How to Build a Brain Wrong
Ray Kurzweil, the justly lauded inventor and machine intelligence pioneer, has been predicting that humans will eventually upload their minds into computers for so long that I think his original audience wondered whether a computer was a type of fancy abacus. It simply isn’t news for him to say it anymore, and since nothing substantive has happened recently to make that goal any more imminent, there’s just no good excuse for Wired to still be running articles like this:
Reverse-engineering the human brain so we can simulate it using computers may be just two decades away, says Ray Kurzweil, artificial intelligence expert and author of the best-selling book The Singularity is Near.
It would be the first step toward creating machines that are more powerful than the human brain. These supercomputers could be networked into a cloud computing architecture to amplify their processing capabilities. Meanwhile, algorithms that power them could get more intelligent. Together these could create the ultimate machine that can help us handle the challenges of the future, says Kurzweil.
This article doesn’t explicitly refer to Kurzweil’s inclusion of uploading human consciousness into computers as part of his personal plan for achieving immortality. That’s good, because the idea has already been repeatedly and bloodily drubbed—by writer John Pavlus and by Glenn Zorpette, executive editor of IEEE Spectrum, to take just two recent examples. (Here are audio and a transcription of a conversation between Zorpette, writer John Horgan and Scientific American’s Steve Mirsky that further kicks the dog. And here’s a link to Spectrum‘s terrific 2008 special report that puts the idea of the Singularity in perspective.)
Instead, the Wired piece restricts itself to the technological challenge of building a computer capable of simulating a thinking, human brain. As usual, Kurzweil rationalizes this accomplishment by 2030 by pointing to exponential advances in technology, as famously embodied by Moore’s Law, and this bit of biological reductionism:
A supercomputer capable of running a software simulation of the human brain doesn’t exist yet. Researchers would require a machine with a computational capacity of at least 36.8 petaflops and a memory capacity of 3.2 petabytes ….
Sejnowski says he agrees with Kurzweil’s assessment that about a million lines of code may be enough to simulate the human brain.
Here’s how that math works, Kurzweil explains: The design of the brain is in the genome. The human genome has three billion base pairs or six billion bits, which is about 800 million bytes before compression, he says. Eliminating redundancies and applying loss-less compression, that information can be compressed into about 50 million bytes, according to Kurzweil.
About half of that is the brain, which comes down to 25 million bytes, or a million lines of code.
First, quantitative estimates of the information processing and storage capacities of the brain are all suspect for the simple reason that no one yet understands how nervous systems work. Science has detailed information about neural signaling, and technologies such as fMRI and optogenetics are yielding better information all the time about how the brain’s circuitry produces thoughts, memories and behaviors, but these still fall far short of telling us how brains do anything of interest. Models that treat neurons like transistors and action potentials like digital signals may be too deficient for the job.
But let’s stipulate that some numerical estimate is correct, because mental activities do have to come from physical processes somehow, and those can be quantified and modeled. What about Kurzweil’s premise that “The design of the brain is in the genome”?
In short, no. I was gearing up to explain why that statement is wrong, but then discovered that PZ Myers had done a far better job of refuting it than I could. Read it all for the full force of the rebuttal, but here’s a taste that captures the essence of what’s utterly off kilter:
It’s design is not encoded in the genome: what’s in the genome is a collection of molecular tools wrapped up in bits of conditional logic, the regulatory part of the genome, that makes cells responsive to interactions with a complex environment. The brain unfolds during development, by means of essential cell:cell interactions, of which we understand only a tiny fraction. The end result is a brain that is much, much more than simply the sum of the nucleotides that encode a few thousand proteins. [Kurzweil] has to simulate all of development from his codebase in order to generate a brain simulator, and he isn’t even aware of the magnitude of that problem.
We cannot derive the brain from the protein sequences underlying it; the sequences are insufficient, as well, because the nature of their expression is dependent on the environment and the history of a few hundred billion cells, each plugging along interdependently. We haven’t even solved the sequence-to-protein-folding problem, which is an essential first step to executing Kurzweil’s clueless algorithm. And we have absolutely no way to calculate in principle all the possible interactions and functions of a single protein with the tens of thousands of other proteins in the cell!
Lay Kurzweil’s error alongside the others at the feet of biology’s most flawed metaphor: that DNA is the blueprint for life.
What this episode ought to call into question for reporters and editors—and yet I doubt that it will—is how reliable or credible Kurzweil’s technological predictions are. Others have evaluated his track record in the past, but I’ll have more to say on it later. For now, in closing I’ll simply borrow this final barb from John Pavlus’s wonderfully named Guns and Elmo site (he’s also responsible for the Rapture of the Nerds image I used as an opener.
How to Make a Singularity
Step 1: “I wonder if brains are just like computers?”
Step 2: Add peta-thingies/giga-whatzits; say “Moore’s Law!” a lot at conferences
Step 3: ??????
Step 4: SINGULARITY!!!11!one
Added later (5:10 pm): I should note that Kurzweil acknowledges his numeric extrapolations of engineering capabilities omit that even “a perfect simulation of the human brain or cortex won’t do anything unless it is infused with knowledge and trained.” Translation: we’ll have the hardware, but we won’t necessarily have the software. And I guess his statement that “Our work on the brain and understanding the mind is at the cutting edge of the singularity” is his way of saying that creating the right software will be hard.
No doubt his admission is supposed to make me as a reader feel that Kurzweil is only being forthcoming and honest, but in fact it might be the most infuriating part of the article. Computers without the appropriate software might as well be snow globes. As a technologist, Kurzweil knows that better than most of us. So he should also know that neuroscientists’ still primitive understanding of how the brain solves problems, stores and recalls memories, generates consciousness or performs any of the other feats that make it interesting largely moots his how-fast-will-supercomputers-be argument. And yet he makes it anyway.