Accelerating Future Transhumanism, AI, nanotech, the Singularity, and extinction risk.

3May/1015

More Cat Brain Nonsense from IBM

MSNBC reports:

Changizi discussed his idea with an IBM researcher working on the SyNAPSE project run by the U.S. Defense Advanced Research Projects Agency (DARPA). The Pentagon agency has recruited the help of IBM, HP and leading research institutes to try to develop an artificial brain that is approximately cat-like in terms of size, number of brain cells and synapses, anatomical structure and even behavioral complexity.

No doubt, this goal was inspired by IBM's ridiculous and misleading claims that it had created an artificial brain with the complexity of a cat brain. If it's already achieved the cat brain goal, then why is it still working towards it? The fact is that Dharmendra Modha, head of the IBM team that developed the neural net simulation, has lost all credibility. Any news coverage involving IBM and artificial brains should be essentially ignored because Modha's claims were explicitly designed to mislead the public into thinking that he had created a cat-complexity artificial brain when he achieved nothing of the sort, and we only have reason to expect more of the same.

Instead of creating "an artificial brain that is approximately cat-like in terms of size, number of brain cells and synapses, anatomical structure and even behavioral complexity", why not begin with a nematode, or a rotifer? A rotifer brain has about fifteen cells. Create a virtual rotifer with the same complexity of behavior as a real rotifer, and then maybe they can move on to planarians and other flatworms, and after a decade or two, insects.

Trying to create artificial intelligence by painstakingly copying biology seems to be an almost hopeless path. More promising seem to be routes that use machine learning, Bayesian networks, and insights from theoretical computer science. These systems are very different than "good old-fashioned AI" because they are fundamentally based on fuzzy probabilistic reasoning, which works very well.

Filed under: AI Leave a comment
Comments (15) Trackbacks (1)
  1. Why is it hopeless? You could argue that “painstakingly copying biology” would probably be harder than using the other routes you mentioned, but given that the human brain exists can’t we be almost certain that if you created a program that simulated the brain (and perhaps some of the body) in enough detail you would have a human level artificial general intelligence? Of course, this doesn’t mean that IBM is anywhere close to having an artificial cat-like brain.

  2. I’m curious why even the likes of the Blue Brain group isn’t starting with nematodes or fruit flies.

  3. James, I said “almost hopeless”, because it’s so complicated. It would be like trying to build a flying machine by copying a bird. It only makes sense if you know nothing about the underlying principles of the operation of the machine which let you abstract away unnecessary complexity.

    Jordan, I wish they would. The reason why is because using billions of point neurons sounds more impressive to the uneducated (like earlier versions of myself, for instance) than using 15 complex neurons that actually emulate the organism ostensibly being modeled.

  4. So Mike you are a neuroscientist or a computer scientist?

    You have read IBMs research and the Blue Brain project research and understood it?

    You have created something better?

    Are the answers to any of these questions yes?

    I would caution against railing against people who actually do real research and have made real contributions especially when your only claim to any form of significance is that you have a blog of modest importance.

    If you properly understand what IBM is doing its interesting and the progress they are making in their simulation is important. You can also argue that the model of the neuron they use is not a good enough one but the results of their work will still advance our understanding. The fact that the claims are sensational should come as no surprise since the history of science is riddled with sensationalist claims.

    “James, I said “almost hopeless”, because it’s so complicated. It would be like trying to build a flying machine by copying a bird. It only makes sense if you know nothing about the underlying principles of the operation of the machine which let you abstract away unnecessary complexity.”

    This is of course silly since in the context of the brain we really don’t comprehend enough to be able to say what is relevant, and what is not to making the brain work. We can’t simplify the design because we don’t understand enough of the design. If you doubt this then start reading the current neuroscience research.

  5. The MSNBC article is about “crafting artificial intelligence on the level of a cat’s brain”. Existing claims about cat brain capacity were about the hardware only. The simulation didn’t *actually* work – and produce a cat mind – but the researchers never claimed that it did.

  6. So Mike you are a neuroscientist or a computer scientist?

    I know enough neuroscience to know that IBM is completely full of it here. The fact that they were excited by non-random oscillations, and call this the “key science result”, is just stupid. Non-random oscillations have been observed in neural networks for a long time. For instance, here is a paper from a 1994 describing a model that “organizes figures from elementary visual signals by self-organizing coherent dynamics of neural oscillators”. In response to this claim, Henry Markram (a real-life scientist who has been studying the brain for 30 years, unlike myself) said, “That model exhibited oscillations, but that always happens so even simulating 100 Billion such points interacting is light years away from a brain. see: http://www.izhikevich.org/human_brain_simulation“.

    Here is the key claim in the paper:

    A simulated electroencephalographic (EEG) like recording of the response to stimulation in the above model showed 12 full oscillations over the course of the one second simulation (Figure 3A). A further analysis revealed that activity in the model did not occur randomly, but rather propagated through the network in a specific fashion.

    The way that they call their recording “EEG-like” is wishful framing. They want it to be like an EEG, so they call it that, so people are deceived into thinking the oscillations are profoundly brain-like at some deep level, but the results described in the paper are not impressive at all.

    If they had anything even vaguely resembling a brain, then they would describe the brain-like characteristics of low-level dynamics, instead of just writing a couple paragraphs about global dynamics and following it up with pages and pages of analysis on the hardware that is not even implementing anything of interest. As Markram pointed out in an answer to a hypothetical questions:

    Is it nonetheless the biggest simulation ever run?

    No. These people simulated 1.6 billion points interacting. They used a formulation to model the summing up and threshold spiking of the “points” called the Izhikevik Formulation. Prof Eugene Izhikevik himself already in 2005 has run a simulation with 100 billion such points interacting just for the fun of it: (over 60 times larger simulation). This simulation ran on a cluster of desktop PCs every graduate student could have built. This is no technical achievement.

    Why does IBM even bother to make these claims when all the information to refute their significance has been easily available, even before Markram spoke up?

    You have read IBMs research and the Blue Brain project research and understood it?

    Yes. I understand all of the neuroscience and computer science terminology in the “Cat is Out of the Bag” paper. The paper had better use a lot of big words, because if the accomplishments were presented in plain terms, they wouldn’t be a big deal. That is why Modha had to actively mislead the public using pictures and terminology of cats. Near the bottom of the paper is a misleading graph that implies that what they have created is essentially 6% of a human brain simulation. What makes a cat or a human is not the mere existence of a large network of interconnected nodes, but the way those nodes interact and their internal structure. (Practically non-existent in this case.) Because Modha did not describe any low-level interactions or even explicitly give arguments for WHY the high-level oscillations he observed are so similar to what neuroscience knows about animal brains, I have to assume that what you see is what you get.

    You have created something better?

    Um, no. Henry Markram is working on something better, and he criticized the project. I don’t see why I have to create something better for this project to be a farce. I am just fascinated by how an already-established figure like Jeff Hawkins or Dharmendra Modha can rattle off a few basic neuroscience concepts, the kind you can find in any introductory book on neuroscience, and people are all impressed. If everyone just skimmed MITECS in college for a week or two, this wouldn’t happen. In the “Cat is Out of the Bag” paper, it says this:

    A large proportion of synapses are plastic, that is, the effect of their activation on the corresponding post-synaptic neuron is subject to change over time using a plasticity rule. Synaptic learning is captured by Donald Hebb’s principle: neurons that fire together, wire together. Our simulations include a form of spike-timing dependent plasticity (STDP) [36] that potentiates (increases the weight of) a synapse if its post-synaptic neuron fires after its pre-synaptic neuron fires, and depresses (decreases the weight of) a synapse if the order of two strings is reversed. Synaptic plasticity allows networks of neurons to extract, encode, and store spatiotemporal invariants from the environment.

    People have to understand that this is not the first time someone has done the above, by a long shot. The team may have built a canvas for a neural simulation (maybe), but they haven’t come anything close to something that resembles a mind, because they didn’t even apparently try to introduce any low-level functional structure. A simulation of billions of point nodes that non-randomly oscillate with zero functional structure is not a scientific accomplishment. Many people could do it with backing from IBM. As Markram points out, the primary achievement here is that they built a supercomputer.

    I would caution against railing against people who actually do real research and have made real contributions

    Building a supercomputer is not a new accomplishment. Back in 2007, Izhikevich already built a human-scale neural net simulation. I don’t understand why this is hard to understand. Did Izhikevich write a paper titled, “The Man is Out of the Bag” when he completed that simulation? No. The reason I rail against it is because this is a transparent PR stunt and if the public actually believes it, it will make it hard for subsequent simulations in the future that actually try to implement low-level functional dynamics to get publicity. When something actually important happens, the public will be desensitized, because they were misled back in 2009 by IBM. This could be a problem because progress towards whole brain emulation could lead to a Singularity. Appraising its progress correctly is of the utmost importance.

    IBM could have just used a problem like calculating the digits of pi to show the world how awesome their computers are. They decided to go the next step, by creating a “cat”-like simulation. Whoever made that decision, probably Modha, made a mistake. They shouldn’t have even claimed they created a mouse back in 2005 or whenever it was. This is what happens when you give engineers and computer scientists a lot of money, they learn a little neuroscience, and then get all excited. It’s the same exact story with Jeff Hawkins. When I was learning about this stuff back in the day, I too got excited about how I learned how neural oscillations work and that visual cortex is organized in hierarchical abstraction layers. However, I thankfully lacked the money or reputation to hyperventilate about it to the public and make an ass of myself. Neuroscience knowledge among tech nuts, uncritical Singularity fans, and Silicon Valley businesspeople is low enough that they buy into this stuff.

    If you properly understand what IBM is doing its interesting and the progress they are making in their simulation is important.

    If you properly understand, you will see why what they have done is not a big deal and is all part of a PR stunt to mislead the public and provide advertising for IBM. I’ve already tried to explain it above. Do you work for IBM?

    The fact that the claims are sensational should come as no surprise since the history of science is riddled with sensationalist claims.

    More importantly, the history of science is riddled with sensible people who have spoken out against sensational claims.

    “James, I said “almost hopeless”, because it’s so complicated. It would be like trying to build a flying machine by copying a bird. It only makes sense if you know nothing about the underlying principles of the operation of the machine which let you abstract away unnecessary complexity.”

    This is of course silly since in the context of the brain we really don’t comprehend enough to be able to say what is relevant, and what is not to making the brain work. We can’t simplify the design because we don’t understand enough of the design. If you doubt this then start reading the current neuroscience research.

    False. Substantial work has already been done on reverse-engineering the hippocampus and auditory cortex. Many computer scientists and cognitive scientists are converging on the concept that the brain uses Bayesian inference rules to process sensory data on all levels, and an edited volume was published on the topic in 2007. Any neuroscience article that presents a theory to simplify the observed data is saying what is important to cognition and what is not. The number of papers that have made incremental progress on this problem could fill several wheelbarrows. It is not an on/off, we understand/we don’t understand situation. We gain new knowledge every day.

    Tim,

    The language of the press release and setup of Modha’s website strongly implied it, though avoided specifically claiming it. Note how the excerpt above cites “behavioral complexity”, implying that the simulation could have the behavioral complexity of a cat? Yeah right. We can’t even simulate the behavioral complexity of a nematode or insect yet.

  7. So the key point is you’re not a real scientist and you are just here to rant because you think IBM is doing something wrong. What is more because you have a basic understanding you think that your opinion and analysis is worth something.

    New Flash: Until you have some credentials this is all hogwash.

    So now that this little rant is out of your system do the entire singularity community a favor and get over it. Its amateur ranting like this that makes you guys a laughing stock in the scientific community. Yes I know you’ll argue that this is false but something I have found about you is that your view of the world is so narrow you can’t see past the end of your nose until it hits you in the face.

  8. Building a flying machine by copying a bird hardly qualifies as impossible. It’s harder than what the Wright brothers did, sure, but should be doable within a few decades if not sooner. Pushing the Singularity back a century only would make it minimally less revolutionary.

    As for the argument about the IBM project, the reasoned and detailed critique makes up for the lack of credentials. You should try responding to Michael’s actual points, Bob.

  9. So the key point is you’re not a real scientist and you are just here to rant because you think IBM is doing something wrong.

    Lying to the public, tarnishing the field with false claims, decreasing the probability that future legitimate projects will get the attention they rightly deserve, and ultimately endangering the human race by confusing us about progress towards producing implementations of intelligence that match or exceed our own.

    News Flash: Until you have some credentials this is all hogwash.

    Not at all. The truth is the truth no matter who says it. What’s more, I’m on the same side as highly credentialed people such as Henry Markram, though I would speak out whether or not he did.

    The dumbest person in the world can say that the Earth orbits around the Sun, and he’d still be right. Credentials, while useful as a filtering heuristic, are hardly the be-all and end-all of knowledge.

    If you simply feel indignation at the very concept that an uncredentialed person can have a strong public opinion on a scientific matter, then you have been brainwashed. People invest a tremendous amount of mental and emotional energy in their formal education under the pretense that having one is a prerequisite to being taken seriously by others, but (surprise!) it isn’t. An informal education can do just as well, as long as it is rigorous. Yet the mostly profit-driven machinery of Big Education has every incentive to spread the meme: “go through us, buddy, or you’re going to be a nobody”.

    So now that this little rant is out of your system do the entire singularity community a favor and get over it.

    Never.

    Its amateur ranting like this that makes you guys a laughing stock in the scientific community.

    Not at all. We get the same reaction when scientists produce the arguments. As I already pointed out, Henry Markram, a scientist who has studied the brain for over 30 years, shares my exact same view on the subject.

    Yes I know you’ll argue that this is false but something I have found about you is that your view of the world is so narrow you can’t see past the end of your nose until it hits you in the face.

    How so?

  10. I don’t see why you are bringing Jeff Hawkins into this same context, Michael. Hawkins started a neuroscience institute and has actually been studying the brain for over 20 years. Plus, he is actually doing the thing you are saying is most important – focusing on structure rather than size.

  11. “Credentials, while useful as a filtering heuristic, are hardly the be-all and end-all of knowledge.”

    Need I say more, you have none thus I filter your “professional annotations” out.

    “If you simply feel indignation at the very concept that an uncredentialed person can have a strong public opinion on a scientific matter, then you have been brainwashed. People invest a tremendous amount of mental and emotional energy in their formal education under the pretense that having one is a prerequisite to being taken seriously by others, but (surprise!) it isn’t.”

    You’re an example of this? Evidence please. Names of organizations, jobs held etc. So things like IBM, Intel, CMU, DARPA, Fermi etc. anywhere?

    ” An informal education can do just as well, as long as it is rigorous. Yet the mostly profit-driven machinery of Big Education has every incentive to spread the meme: “go through us, buddy, or you’re going to be a nobody”.”

    Yeah, and that is why you are such a well respected voice out in the scientific community because of your lack of credentials. Please that is a load of total BS. You can make the argument on a case by case basis that formal education is not needed but those are special cases not the norm and not you.

    I cite as evidence your total and complete lack of meaningful accomplishments. Unless you have ones I am not aware besides this blog (unfiltered thought -big deal), SIAI PR guy (big deal, an organization composed of primarily amateurs hiring an amateur huge surprise).

    “Not at all. We get the same reaction when scientists produce the arguments. As I already pointed out, Henry Markram, a scientist who has studied the brain for over 30 years, shares my exact same view on the subject.”

    Yeah, and he isn’t the one adding to the uniformed outrage. That would be the department of the genetically pre-disposed to ignorance and you. So since you are not adding anything to the issue or changing anything get over it.

    “How so?”

    You’re a transhumanist and a believer in the singularity need I say more? You seem to have an avid interest in what SIAI does, you think you’ll obtain immortality through cryonics. Your science education consists of reading more science fiction, comic books and watch Sci-Fi movies then it does actual science. I can go on and on… People of little knowledge who think they know a lot almost always, especially when they are somewhat arrogant, tend to be narrow minded-can’t see past the end of their noses type of person.

  12. @Bob: You’re giving a textbook example of why appealing to authority can be a logical fallacy. Instead of addressing the substance of the issue, you fixate on credentials.

  13. BobBonker, let me turn that back on you: please list your achievements and qualifications, that allows you dismiss the opinions of others.

    As opposed to the dutch courage that anonymity seems to lend to you.

    In fact your assertions (or rather, ranting ad hominems) are spurious and ignorant of the history of science. Many enthusiastic amateurs have made significant contributions to science.

    Your assertions reek of intellectual snobbery and arrogance – ironic, considering you accuse others of such. The sophomoric venom in your tone betrays a lack of objectivity and no small penury of maturity.

    Essentially you are saying that only scientists may comment on science, and the rest of us should stand by in awed silence, much as the Christian church controlled literacy during the dark ages to maintain an elite position.

    An irreverence towards ‘credentialled’ scientists is not only a right but a necessity. Scientists are just as guilty of stupidity, prejudice and narrow-mindedness as person ‘of little knowledge’.

    Antoine Lavoisier once said of meteorites, “Stones cannot fall from the sky, because there are no stones in the sky”, so don’t talk to me of arrogance and narrow-mindedness!

    The Manhattan Project was in the process of detonating the first atomic bomb when it occurred to one of the scientists that it might set the atmosphere on fire. So we should always bow to credentials, because it automatically imbues the possessor with a greater sense of responsibility?

    To repeat, please list your *verifiable* achievements, before heaping scorn on others. Or take your repellent trolling elsewhere.

  14. BobBonker, you haven’t even bothered defending your company with facts here, you’ve just used ad hominem attacks. More and more experts will step forward to the same side as mine on this issue. Your arguments would only be persuasive to someone already firmly on your side, i.e., other unquestioning IBM employees.

    It’s really pathetic how you keep asking me to stop, as if I actually would, because that’s obviously what you actually want — to do PR defense for your company. But you have obviously failed because you only prompt me to keep responding.

    You keep repeating my lack of credentials over and over again, but as I said, Markram has plenty of credentials, and I’m sure you could easily find a PhD neuroscientist that could tell you that these systems aren’t even vaguely brainlike at anything but the most superficial level.

    Yeah, and he isn’t the one adding to the uniformed outrage.

    He has! Did you miss the entire hubbub a few months ago? It was covered by all sorts of places. I’m not the first person to criticize this project.

    That would be the department of the genetically pre-disposed to ignorance and you. So since you are not adding anything to the issue or changing anything get over it.

    I am adding to the issue and changing things, or you wouldn’t bother to get so huffy about it. Also, your blatant ad hominem is ridiculous. As my comments above prove, I’m hardly ignorant on this issue. You haven’t even said WHY I’m supposedly wrong, just tossed out a bunch of ad hominem. Here’s an argument pyramid, for your convenience.

    Notice how you’ve blatantly insulted me dozens of times and I haven’t even bothered to insult you? I deserve a Ribbon or something.

    Your science education consists of reading more science fiction, comic books and watch Sci-Fi movies then it does actual science.

    My science education is actually better than most scientists’, because I’ve extensively read scientific papers, books, and articles from numerous disciplines and talk frequently to interdisciplinary scientists whereas many scientists are only familiar with their area of specialty. Also, I display high comprehension at a high reading speed.

  15. I agree with Michael. I think that both approaches are complimentary and have rights to exist. Creating low level simulations of simple networks will ultimately be a faster way to the singularity. The IBM project will never produce any conscious machine or even something close, but it may be useful to understand how high level emulations might work in the future and what hardware/software algorithms are needed to make them closer to the real thing. More thoughts on this issue:
    http://singularsociety.blogspot.com/2010/06/why-strong-ai-efforts-have-failed-for.html


Leave a comment

(required)