arms and grippers, computer vision systems, and the first electronic learning system, a device, which he called Snarc, that simulated the functioning of a simple neural network fed visual stimuli. Remarkably, while at Harvard in 1956, he also invented the confocal scanning microscope, an instrument that is still widely used today in medical and scientific research.

Minsky was also central to a split in AI that is still highly relevant. In 1969, together with Seymour Papert, an expert on learning, Minsky wrote a book called Perceptrons, which pointed to key problems with nascent neural networks. The book has been blamed for directing research away from this area of research for many years.

Today, the shift away from neural networks may seem like a mistake, since advanced neural networks, known as deep learning systems, have proven incredibly useful for all sorts of tasks.

In fact, the picture is a little more complicated. Perceptrons highlighted important problems that needed to be overcome in order to make neural networks more useful and powerful; Minsky often argued that a purely “connectionist” neural network-focused approach would never be sufficient to imbue machines with genuine intelligence. Indeed, many modern-day AI researchers, including those who have pioneered work in deep learning, are increasingly embracing this same vision.

Prediction #1: Human Brains Work like Computers?

MIT Technology Review states the following.

Minsky believed that the human mind was fundamentally no different than a computer

However, as we covered up to this point already, this assumption was proven to be false.

Curiously, MIT Technology Review presents a false belief as something positive. Dreyfus had already proposed indeed before 1972, but at least by 1972 that Minsky’s presentation of the analogy between biological brains and how computers come to conclusions was incorrect. This same statement was made in the following quotation, without any observation that the assertion is false.

Marvin Minsky is the leading light of AI — that is, artificial intelligence. He sees the brain as a myriad of structures. Scientists who, like Minsky, take the strong AI view believe that a computer model of the brain will be able to explain what we know of the brain’s cognitive abilities. – Edge

And this quote.

Minsky, who helped pioneer the field of artificial intelligence during more than 50 years at MIT, was fascinated by the way minds worked. In opposition to the millennia-old philosophical tradition that conceives of mind and body as irreducibly separate, he viewed human intelligence as the result of the brain’s numerous, distinct computational processes. – Princeton.edu

These quotes are not from 1967, when the debate was raging if this was true. The first quote is from 2016 and the second from 2017 (when after Minsky’s death several articles were published about him).

So here is a logical question. Why are we still lauding Minsky for holding a view of AI that was proven incorrect decades ago? 

Whitewashing History…From MIT Technology Review?

The area that MIT Technology Review states is..

“a little more complicated”

..is inaccurate.

Prediction #2: Perceptrons (The Precursor or Neural Networks) Would Not Amount to Anything?

Minsky argued against neural networks, and these are today the most successful area of AI. Furthermore, Minky’s book Perceptrons argued with little evidence that neural networks would not be useful. 

This is explained in the following quotation.

..in 1969 Marvin Minsky and Seymour Papert published the book “Perceptrons” with a mathematical proof about the limitations of two-layer feed-forward perceptrons as well as unproven claims about the difficulty of training multi-layer perceptrons. The book’s only proven result—that linear functions cannot model non-linear ones—was trivial but the book had nevertheless a pronounced effect on research funding and, consequently, the community. – Wikipedia

Therefore, Marvin Minsky and his writing partner make claims about neural networks that ended up being incorrect. They were mistaken when the book was published, because people working in neural networks at the time had already developed backward propagation or backpropagation, which addressed the criticisms in Minsky and Papert’s book. Backpropagation is explained in the following quotation. 

Backpropagation is a method to adjust the connection weights to compensate for each error found during learning. The error amount is effectively divided among the connections. Technically, backprop calculates the gradient (the derivative) of the cost function associated with a given state with respect to the weights. The weight updates can be done via stochastic gradient descent or other methods, such as Extreme Learning Machines,[48] “No-prop” networks,[49] training without backtracking,[50] “weightless” networks,[51][52] and non-connectionist neural networks.

This is essentially “quality control” for the weights that drive the neural network. 

This was a debate about computing as explained in the following quotation. 

“Of course, all of these limitations kind of disappear if you take machinery that is a little more complicated — like, two layers,” Poggio says. But at the time, the book had a chilling effect on neural-net research.

“You have to put these things in historical context,” Poggio says. “They were arguing for programming — for languages like Lisp. Not many years before, people were still using analog computers. It was not clear at all at the time that programming was the way to go. I think they went a little bit overboard, but as usual, it’s not black and white. If you think of this as this competition between analog computing and digital computing, they fought for what at the time was the right thing.” – MIT Technology Review 

Where Did Marvin Minsky Get the Hypothesis of Computers Work like Human Brains?

Curiously, Marvin Minsky obtained the idea that the human brain worked like a computer from an interesting source. 

The neural nets described by McCullough and Pitts in 1944 had thresholds and weights, but they weren’t arranged into layers, and the researchers didn’t specify any training mechanism. What McCullough and Pitts showed was that a neural net could, in principle, compute any function that a digital computer could. The result was more neuroscience than computer science: The point was to suggest that the human brain could be thought of as a computing device. – MIT Technology Review 

Frank Rosenblatt, the inventor of perceptrons had been debating against Minsky on the topic of perceptrons throughout the 1960s. Still, unfortunately, Rosenblatt died in a boating accident two years after Minsky and Papert’s Perceptrons (1969) was published.

Rosenblatt’s perceptron was the first “mini” neural network, as explained in the following quotation.

The first trainable neural network, the Perceptron, was demonstrated by the Cornell University psychologist Frank Rosenblatt in 1957. The Perceptron’s design was much like that of the modern neural net, except that it had only one layer with adjustable weights and thresholds, sandwiched between input and output layers. MIT Technology Review

Who Was More Accurate on Perceptrons, Minsky or Rosenblatt?

Today, we hear about Rosenblatt far less than we do about Minsky. This is explained in the following quotation.

After research on neural networks returned to the mainstream in the 1980s, new researchers started to study Rosenblatt’s work again. This new wave of study on neural networks is interpreted by some researchers as being a contradiction of hypotheses presented in the book Perceptrons, and a confirmation of Rosenblatt’s expectations. – Wikipedia

And here is a critique of Minsky and Papert’s book.

H.D. Block expressed concern at the authors’ narrow definition of perceptrons. He argued that they “study a severely limited class of machines from a viewpoint quite alien to Rosenblatt’s”, and thus the title of the book was “seriously misleading”.[9] Contemporary neural net researchers shared some of these objections: Bernard Widrow complained that the authors had defined perceptrons too narrowly, but also said that Minsky and Papert’s proofs were “pretty much irrelevant”, coming a full decade after Rosenblatt’s perceptron.

With the revival of connectionism in the late 80s, PDP researcher David Rumelhart and his colleagues returned to Perceptrons. In a 1986 report, they claimed to have overcome the problems presented by Minsky and Papert, and that “their pessimism about learning in multilayer machines was misplaced” – Wikipedia

Marvin Minsky is strongly associated with MIT’s AI/Media Lab, and they are not going to want to admit that Minsky was incorrect. Therefore it appears they need to propose that was is obviously true, is not true, and that it is “more complicated.”  

On Minsky’s Wikipedia page, the only mention of controversy is that he received funding from noted sex offender and trafficker Jeffery Epstein. The following describes testimony from Virginia Guiffre on how Marvin Minsky was one of the many high-status people that Jeffery Epstein directed her to have sex. 

A young woman who says financier Jeffrey Epstein and socialite Ghislaine Maxwell kept her as a sex slave also accused a host of high-powered men of being involved in Epstein’s alleged sex-trafficking ring, according to court records unsealed Friday. Virginia Giuffre, who says that Epstein and Maxwell trafficked her to powerful people for erotic massages and sex, claimed in depositions in 2016 that Maxwell directed her to have sex with former New Mexico Gov. Bill Richardson, Britain’s Prince Andrew (whom she has accused before), wealthy financier Glenn Dubin, former senator George Mitchell, now-deceased MIT scientist Marvin Minsky, and modeling agent Jean-Luc Brunel, as well as “another prince,” a “foreign president,” a well-known prime minister” and the owner of a “large hotel chain” in France. “There was, you know, another foreign president, I can’t remember his name. He was Spanish. There’s a whole bunch of them that I just—it’s hard for me to remember all of them. you know, I was told to do something by these people constantly, told to—my whole life revolved around just pleasing these men and keeping Ghislaine and Jeffrey happy. Their whole entire lives revolved around sex.” The late MIT cognitive scientist and “father of AI” Marvin Minsky was part of a glittering set of academics that Epstein collected as friends. In 2012, as Epstein was trying to rehabilitate his image, he touted his foundation’s sponsorship of an academic conference organized by Minsky. During her first alleged rendezvous with the prince, Giuffre claimed that Maxwell “asked Andrew how old he thought I was and he guessed 17 and they all kind of laughed about it and Ghislaine made a joke that I was getting too old for Jeffrey. She said, ‘He’ll soon have to trade her in.’” Giuffre also said Andrew participated in an orgy with underage girls on Epstein’s so-called “pedophile island.” (Andrew and Buckingham palace deny these claims.) A 2001 picture of Andrew, with his arm around Giuffre’s waist, surfaced alongside her bombshell story a few years ago. In 2007, another young woman named Johanna Sjoberg told The Daily Mail that Andrew had groped her at a party at Epstein’s mansion. – The Daily Beast

I find several things about this odd. First, in researching for this article, it was the first I had ever heard of this.

But secondly, how can this be the only thing related to Marvin Minsky that is considered controversial? Minsky’s predictions should be considered extremely controversial — primarily because it is difficult to find one where he ended up being correct. 

Watch out according to Minsky, because HAL was right around the corner. 

Prediction #3: Minsky’s All Powerful Conscious AI Robot

The AI robot that takes over the world and subjugates the tiny intellect humans has been the plot of some of the most popular movies. This includes all of the movies in the Terminator franchise, Eagle Eye, Echelon Conspiracy, and a host of other movies. And this concept is endorsed by Minsky, as not only 2001, but the following quotation describes. 

“Once the computers got control, we might never get it back,” Minsky told Life magazine in 1970. “If we’re lucky, they might decide to keep us as pets.” – Princeton.edu

Did Princeton notice that 50 years later we do not have generalized intelligence? 

Minsky’s brilliance — in 1969, he received the Turing Award, the Nobel Prize of computer science — could be intimidating. “He was always ahead of you in a conversation and guessing what you were going to say,” says MIT professor Patrick Henry Winston, a Minsky student who followed him as head of the AI Lab. “And often what you were going to say wasn’t as good as what he’d predicted.” – Princeton.edu

The Nobel Society must not incorporate accuracy into their prize awarding criteria.

Notice the last part of the quote, that what you were going to say wasn’t as good as what he’d predicted. When did Minsky predict anything about AI that turned out to be true? 

The Princeton article continues. 

The intelligent machines Minsky envisioned still lie over the horizon. “The actual computer programs that emerged from MIT were typically things that, given the technology available at the time, could not really do what AI proponents saw in their mind’s eye,” says mathematician Martin Davis *50, who met Minsky while both were students at New York’s Bronx High School of Science. “He was more like Moses — allowed to look into the Promised Land but not actually to enter it.” 

No, that is incorrect.

This has been the long term claims by AI proponents. Still, current AI is not developing along the lines proposed by Minsky, the most fertile area is neural networks which weak AI proposed by his nemesis Rosenblatt whose approaches Minsky and Papert dismissed in their book Perceptrons. 

Prediction #4: Coding Will Become Irrelevant?

The following quote relates to how Minsky viewed the future of coding.

Quoting from a book entitled Recoding Gender: Women’s Changing Participation in Computing, Minsky said: “Surely the days of programming, as we know it, are numbered … Instead, we’ll express our intention about what should be done … Then these expressions will be submitted to immense, intelligent, intention-understanding programs, which will themselves construct the actual, new programs.”

While programmers are still very much in demand in 2016, and it has become increasingly more common among younger generations to be knowledgeable of the various languages, there’s something to be said for the various ‘click-and-drop’ programs out there for creating codeless projects. Who knows, maybe one day he’ll be right? – Silicon Republic

This quotation is useful for two reasons. First, this prediction is as I write this 37 years old, and coding is more popular than ever. The opposite of what Minsky predicted happened.

However, the author at Silicon Republic does not count this against Minsky. Rather he says it may be proven true in the future.

Well, Minsky did not say that in 2080 or 2200 that coding would go away. He made this prediction in 1984 and he stated that coding would be going away, that “its days are numbered.” 

1984 is forty years before 2016 when the quotation from the Silicon Republic was written. Why did the author not call out this prediction as not coming true, rather than project a possible future time, far outside of Minsky’s own prediction timeline that it might come true?

This is a constant feature that I observed of the reporting on Minsky.

If something he says turns out to be the exact opposite of what happened, the author proposes that it may yet occur in sometime in the future. There is no measurement for forecast accuracy. 

Prediction #5: Human Will Be Able to Communicate With Aliens?

If aliens are out there, we’ll be able to communicate with them

Once again looking at the intelligence of a species, Minsky contributed to a collection of essays from established scientists which posed the question of the possibility of extraterrestrial life in the universe and what would happen if we ever met it?

In Minsky’s case, he discussed the possibility of whether we could communicate with ‘aliens’ or not.

In his view, if we are to ever make contact, we will be able to communicate with them because we would share similar constraints, such as any species’ need to create symbols to represent ideas, like economics, to other beings on their planet.

However, this would only apply to physical beings, like us, that still rely on survival and communication in the same way we do.

Minsky was instrumental in the decision to send mathematical algorithms into outer space, based off this theory. – Silicon Republic

How would Minsky know this exactly?

This demonstrates a pattern on the part of Minsky to use his celebrity status to opine on things that aren’t in his wheelhouse. 

Here is something Minsky should have picked up on working out of a university and having the opportunity to ask questions of professors in the astronomy departmen. The reality is any life that does exist in the universe will be so distant from the Earth that it will be impossible to communicate with them even radio waves. 

Prediction #6: Neuroscience Was on to Nothing?

Okay, so maybe he wasn’t right all the time. As recently as 2007, Minsky was interviewed about the potential for neuroscience, something which it becomes clear he was not a fan of.

“When you talk to neuroscientists, they seem so unsophisticated; they major in biology and know about potassium and calcium channels, but they don’t have sophisticated psychological ideas,” Minsky said in the interview. “Neuroscientists should be asking: What phenomenon should I try to explain? Can I make a theory of it? Then, can I design an experiment to see if one of those theories is better than the others? If you don’t have two theories, then you can’t do an experiment. And they usually don’t even have one.”

He went on to say that AI and mathematics were the obvious ways in which to solve the answers of the human mind.

“[You can answer questions] through the lens of building a simulation,” he said. “If a theory is very simple, you can use mathematics to predict what it’ll do. If it’s very complicated, you have to do a simulation. It seems to me that for anything as complicated as the mind or brain, the only way to test a theory is to simulate it and see what it does.” – Silicon Republic

The premise of the introduction by the author of this article about Minsky’s prediction accuracy is reversed. Minsky has proven to be wrong about nearly all of his (major at least) predictions. Therefore, the better introduction to this quote would be..

“this is another example of Minsky being wrong.” 

This topic is back to the issue that Minsky thought the brain worked like a computer, and he was proven wrong. Neuroscience does explain how the brain works, AI does not do this. 

Prediction #7: AI Will be Solved Before 1980

Speaking in 1967: “Within a generation … the problem of creating artificial intelligence will substantially be solved. Oh.” – Silicon Republic

Where is the rest of this article from Silicon Republic?

This is how the article ends.

It is as if the article author found a prediction that was so inaccurate that they just stopped writing the rest of the article. Every one of the predictions listed by this article of Minsky’s either did not occur or played out the exact opposite of what Minsky said they would. 

Robert Stone makes the following observation about Minsky’s predictions from the book Society of the Mind. 

I respect Minsky, and this is not meant as an attack on him. He tried to be wide-ranging and explorative in the AI domain and I laud that.

But like other pioneers, he could be wrong.

In Society of Mind he discussed emotion, but he was not even close to the right model. (Just to back this up, you need to know that the correct model of emotion is that each different emotion is a metric for the success or failure of a particular class of goal types. From that premise, everything else about emotion is deducible and also cleanly mechanizable.)

And, in his book The Emotion Machine again he explored but didn’t quite hit the real core principle.

But anyway R.I.P. Minsky, pioneer.

So we have established that Minsky could be wrong — but what is a question is could Minky be right. Commenters seem to assume that Minsky was proven right in some area, and they are only finding an area where he was wrong. 

To see the full screen just select the lower right-hand corner and expand. Trust us, expanding makes the experience a whole lot more fun

Conclusion

This has to be one of the worst prediction histories of anyone who is considered not only very prominent in their field but is the best-known person in his field. If Minsky was so knowledgable about AI, why was he so accurate? Secondly, why is there reticence to measure Minsky’s accuracy? Minsky had the right to make predictions and to be wrong. But the interpretation of Minsky as a visionary in AI does not hold up. Visionaries are at least somewhat right about their visions and predictions.

It is extremely difficult to see how Marvin Minsky is deserving of the place he holds in the history of AI. He was wrong about the similarity between the brain and how software can be developed to exhibit intelligence. He was wrong about the limitations of neural nets, he so poorly advised Stanley Kubrick, that he had Kubrick create an AI in the form of HAL that 19 years after 2001, we don’t have anything close to. And there is no pathway to achieve this generalized intelligence.

Secondly, it is not my view that Minsky was simply incorrect or in error. Minsky showed a disregard for what was true, and appeared to be more interested in promoting AI, and his career, and his approach to AI, than in limiting his comments to things that he could prove. 

I found it very difficult to find anything written that was critical of Minsky. Many discussions around Minsky seem to have more to do with what a colourful person he was, how playful he was, or his penchant for Hawaiian shirts, rather than whether what he said would happen even came true.

The following is a typical quotation of those that interacted with Minsky.

During our talks, Minsky proved to be a fascinating conversationalist, with an engaging sense of humour and a luminous smile. He has one of the clearest minds I have ever encountered, and he is capable of elucidating the most complicated ideas in simple language—something that is possible only if one has a total mastery of the ideas. – The New Yorker

The numerous articles mentioning Minsky’s personality traits indicate that we are still estimating people to a great degree by how they make us feel about ourselves then whether what they are proposing is true.

MIT is itself presenting an overly enhanced presentation of his contributions to AI because he worked at MIT. Minsky was a distant second to Dreyfus when it comes to predictive ability. But it is not only Dreyfus, but it is also Rosenblatt and many others who register a far smaller footprint in the history of AI than Minksy.

And when Dreyfus died, the coverage he received was nothing like the effusive praise given to Minsky, and in fact, we don’t even know if he liked Hawaiian shirts. 

The problem of Marvin Minsky, which is exaggerated claims about AI, has persisted through the history of AI, and they persist to this day.

References

https://www.theverge.com/2019/8/9/20798900/marvin-minsky-jeffrey-epstein-sex-trafficking-island-court-records-unsealed

https://web.media.mit.edu/~minsky/

https://en.wikipedia.org/wiki/Marvin_Minsky

https://en.wikipedia.org/wiki/Artificial_neural_network

https://www.scribd.com/article/401160340/Stop-Saying-exponential-Sincerely-A-Math-Nerd

https://www.theguardian.com/technology/2016/feb/03/marvin-minsky-obituary

https://en.wikipedia.org/wiki/Frank_Rosenblatt

https://www.vanityfair.com/news/2017/03/elon-musk-billion-dollar-crusade-to-stop-ai-space-x

https://medium.com/@windycityn0velist/musk-misses-the-stories-you-dont-hear-about-tesla-anymore-8dfce1d803d2

https://www.ccn.com/elon-musks-backtrack-tesla-robotaxi-claim-securities-fraud/

https://www.latimes.com/business/autos/la-fi-hy-tesla-musk-annual-meeting-shareholders-20190611-story.html

https://medium.com/@hypergiant/is-neural-network-hype-killing-machine-learning-120041406f1

https://www.kdnuggets.com/2017/08/37-reasons-neural-network-not-working.html

*https://www.analyticsinsight.net/neural-networks-not-answer-everything/

http://neuralnetworksanddeeplearning.com/chap5.html

https://builtin.com/data-science/disadvantages-neural-networks

https://towardsdatascience.com/ants-and-the-problems-with-neural-networks-778caa73f77b

https://towardsdatascience.com/neural-networks-problems-solutions-fa86e2da3b22

https://en.wikipedia.org/wiki/Backpropagationhttps://www.youtube.com/watch?v=t81HiFOqbAs

https://www.technologyreview.com/2016/01/26/163622/what-marvin-minsky-still-means-for-ai/

https://en.wikipedia.org/wiki/Computing_Machinery_and_Intelligencehttps://www.the-scientist.com/magazine-issue/artificial-intelligence-versus-neural-networks-65802

https://www.youtube.com/watch?v=aircAruvnK

khttps://www.youtube.com/watch?v=S3Y-TeLKMP8

https://www.bbc.co.uk/teach/ai-15-key-moments-in-the-story-of-artificial-intelligence/zh77cqt

https://www.forbes.com/sites/cognitiveworld/2019/10/20/are-we-heading-for-another-ai-winter-soon/#3fb8536256d6

https://en.wikipedia.org/wiki/Lighthill_report

https://www.newscientist.com/article/dn16306-us-investigation-into-gravity-weapons-nonsense/

https://towardsdatascience.com/probability-of-an-approaching-ai-winter-c2d818fb338a

http://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/p004.htm

https://www.theguardian.com/technology/2018/jul/25/ai-artificial-intelligence-social-media-bots-wronghttps://www.kurzweilai.net/

*https://www.siliconrepublic.com/machines/marvin-minsky-ai-predictions

https://en.wikipedia.org/wiki/Perceptrons_(book)https://www.newyorker.com/news/news-desk/is-deep-learning-a-revolution-in-artificial-intelligence

https://www.edge.org/conversation/marvin_minsky-remembering-minsky

https://paw.princeton.edu/article/lives-marvin-minsky-54

https://www.quora.com/Whats-Marvin-Minskys-view-on-deep-learning

https://www.quora.com/What-was-Marvin-Minsky-wrong-about

https://www.newyorker.com/magazine/1981/12/14/a-i

https://www.quora.com/Why-does-Marvin-Minsky-hate-Noam-Chomskys-linguistic-theories