A New Marriage of Brain and Computer

Quantum consciousness has attracted a lot of total quacks, running 10 steps ahead of science and using "quantum word salad" to justify whatever beliefs they already had. For so many people, "quantum" reads as "magic", and flick the critical thinking is turned off, and off we go on an adventure of telepathy, auras, out-of-body adventures, and the whole litany of New Age-isms that might as well come from the late 19th century. Only this time with the word quantum in it, for that extra helping of plausibility. That's never a valid use of science.

But that's unfair of me to use that as justification to dismiss the entire idea. There is no idea so right that idiots can't misuse it. "Quantum" itself is the canonical case. Quantum mechanics is now over a hundred years old, but you'd never know it from the public perception. It remains counterintuitive, but it's not fair to call it a mystery any more. A mystery to you and I, perhaps, but it has long ceased to be an anything-box in physics.

If consciousness research has proved anything, it is that all our simple models are inadequate, and the final answer, whatever it may be, is going to be complicated.

A New Marriage of Brain and Computer recently went by on the Google TechTalk feed, and while there's still a little bit of the quantum = magic in there, there is also interesting material to chew over regarding the simple question of "How does the brain really work?" You can consider the consciousness discussion an irrelevancy and still learn some interesting things. The rest of this post assumes you've watched that video. (Google's video interface is superior to the embed version, so I give you the link.)

The most intriguing aspect of the video to me is the assertion that there is more to thinking that what we see in the traditional neural model. Two things are interesting about that:

1. The moderately intelligent behavior of a paramecium, as discussed in the video would seem to indicate non-neuronal computation can occur. I have not examined the claims in detail, but any intelligence is more than than can be explained with 0 neurons.
2. We've never really built a successful model of biological neuronal intelligence; in particular, we've never built a computer-executable model of a neural net that we'd consider "intelligent". We have some things we call neural networks, but we are stuck with either "feed-forward" networks, which have no feedback loops in the net and can't demonstrate anything like symbolic intelligence, or networks with feedback, which nobody seems to have any really good idea how to train, at least in the sense of seeing intelligence. Adding another entity to the system may be necessary; in the short term it violates Occam's Razor, but if the "microtubules" prove a necessary entity, then so be it.

The idea that there may be more to the brain is worth considering. The scientific demonstration of phenomena in the brain that can only be explained in the quantum regime is interesting.

The computational estimates have striking implications, if true. Kurzweil estimates approx. 10^16 operations per second in a human brain, Dr. Hameroff estimates 10^28 operations per second. (Plus, left unspoken in the video, these extra operations are at least partially in the quantum regime, which may or may not mean anything for the resulting computational power but certainly means that true simulation on a conventional computer would be harder yet again.)

What immediately leaps to mind is that we sure don't seem to get much for that. No, I don't mean that old tired adage about using only 10% of your brain. I mean that for 10^28 operations per second, you'd really expect us to be able to do more impressive things with it. Then again, given how little we know about consciousness, perhaps that's simply a necessary aspect. We don't have the first clue how to even formulate the question "What is the minimal conscious computational system?" in any sort of computationally-useful formalism.

On the list of things that bother me:

• Yes, it is speculated that QM may imply the ability to send signals back in time, but this cuts dangerously close to invoking magic. Nobody has demonstrated the useful transmission of signals backwards in time yet. Useless demonstrations have been made, but always the math forces them to stay useless. They may be mind-bending, but they aren't paradigm-shifting.

  Besides, I am of the opinion that everywhere you think you see signals going "backwards in time" or "faster than light", you are actually looking at things with the wrong reference frame, forcing a Euclidean frame onto a fundamentally Minkowskian universe. This is a theme throughout chapter 9 of Reflections on Relativity, but is most directly discussed in 9.9 Locality and Temporal Asymmetry. In other words, I believe the correct way to bend your mind is to come to a deeper understanding of time so that you see there is no backwards-in-time communication, not insist on using the wrong geometry and then standing amazed at the contradictions.

  If we have a epiphenomenal problem of neural activity following only after putatively-conscious actions, it seems a much simpler hypothesis to suppose that the conscious decisions were made in the microtubules, in real-time, and that subsequent neural activity is some other clean-up process or further delayed reactions. If the microtubule hypothesis is true, why wouldn't we use the microtubules for conscious decisions, since the microtubule system is both faster and more computationally powerful? In this model, what we're seeing is the slower neural system being notified of the microtubule decision, for some presumably good reason. (I'm tempted to email this as a "followup question".)

  (I'm generally open to the idea that all putatively-conscious actions are actually after-the-fact rationalizations, since it is a reasonable interpretation of current science, but that theory has problems passing the smell test. If consciousness is always fractions of a second late to the game, then what use is it? And if it is useless, then what is it doing there? If microtubules do contribute significant computation, perhaps even the dominant mode of computation, and especially if microtubule-based computations are the very hallmark of consciousness, then it seems to make sense that they are in fact the origin of conscious thought, in realtime.)

• It is admittedly hard to resist the temptation to jump to "personal connections" being possible through "quantum resonance", but this presentation makes this no more plausible to me, for two basic reasons. First, per Larry Niven, if such "psychic powers" (sort of) were real and useful, we probably wouldn't be arguing about them, we'd be using them after having developed them a long time ago. If this effect does exist, it is almost by definition uselessly small.

  Maybe there is some way you could sense if a brain was "on" or "off", even if the activity itself was impenetrable. That might at least give you the "something's wrong" feeling. That still requires a setup I find exceedingly unlikely, but can't disprove; on the one hand the signal would need lots of amplifying, on the other hand it is faintly possible that the human brain is the near-ideal device to perform such amplification. Also, per the talk, the "something's wrong" feeling ought to fire during any anesthesia procedure, as that equally shuts down the microtubules, and who's ever heard someone say "Well, I just knew he was dead, I don't know how, but it turns out that he was actually in surgery"? That's a weak argument on my part, but there's no data here I consider reliable. Although... an experiment based on testing whether subjects claiming to be so "connected" can correctly tell if the other has been anesthetized is somewhat more feasible than an experiment determining if they can tell if the other is dead...

• The final thing that sort of bothered me is that it remains unspecified what "consciousness" is. Towards the end he talks about "intensity" of consciousness, but this is also left unspecified.

  I say this bothered me out of completeness, but unlike the previous two issues, I doubt there's much to be done about this one. Even if Dr. Hameroff is 100% about the function of microtubules, we've still only located consciousness; in a way, we've even farther from understanding how it works than we were in a purely standard-neuron-model world; we're a lot closer to being able to understand 10^16 operations per second than 10^28, and if nothing else, closer to building 10^16 machines to study the situation in a controlled environment. And celluar automata are likely to be even less amenable to mathematical analysis than neurons, themselves not famous for their transparency. But I think this should still be pointed out; structurally, this talk is indeed about consciousness, but I think most of the interesting data is "merely" about how the brain functions and that doesn't translate as directly to "consciousness" as Dr. Hameroff would like.

I found this very stimulating, and I recommend it. Better than any ol' hour-long Nova repeat.