Slashdot Mirror
Ray Kurzweil Responds To PZ Myers
On Tuesday we discussed a scathing critique of Ray Kurzweil's understanding of the brain written by PZ Myers. Reader Amara notes that Kurzweil has now responded on his blog. Quoting: "Myers, who apparently based his second-hand comments on erroneous press reports (he wasn't at my talk), [claims] that my thesis is that we will reverse-engineer the brain from the genome. This is not at all what I said in my presentation to the Singularity Summit. I explicitly said that our quest to understand the principles of operation of the brain is based on many types of studies — from detailed molecular studies of individual neurons, to scans of neural connection patterns, to studies of the function of neural clusters, and many other approaches. I did not present studying the genome as even part of the strategy for reverse-engineering the brain."
Clearly, this dispute should be resolved by a poll.
This whole discussion reminds me way too much of the million partisan pundit sissy fights that rage endlessly on the internet. If I wanted to see two guys argue about what the other did or didnt say, I would gladly head over to DailyKos or BigJournalism and drown myself in their pedantry. This is slashdot; please save the inanity for the comments and at least give us stories that have meaning!
Myers may have been focused on the "reverse engineer from the genome" argument but really the main issue is whether Kurzweil is within a few orders of magnitude of guessing the right level of complexity necessary to simulate a brain. The gist of the Myers argument isn't so much about genomics and ontogeny as it is about the emergent complexity of inter-related systems and I think the real nugget there might be something like: "We could model a brain but that wouldn't mean we modeled a mind. To model a mind you need to model a great deal of the environment the mind lives in... and that is many many orders of magnitude more complex."
For the record: I hope Kurzweil is right but I rather doubt he is. I don't think he's wrong about how powerful machines will be in 2050 I think he may be wrong about whether those machines can simulate a mind well enough because I really wonder if the complexity of a mind is actually a superpolynomial problem due to the hyper connected-ness of a mind and its environment.
[signature]
Myers, who apparently based his second-hand comments on erroneous press reports (he wasn’t at my talk), goes on to claim that my thesis is that we will reverse-engineer the brain from the genome.
So put your speech up on your site, all I can find are videos from previous summits. TED seemingly posted videos as they happened and therefore we could openly debate them. Summits are great but not everyone has the time or resources to attend them. I would suggest you move towards a more open format of disseminating your ideas and the very specific and lengthy details about them. I'm not going to buy a book on futurism and wade through it for the details you provide about neurobiology and I don't think PZ Meyers would do that either.
I mentioned the genome in a completely different context. I presented a number of arguments as to why the design of the brain is not as complex as some theorists have advocated. This is to respond to the notion that it would require trillions of lines of code to create a comparable system. The argument from the amount of information in the genome is one of several such arguments. It is not a proposed strategy for accomplishing reverse-engineering. It is an argument from information theory, which Myers obviously does not understand.
Well, frankly, I don't understand it either. You're applying information theory to lines of code ... and that just doesn't make any sense to me. I haven't heard of it. I haven't heard of anyone say "theoretically could be reduced to x lines of code." I don't know why we're talking about information theory when we're talking about simulating the brain or even understanding the brain.
The amount of information in the genome (after lossless compression, which is feasible because of the massive redundancy in the genome) is about 50 million bytes (down from 800 million bytes in the uncompressed genome). It is true that the information in the genome goes through a complex route to create a brain, but the information in the genome constrains the amount of information in the brain prior to the brain’s interaction with its environment.
So first it was information theory on the genome and now you're on about compression of the genome. Great, you've applied theoretical limits to lines of code in order to describe a complex biological system and then argued that due to redundancy we can reduce it to 50 million bytes. And what did that buy us exactly? Look at how many lines of code we've devoted to simulating a single neuron or synapse ... and it's not even a complete and accurate simulation. Your theoretical limits are amusing but pointless ... to further apply your 'exponential growth' of the lines of code we can program is further amusing.
Kurzweil is a futurist with just enough knowledge to sell people. His exponential growth to a singularity and proof of it doesn't do him much good when he doesn't understand the complexity of the brain and then applies theoretical limits to that from other disciplines. He's free to keep preaching, I just question at what point people will give up on him. If he dies soon and pulls a L. Ron Hubbard what sort of cult then will we have on our hands?
My work here is dung.
We can easily do it within two decade: 2010-2019, and 3560-3569.
We don't have more than a rudimentary understanding of how the brain works, or even what consciousness is.
People say this a lot, and I don't understand why. Our understanding of how the brain works is a good deal more than rudimentary. The advances we've made in understanding the brain on both the large and small scales in just the last five years are breathtaking. Our understanding is a long way from complete, but Kurzweil is correct at least to the extent that our understanding is significant and appears to be growing at an accelerating rate. It may not be accelerating as fast as he expects, but keeping up with new developments in neurology at even a cursory level is quite challenging. The main difficulty we face at present in implementing the structures we do understand in silicon is the lack of adequate parallelism in current computing hardware, not our understanding of the relevant neural structures.
As for consciousness, unless you believe in some kind of pre-scientific vitalism, a reasonable working assumption is that it is an emergent property of brain-like structures. Unless and until we discover otherwise, there is no reason to wait for an understanding of consciousness to begin working on replicating the functionality of the brain. Quite likely, the attempt to replicate the brain will reveal more about consciousness than idle philosophical inquiries. Those so inclined might want to settle on a definition of consciousness before trying to figure out how it works.
Proud member of the Weirdo-American community.
The major flaw I can see in his response (which I think was addressed by Myers) is
but the information in the genome constrains the amount of information in the brain prior to the brain’s interaction with its environment.
He even underlined it. The problem is that the brain doesn't just spring into existence fully formed and THEN get exposed to the environment. The brain starts out as a few cells and is constantly exposed to the environment as it develops. I think this was a major point in Myers response and RK just blew right past it.
is right. Myers criticism may be off the mark but Kurzweil's speculation about brain design, like some much of his other speculation, is bullshit. His basic argument in the blog post is that the amount of information in the human genome constrains the amount of information (and the complexity) required to design the brain. This thesis is wrong on a bunch of levels but let's take the most obvious. The amount of information in the genome is the amount of information that the "body" (to simplify) requires to replicate or create parts of itself. The amount of information required is relative to the machinery which is going to interpret it. There is no reason to believe we are dealing with a Turing machine here where the amount of bits required for a program to perform a function is going to be more or less consistent across languages and platforms (assuming similar complexity of the code). The machine interpreting the bits matters. So while the body may only need "50 million bytes" to create itself we may need many, many more millions of bits to specify how to build it. Just consider the complexity of protein folding.
More dubious statements follow:
"The goal of reverse-engineering the brain is the same as for any other biological or nonbiological system – to understand its principles of operation. We can then implement these methods using other substrates other than a biochemical system that sends messages at speeds that are a million times slower than contemporary electronics. The goal of engineering is to leverage and focus the powers of principles of operation that are understood, just as we have leveraged the power of Bernoulli’s principle to create the entire world of aviation."
This completely begs the question of whether it can be replicated in another substrate. He just assumes that it can be done and by doing so he already assumes a model of the brain that could be (and is most likely) wrong. The brain is clearly not a Turing machine. That's not say it is not another kind of "computer" (for some expanded definition of computer) or follow mechanistic principles however. Assuming the brain is like a Turing machine (which Kurzweil implicitly does) is one of the biggest obstacles to developing real AI.
Speculation of Kurzweil kind does not belong in the "Science" category, maybe "Idle".
Kurzweil is absolutely correct. His best argument is not the complexity of the genome, but focusing on the actual functional structures in the brain. A cortex composed of a billion repeating units is something we CAN feasibly simulate. Already, we have massive systems that run an algorithm spread across billions of separate instances. (google.com is one)
An "algorithm" could also model the behavior of a few neurons working in circuit.
Also, keep in mind that most of the complexity of the brain and body are completely unrelated to the task of thinking. Much of that genome codes for molecular machine parts needed to maintain and grow the hardware. There's all kind of defense and circulatory and support systems that we won't have to worry about when designing artificial minds.
And finally, when you consider the changes made to the brain from the enviroment : that doesn't make the problem harder. Once you have a self organizing neural system that works like the human brain but a million times faster, you expose that system to our environment and train it up just like we do with humans. Sure, it might take a few years for such a system to reach super-intelligence, but if your fundamental design was right then this would eventually happen.
Myers primary complaint was that Kurzweil used the number of genes in the genome and how many bits would be required to store that data as a predictor of how long it will take to completely understand the complexity of the human mind. Myers' post lays out a glimpse of the additional complexity involved and rightly points out the fallacy of making such a grand prediction based on such a small amount of information and understanding. Of course Kurzweil's entire career and fame are now dependent on people continuing to fall for his dramatic generalizations and overreaching predictions that "Something Big" is right around the corner. I have watched Kurzweil talk and sometimes it seems as if he has a messianic complex.