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


More Nonsense Reporting Overblowing IBM’s Accomplishments

Last month in New York I had the pleasure to talk personally with the creator of Watson, Dr. David Ferrucci. I found him amicable and his answers to my questions on Watson very direct and informative. So, I have nothing against IBM in general. I love IBM's computers. Several of my past desktops and laptops have been IBM computers. The first modern computer I had was an IBM Aptiva.

However, there is a constant thread of articles related to claims being reported that IBM has "completely simulate(d)" "the brain of a mouse (512 processors), rat (2,048) and cat (24,576)", which was revived in force this last weekend. This is entirely false. IBM has not simulated the brain of a mouse, rat, or cat. Experiments have just recently been pursued to even simulate the 302-neuron brain of a flatworm, for which a wiring diagram exists. Instead, IBM has made "mouse-SIZED" neural simulations, "rat-SIZED" neural simulations, and "cat-SIZED" neural simulations, given certain assumptions about the computational power of mammalian brains. The arrangements between neurons being simulated bear little relation to the actual wiring diagram of neurons in these animals, which are not known. Given the tools we currently have, like ATLUM, it would take tens of thousands of years to determine the full connectomes of mice, rats, or cats.

I can never tell if it is the reporters who are being ridiculous, or IBM is deliberately misleading the public. However, I think IBM should issue a press release that clarifies the situation. Directly quoting Scientific American:

IBM describes the work in an intriguing paper (pdf) that compares various animal simulations done by its cognitive computing research group in Almaden, Calif. The group has managed to completely simulate the brain of a mouse (512 processors), rat (2,048) and cat (24,576).

The paper they cite is the same damn paper from 2009, "The Cat is Out of the Bag", which I immediately reacted to negatively within days of its publication. Since then, I've been watching as this false meme, which has yet to be directly repudiated by an IBM representative, makes its way through the media, which doesn't know any better.

Now, IBM is allegedly claiming that they simulated 4.5% of the (processes?) of the human brain, or at least hundreds of media sources are reporting it. All the media sources seem to just be linking the two-year old paper "The Cat is Out of the Bag", so I'm not sure if there was a recent announcement or it just took the media two years to pick up the story.

Again, it's impossible that IBM could simulate 4.5% of the human brain, because we (human civilization) don't have 4.5% of the wiring diagram of the human brain to use as raw data to build a simulation. We don't even have 0.1% of the wiring diagram of the human brain, I'd estimate, but you'd have to ask a computational neuroscientist (not one from IBM) to get a more informed guess.

We have the wiring diagram of the 302 neurons in the flatworm brain. That's about it.

The vast majority of Reddit commenters are clueless and missing the obvious error. Even this seemingly educated comment misses the point that there is NO WIRING DIAGRAM for the parts of the brain IBM allegedly simulated. Even this "best of class" comment seems to take the reporting at face value, as if 4.5% of the human brain had been simulated, and criticizes neuron models instead of the "elephant in the room" that I've explained.

Reddit commenters fail for being fooled, the media fails for reporting a false story, and IBM fails for not issuing a clarification. In many cases IBM seems to actively encourage the misconception that a full feline connectome has been simulated.

My prediction is that AGI will be invented and we will have a full-blown Singularity before a complete cat connectome (much less human connectome) is created.

This whole issue is important because the public is already confused about computational neuroscience enough as it is. I see computational neuroscience as very important, and it's important that the public -- and scientists, who despite their alleged higher level of thinking, frequently pull their beliefs from popular articles like everyone else -- know what is and hasn't been accomplished in the field.

For a nice article on connectomics and what has been accomplished so far, see this article from Microsoft Research. It correctly highlights ATLUM as the only technology that is precise enough to get slices that can be imagined in sufficient detail to build a connectome. ATLUM, by the way, was invented by a transhumanist, Ken Hayworth. (Why do people say that transhumanists don't contribute to science?)

Here's yet another article.

Filed under: AI Leave a comment
Comments (21) Trackbacks (0)
  1. We don’t even have the flatworm circuit diagram. According to David Dalrymple, who is trying to emulate C. elegans:

    The “connectome” of C. elegans is not actually very helpful information for emulating it. Contrary to popular belief, connectomes are not the biological equivalent of circuit schematics. Connectomes are the biological equivalent of what you’d get if you removed all the component symbols from a circuit schematic and left only the wires.

    (Note another transhumanist advancing science btw)

  2. Fascinating. I’m also surprised to see that Dalrymple calls himself a disciple of Kurzweil.

    • That article on LessWrong and its followup (linked in the comments to the original) are possibly the best I’ve ever read on this subject. Might you could give them their own entry here, Michael?

  3. Thanks for debunking this. Can anyone help me in understanding what might constitute an adequate description of a flatworm brain, for the purposes of emulation?

  4. I am with you in this one. I have been complaining about such exaggerated claims for quite a while (at least three years ago on the old AGI list).

    And in fact there has been a discussion today on Google+ (prompted by Derya Unutmaz) on the claims about the mouse connectome is in process of being mapped (based on a report here One thing that I discovered there is that they are making more progress than I thought, at a higher resolution (approx 10nm pixel size). But even so synapse strengths are invisibly small. That, at least, is an attempt to build a real map.

    But the IBM claims seem to be based on “We have made statistically plausible replicas of about the same size as X”, which gets publicized as “We have built a computer map of the X brain”. A new low in the ongoing scientific Hype Wars.

  5. Fascinating discussion.
    Good post Michael!

  6. What do you think about Dr. Markram’s (exponential) projections for the Blue Brain Project?

  7. IBM has made “mouse-SIZED” neural simulations, “rat-SIZED” neural simulations, and “cat-SIZED” neural simulations

    Even that would’ve been an amazing breakthrough, if they’d been biologically detailed neurons. I like how Markram put it:

    These are point neurons (missing 99.999% of the brain; no branches; no detailed ion channels; the simplest possible equation you can imagine to simulate a neuron, totally trivial synapses; and using the STDP learning rule I discovered in this way is also is a joke).

    As for Henry Markram’s exponential projections for the Blue Brain Project… Even Kurzweil told him it was too optimistic.

  8. Very good analysis. It is terrible to see how misleading such a large corporation can be in their PR. Decades ago IBM would never have done this.

    Seriously, IBM should hire Mr. Anissimov (or someone of his integrity) for handling their PR!

  9. All I see is e-peen waving.

  10. Is there any consensus on what level of detail might be necessary for a flatworm brain simulation to respond appropriately in a virtual environment?

    Is it a question of increasing the detail level upto the point that ‘it works’ in some sense?

  11. “ATLUM, by the way, was invented by a transhumanist, Ken Hayworth. (Why do people say that transhumanists don’t contribute to science?)”

    This comment falls below the usually high standards of this blog.

    Transhumanists don’t contribute to science; scientists do. The transhumanist in question didn’t apply transhumanism to the research problems; he applied science. He hadn’t learned transhumanism in order to do his scientific job; he’d learned science. There is no scientic field called transhumanism. No prizes are awarded for accompllshments in transhumanism.

    Do not conflate science and transhumanism.
    Do not say transhumanism has anything to do with science or science with transhumanism. Stop connecting the dots ‘science’ and ‘transhumanism’. To even speak of transhumanism and science in the same breath is wrong. It sends the wrong message to people who can’t tell the difference – maybe that is the goal?

    There are religious (= superstitious) scientists, but no scientific findings have ever been made by applying supersitition. No discoveries will be made by transhumanism or transhumanists; only scientists. Transhumanism is not science; only science is.

  12. And I say this as a scientist also happens to be a transhumanist.

  13. Is Dalrymple a transhumanist? He’s a not too shabby pianist:
    If he is, YouTube gets it right by linking videos with people looking more transhumanistic that I’ve ever seen! The attires and makeup are sheer perfection. Check out the freaky eyes at 0:42
    Now that’s what a transhumanist looks like!

  14. Hi Michael. I’m not sure if you’re aware of this: when I look up “accelerating future” using the Bing search engine on Internet Explorer (school computer), the link to your site includes “Next Day Delivery Cialis” in the title.

  15. Big corporations are running out of ammo. Microsloth’s Future Productivity vision is also a bunch of nonsense reporting. We’ve been seeing similar videos for about a decade now and nothing has happened. Windows 8 isn’t even close.

  16. I have found that my level of self discipline in offline and online businesses is in direct proportion to the level of income I have achieved.

  17. I do not understand why the author of this article thinks that anyone who wishes to simulate a brain has to simulate each individual part of it.

    It’s like saying, I want to simulate a ball flying through the air. What would that mean? That I would have to put air particles all over the place? I’d be an idiot.

    I guess the author never heard of hidden markov models and replications of human ability to recognize words from voices and letters.

    Did we really replicate the parts of the brain for speech and text recognition or we replicated the algorithm? It seems we did the latter by accident as the new research implies.

Leave a comment

Trackbacks are disabled.