Banner
    Kurzweil's "How to Create a Mind" is Severely Lacking
    By Samuel Kenyon | November 25th 2012 04:16 PM | 15 comments | Print | E-mail | Track Comments
    About Samuel

    Software engineer, AI researcher, interaction designer (IxD), actor, writer, atheist transhumanist. My blog will attempt to synthesize concepts...

    View Samuel's Profile
    A day after its release, I strolled into the Harvard Co-Op bookstore and read most of Ray Kurzweil's new book, How to Create a Mind: The Secret of Human Thought Revealed. I then quietly placed it back on the shelf and wandered out of the store in a state of wonder. Wondering, that is, if Kurzweil was going broke and this was a last ditch attempt to make some cash.



    Don't get me wrong. I like Ray Kurzweil. I have met him and I have attended many of his public presentations. He used to be a decent writer (last century). And I am a transhumanist, although not a Singularitarian.

    And one of the advanced praisers of Kurzweil's new book, Marvin Minsky (also mentioned in the book), has had some influence on me via his Society of Mind / Emotion Machine class that I took and reading his books. And I've met, worked for the company of, and volunteered for the educational program created by another advanced praiser, Dean Kamen.

    So, with a quantum of respect and the intention of feedback for future improvement, I declare the book How to Create a Mind to be a needless trite pile of vomit.

    What is the Point of this Book?


    The point of this book is apparently for Kurzweil to establish himself as an artificial intelligence expert. Perhaps a second, more admirable, goal is to inspire the accelerating growth of intelligent technology that he predicts.

    He attempts to achieve these goals by explaining his One True Theory of Intelligence, and trying to convince us that his leadership of pattern recognition teams in the past somehow makes him the expert on Hidden Markov Model pattern recognition, which in turn just happens to be the key to the One True Theory of Intelligence.

    Oh, and what is the overall theory? It's basically the same as Jeff Hawkin's approach of Hierarchical Temporal Memory (HTM). Like Hawkins, Kurweil also claims this is a biomimetic approach, focusing only on the neocortex, and within the neocortex only on cortical columns.

    Instead of merely writing his Hawkins-esque theory in a twenty page paper, he has graced the world with a full book by filling in the theory with a bunch of autobiographical stories, juxtapositions of computer science successes like IBM's Watson, and stuff from his previous books.

    Interlude: A Twitter Conversation




    Conclusion


    I wonder if people who buy this book, and also the HTM cult members, have actually thought from a more systematic point of view.

    I suspect people are going with a fictional concept of a "brain in a vat" or the disembodied big computer that somehow can think exactly like a human.

    But HTM and Kurzweil's new book are really very narrow approaches and don't say anything about how to actually make a mind, which is to say how to make a system. And what, exactly, does that system do? What range of environments and interactions is a given system capable of?

    Comments

    vongehr
    Yes, Kurzweil has long since given up to make sense and now joined, like so many, the class of surviving writers exploiting a found market niche, and I recommend you for having the guts to speak up against this practice, as it corrodes the public discourse into a deceive-the-stupid-consumers game that crushes anything insightful under the sheer volume of rubbish.

    I got an invitation to join the "scientific advisory board" of a certain foundation, where you can see literally hundreds of names of desperate people, basically using such platforms (similar to IEET, FQXi, Lifeboat Foundation, ..., and partially Science2.0 sadly) as nothing but a dump to self promote, letting themselves be used as justification for whatever a small team of insiders puts out as consensus (you know - 'we have a scientific advisory board with some famous names' kind of deal) without of course such a large board making any sense except for that the small in-crowd can find any "advice" it wants to hear among it. I told them politely to go and ... themselves, and skimming the names, found of course all the usual suspects like David Brin, Andrea Kuszewski, M. Treder, M. Shermer, Samuel Kenyon, Kurzweil, Bostrom and Terrile next to completely idiotic pseudo scientists like Jack Sarfatti on it. Those people start out promising enough, but then the fame and all that is like crack, and the decay rapid, and them even joining such boards for example, tells all: One cannot take them seriously because they sold-out, and sold out way too cheaply.
    SynapticNulship
    I'm not sure if I've ever been referred to as a "usual suspect" before.

    After posting this I found this pretty good and honest review of Kurzweil's book by Gary Marcus:
    http://www.newyorker.com/online/blogs/books/2012/11/ray-kurzweils-dubious-new-theory-of-mind.html

    It goes into some more detail on the failure of this book to do anything its title promises:

    At the end Kurzweil leaves us with a theory that is generic. Almost anything any creature does could at some level be seen as hierarchical-pattern recognition; that’s why the idea has been around since the late nineteen-fifties.

    What Kurzweil doesn’t seem to realize is that a whole slew of machines have been programmed to be hierarchical-pattern recognizers, and none of them works all that well, save for very narrow domains like postal computers that recognize digits in handwritten zip codes. This summer, Google built the largest pattern recognizer of them all, a system running on sixteen thousand processor cores that analyzed ten million YouTube videos and managed to learn, all by itself, to recognize cats and faces—which initially sounds impressive, but only until you realize that in a larger sample (of twenty thousand categories), the system’s overall score fell to a dismal 15.8 per cent.


    Of course, to be fair, perhaps a properly designed mind built out of the right kind of hierarchical pattern recognizers--amongst other kinds of modules--would do very well at some range of tasks. But Kurzweil certainly isn't coming up with any design, construction, development, interface, environment or test strategies / ideas for a mind.
    vongehr
    Your expertly skirting around the implied charges by simply ignoring them certainly qualifies you for a spot among the "usual suspects". Nothing else have I ever seen from people like Brin, Terrile and so on, say to critical comments or open letters etc. You made it! ;-)
    John Hasenkam
    Like Hawkins, Kurweil also claims this is a biomimetic approach, focusing only on the neocortex, and within the neocortex only on cortical columns.


    I'd dismiss the whole effort right there. As if "mind" is located somewhere in the brain. "Mind" is a function of the whole body+environment(strictly speaking) so there is a trace of dualism in his thinking there, that little man is being proposed by another little man to reside in our Mountcastle columns. Big problems there:


    It was long thought that these columns were pretty much "up and down" in signal transmission whereas we now know there is a lot of "horizontal" transmission going on. 


    What about those sparse spindle cells, neurons with very long connections across the neocortex, perhaps moreso on the right side than the left. As I read recently loss of these cells can involve loss of prospective memory and sense of self. These types of cells are also quite rare across species, very present in humans, some higher primates, and possibly some whales and dolphins. 


    What about the chandilier cells? These big inhibitory neurons are in the neocortical layers and each of these cells can potentially inhibit hundreds of pyramidal neurons. In a grasshopper study I read one single giant GABA interneuron cell inhibited the entire output of the mushroom body. What I find so strange about AI is they seem to ignore that fundamental role of inhibition in cognition. As one bloke I call Mr. GABA once advised me: the neocortex is basically in an epileptic state kept under constant regulation by GABA inhibition. Once again: life at the edge of chaos. 


    Stop thinking about the brain-computer comparison, it is doomed to fail. The fundamental operations of nervous systems involves many processes that are not found in computing, from retrograde neurotransmission to backward propagation to whole body influences(hormones, cytokines, growth factors). One of the incredible mysteries is how the brain maintains a stable world in a constantly changing internal milieu(perhaps it doesn't, we are not conscious of the difference). Go on, try that with a computer, change the voltage and see what happens. 


    Mathematics is a function of intelligence, that does not mean it is the cause of intelligence. It may well be but I'd be more inclined towards the current popular view that if any mathematical is going on up there it is probably akin to a Bayesian process. 






     




    SynapticNulship
    I'd dismiss the whole effort right there. As if "mind" is located somewhere in the brain. "Mind" is a function of the whole body+environment(strictly speaking) so there is a trace of dualism in his thinking there, that little man is being proposed by another little man to reside in our Mountcastle columns. Big problems there:

    Yes that is why I am very interested in externalist approaches and embodied situated AI. The philosopher Alva Noë has pointed out that the philosophical assumptions in a lot of neuroscience mind research results in this new dualism (Out of Our Heads, p.5):

    It isn't surprising to be told that there is a thing inside each of us that thinks and feels and wants and decides. This was the view of the seventeenth-century philosopher Rene Descartes, who held that each of us is identical to an interior something whose essence is consciousness; each of us, really, is an internal res cogitans, or thinking thing. And this is the doctrine promulgated by many religious traditions. Of course, the religions, and Descartes himself, didn't teach that that thing inside us that thinks and feels is a part of our body, a bit of flesh, such as the brain. They supposed that it was something immaterial or spiritual, and so, in that sense, that it was something unnatural...It is precisely on this point, and only on this point, really, that today's neuroscientist breaks with tradition.
    So, I definitely agree with you there. Neuromorphic or weak-AI based proposals for strong AI need to at least try to explain the environment (whatever that may be), the types and ranges of inputs and outputs, and any other interaction details and how that affects the strong AI instance over time/experience.

    What about those sparse spindle cells, neurons with very long connections across the neocortex, perhaps moreso on the right side than the left. As I read recently loss of these cells can involve loss of prospective memory and sense of self. These types of cells are also quite rare across species, very present in humans, some higher primates, and possibly some whales and dolphins. 

    I believe Kurzweil does mention spindle cells (as he also did in The Singularity is Near (p.192)), but it doesn't matter. We're still left with still a vague vaporware explanation of the mind.

    Stop thinking about the brain-computer comparison, it is doomed to fail. The fundamental operations of nervous systems involves many processes that are not found in computing, from retrograde neurotransmission to backward propagation to whole body influences(hormones, cytokines, growth factors). One of the incredible mysteries is how the brain maintains a stable world in a constantly changing internal milieu(perhaps it doesn't, we are not conscious of the difference). Go on, try that with a computer, change the voltage and see what happens.
    I think you have gone astray here. I have a couple arguments for this:
    1. Does any researcher actually do new brain computer comparisons? I think those comparisons were made decades ago and are now stuck in the general culture.
    2. Using computers is perfectly fine, as long as you don't limit your designs or results to the designs or nature of the computer itself. One exception where the lower level affects the upper levels is that some claim they need true analog functions, and have actually worked with analog computers. Therwise, the power of software to produce unpredictable and/or random and/or complex behavior should not be swept under the rug.
    3. You are now claiming that fundamental operations of the nervous system at arbitrary levels of detail are necessary for creating a mind. That is not a sound statement as far as I know.


    Mathematics is a function of intelligence, that does not mean it is the cause of intelligence.
    Rodney Brooks once told me that a lot of his students were programming robots with complicated mathematical models onboard, completely missing Rod's suggestions that the models don't have to actually be running inside the robot. Models and various other agenticity assumptions come from observers. The programs running in a robot can actually be quite simple, or at the very least not be the model itself. The cellular automata people like Wolfram came to the same conclusion about the power of simple programs. The main problem with simple programs is, just like with narrow algorithms inspired by neurons, is how to design a mind and/or a construction program to result in a mind instead of just amusing unpredictable behavior.
    John Hasenkam wrote, "'Mind' is a function of the whole body+environment"

    I tried saying that (in only slightly different words) to Steven Pinker after one of his lectures. He looked quite hurt and said no, he did not agree. Suddenly I found myself feeling sorry for him, and so just told him he gave a good lecture, eliciting a big smile. What could be the harm in telling a polite lie?

    Hello, could you give some references for this passage from your post "The Disillusionment Of Math"

    "My readings of research and ideas connected to the behavioral robotics scene that had started in the 1980s (but I didn't know about that stuff until around 2001 or 2002). One of the concepts is--what can you do without an internal representation? Can you use the world as its own model in real time?"

    Thank you,
    Marius

    SynapticNulship
    R.A. Brooks, “Cambrian Intelligence: The Early History of the New AI,” MIT Press, Cambridge, MA, 1999.
    J.L. Jones and A. M. Flynn, “Mobile Robots: Inspiration to Implementation,” A K Peters, Wellesley, MA, 1993.
    J.H. Connell, “Minimalist Mobile Robotics: A Colony-style Architecture for an Artificial Creature,” Academic Press, San Diego, CA, 1990.
    V. Braitenberg, Vehicles: Experiments in Synthetic Psychology, MIT Press, 1984.
    John Hasenkam
    . You are now claiming that fundamental operations of the nervous system at arbitrary levels of detail are necessary for creating a mind. That is not a sound statement as far as I know.

    Not quite Samuel, what I am saying is that we cannot talk about "mind" unless we have some level of understanding of how it arises in wet brains. We can philosophize about mind all we like, and by golly some people really like that, but without sufficient empirical grounding we have words and more words but we don't have a theory. We still do not understand how anesthesia works. 
    It is not just some arbitrary detail but the combinatory result of all these arbitrary details. I'm inclined to the view that there is a seemingly endless re-iteration of rather simple processes that gives rise to the complexity. I'm inclined to discount the idea that evolution came up with some fancy algorithms to do the job. The fancy algorithms may well create a mind but it will probably be a mind but not as we know it, Jim. 


    and have actually worked with analog computers. 


    Which touches on some issues I didn't bother to raise. Neurons can perform functions in many ways, from the standard model to even spiking without input that can last for up to 30 minutes in some neurons, possibly dopamine neurons but my memory is poor on that. There is also axonic communication occurring. So even at the most basic level our understanding of neuron activity has a long way to go. So it seems to me that at this point in time it is somewhat premature to be suggesting we know enough about wet brains to start thinking about vague ideas like mind. Mind may well require embodied cognition. If AI wishes to avoid the amusing unpredictable behavior and seek something more substantive then AI can consider concepts like reinforcement and internalised drives like the 3Fs(to my knowledge this has been happening for quite some time. eg. Edelman's Darwin robot). Without those types of cognitive impetus what is there to do but engage in amusing unpredictable behavior? I don't like the concept "mind", it reminds me of Skinner's warning about relying on folk psychology concepts. 
    Too Much Hope
    As a transhumanist/singularitarian leaning individual I have done my fair share of reading, blogging, and visiting conferences including online meet ups. Your analysis was fairly accurate but somewhat misplaced. Ray’s attempts to outline “The law of accelerating returns” by describing timeframes and mechanisms of action that help to build a better knowledge base and better tools to work with are what inspire most people. In order to do this Ray gives examples of his past work and other peoples work and then extrapolates.
    The transhumanist/singularitarian crowds take a positive attitude and try to promote these technologies and the affects they will have over time. You sound disappointed when Ray’s talk of “creating a mind” is discussed using “his leadership of pattern recognition”. You claim that Ray’s technical discussion is too simplistic and imply it is wrong to use the title “How to create a mind” and “The secret of human life revealed” and other claims that Ray makes in the book. Your frustration is noted. Your rational for writing it is petty and unproductive. Sure your “rant” helps to set the singularitarians and H+ (psychotics) straight but seriously those are few and far between.
    You do not explain the emerging future in bio, nano, or AI with anything more than guesses either do I, or anyone else, except for those that are visiting us from the future (that last part was a joke). Yes science needs proof, and trends in technology are not proof. You can rip on Ray or Eric Drexler, or Aubrey de Grey, or George Church, or Peter Diamandis, etc. because they present what the future will become in some fashion or other. Everyone knows it will be a convergence of brand new technologies that will be used in the future. All of the above individuals consistently tell us this and while promoting today's newer technologies. Your disappointment in the book means that you seriously thought the “secret” was going to be revealed about 15 to 20 years earlier than what Ray, transhumanists, and singularitarians have been saying for decades (this will not happen until 2025 to 2029).
    All of the great books by H+ and singularitarians are full of examples about today’s emerging tech as if it will be the answer that will make things happen 5, 10, or 20 years down the road. Any serious reader understands that this is just to give an understandable basis to improved tech that leads to better things. Readers do not take the examples of the tech in a book that is 5 years old as gospel. They understand that many trends need to occur in an industry before it becomes the level of technology that transhumanists discuss for the 20s, 30s and 2045. The only thing “severely lacking” was your rational expectation of the book. You did give an Abundance of disdain to one of history’s greatest inventors and promoter of helping the world solve its problems.
    We want to recreate the mind only to find “tools” that we can use. The majority of human level artificial brains and AGI technology will be created in the future probably by semi-intelligent software.
    See you in 2095.
    Robin
    www.TheEmergingFuture.com

    SynapticNulship
    You sound disappointed when Ray’s talk of “creating a mind” is discussed using “his leadership of pattern recognition”. You claim that Ray’s technical discussion is too simplistic and imply it is wrong to use the title “How to create a mind” and “The secret of human life revealed” and other claims that Ray makes in the book. Your frustration is noted.

    My "frustration is noted?" Are you the singularitarian police?

    Your rational for writing it is petty and unproductive.

    Possibly, but it was worth it just for the few discussions already started here.

    You do not explain the emerging future in bio, nano, or AI with anything more than guesses either do I, or anyone else, except for those that are visiting us from the future (that last part was a joke).

    That is not my job or hobby. I have never claimed to be "explaining" the future, emerging or non-emerging. If I have not clearly labeled speculations then I apologize for that. The point of my blog is primarily to discuss cognitive science, with an emphasis on AI and interfaces. Kurzweil has only come up on my blog at this point because he has made a big splash (at least according to Amazon sales rankings) with a book supposedly about AI and how the mind works.

    You did give an Abundance of disdain to one of history’s greatest inventors and promoter of helping the world solve its problems.

    If Ray is offended by my blog post, I basically don't give a shit. If he would like to explain why he wrote this book and what it was supposed to achieve--and money is a valid reason--then I am all ears. Ray, if you're in Boston or Cambridge this week and want to have a chat let me know.

    We want to recreate the mind only to find “tools” that we can use. The majority of human level artificial brains and AGI technology will be created in the future probably by semi-intelligent software.

    I dare say all, not the majority, will be created in the future. Unless you have a time machine.
    Hank
    I'm always fascinated by the extremes we get; we have a demographic of people that is overwhelmingly anti-science and a contingent overwhelmingly optimistic, but both poles don't actually understand what science is. 
    In order to do this Ray gives examples of his past work and other peoples work and then extrapolates.
    sounds like just what Samuel said it sounded like; a way to sell some books to the faithful but nothing meaningful.  Someone who said a hundred years ago that we would beam information using radio was not a prophet, any more than science-fiction writers who wrote about teleporters were creating a roadmap.  Humans + Science/Magic = Singularity is not any great vision, it is just having an imagination.

    Transhumanists are absolutely more fun to be around than the anti-science people who insist we are doomed, but there is a real lack of understand of what 'intelligence' is and, as I have noted before, Kurzweil is stuck in the 1960s regarding his understanding of neuroscience and technology. 
    one of history’s greatest inventors and promoter of helping the world solve its problems.
    Yeaaaah, that's an objective analysis.  You're a fine acolyte but I suspect that you don't really know much about the core issues in what you are praying for.
    vongehr
    The only thing “severely lacking” was your rational expectation of the book.
    Well, that is the point now, is it not, namely that people like Kurzweil are responsible for us having to even rationally expect nothing anymore from books.  In case you find "how to create a mind" and had "the secret of human thought revealed", what would you title your book in order to tell us that such is the content? "God found: The secret of the universe revealed"?

    These players of publishing markets have to hype their shit so much because they have not added any insight for years since they are busy 24/7 selling themselves.  Looking at your webpage, it seems that you defend such in order to participate.  The problem for people not just into money and fame is:  Where among all this crap is there anything worth looking at, how do I find it?  Samuel saved people some time here.
    My only substantive comment is that after first using a computer in the late 1960's, and having subsequently worked in both hardware and software for about thirty years, I never, ever expected to own a shirt-pocket wireless telephone that shows photos, plays music and films in hi-fi over wi-fi, has two quite decent still and moving picture cameras, calculates almost instantly in a wide variety of arbitrary units and mathematical functions, tells me where I am anywhere in the world with a map, aerial view and directions, and can locate and transmit back to me, within a few seconds, the answers to about 98% of the questions I can ask.

    Now that I casually carry around such an astonishing device, I cannot begin to imagine what will come of millions of people all over the world using today's devices to help them invent even more amazing stuff. I can only stand in awe and feel fortunate to be living in a world with the amazing technology that has developed in my little lifetime.

    Or am I just hallucinating?

    Yes that is the point Sascha, read what you want and believe the kind of drivel produced by China's National Laboratory of Solid State Microstructures. The normal consumer reads the book jacket before buying a book and in the book jacket the first sentence says "bold futurist" and "technological visionary". This should tell you a lot about the book you will read. These supposed claims are simply projections about things that exist, your backwoods naivete and academic micro vision has gotten you all bent out of shape. The only reason you and the pseudo 2.0 crowd rail against Ray is because, well who the hell knows why you do it, it is simply speculation about AI modeling.
    Go ahead and go after the need to sell books as a bad thing. How do your handlers make the money to pay your salary and run the institution? I seriously doubt you would look into their source of revenue very deeply. And you sure as hell wont do anything about it. Samuel Kenyon readily admits his purpose of writing such an inane blog post "it was worth it just for the few discussions already started here"
    By the way how did you get to read this online are you working underground with some dissidents or are you a double agent?
    Cheers
    Robin