Thursday, February 15, 2007

The Future Will Be So Fine

Intel Corp. said it has successfully built a computer chip that could shatter processing-speed records. The last big leap came when chipmakers fit more than one processing engine -- or cores -- onto a single chip. Intel said its experimental chip, which could be available commercially within five years, has 80 cores. It can process a trillion calculations a second using no more electricity than a light bulb. (BusinessWeek.com) In 1996, a supercomputer that powerful at Sandia National Laboratories took up 2,000 square feet, used 10,000 Pentium Pro processors, and consumed 500 kilowatts of electricity. "This is significant," said Jim McGregor of market researcher In-Stat. (AP in Yahoo! News)

From Harold Maass' The Best of Today's Business


This type of thing is why I make extremely optimistic forecasts about the future. (Although even I pale before Ray "The Singularity Is Near" Kurzweil's optimism. His site is here). If the time-frame to market pans out, we'll have had a house-sized computer, using 500 Kw of juice, reduced in 15 years to A SINGLE CHIP, using 100 watts.

Kurzweil makes the same point here about biological science:

"It took us 15 years to sequence HIV. We sequenced SARS in 31 days. We’ll soon be able to sequence a virus in just a few days’ time. We're basically doubling the power of these technologies every year."

We are made demi-gods.

I know, I know, "But what about the flying cars, huh? We was promised flying cars, and that never panned out, so what makes you think that this stuff will be more than interesting trivia?"
In point of fact, we've had practical flying cars since the 70s. The problem is, you need a pilot's license to operate them, so the masses never took to 'em. Which may be a bit of a dodge on technical grounds, but I think that one has to admit that the "flying car" promise was at least half-fulfilled.

So what does Ray think we'll do with all the computing power? :

Consider that the price-performance of computation has grown at a super-exponential rate for over a century. The doubling time (of computes per dollar) was three years in 1900 and two years in the middle of the 20th century; and price-performance is now doubling each year. This progression has been remarkably smooth and predictable through five paradigms of computing substrate: electromechanical calculators, relay-based computers, vacuum tubes, transistors, and now several decades of Moore's Law (which is based on shrinking the size of key features on a flat integrated circuit). The sixth paradigm--three-dimensional molecular computing--is already beginning to work and is waiting in the wings. We see similar smooth exponential progressions in every other aspect of information technology, a phenomenon I call the law of accelerating returns. [...]

Consider the following: As with all of the other manifestations of information technology, we are also making exponential gains in reverse-engineering the human brain. The spatial resolution in 3D volume of in-vivo brain scanning is doubling each year, and the latest generation of scanners is capable of imaging individual interneuronal connections and seeing them interact in real time. For the first time, we can see the brain create our thoughts, and also see our thoughts create our brain (that is, we create new spines and synapses as we learn). The amount of data we are gathering about the brain is doubling each year, and we are showing that we can turn this data into working models and simulations.

Already, about 20 regions of the human brain have been modeled and simulated. We can then apply tests to the simulations and compare these results to the performance of the actual human brain regions. These tests have had impressive results, including one of a simulation of the cerebellum, the region responsible for physical skill, and which comprises about half of the neurons in the brain. I make the case in my book (The Singularity is Near) that we will have models and simulations of all several hundred regions, including the cerebral cortex, within 20 years. Already, IBM is building a detailed simulation of a substantial portion of the cerebral cortex. The result of this activity will be greater insight into ourselves, as well as a dramatic expansion of the AI tool kit to incorporate all of the methods of human intelligence.

By 2029, sufficient computation to simulate the entire human brain, which I estimate at about 1016 (10 million billion) calculations per second (cps), will cost about a dollar. [Emph. add.] By that time, intelligent machines will combine the subtle and supple skills that humans now excel in (essentially our powers of pattern recognition) with ways in which machines are already superior, such as remembering trillions of facts accurately, searching quickly through vast databases, and downloading skills and knowledge.

But this will not be an alien invasion of intelligent machines. It will be an expression of our own civilization, as we have always used our technology to extend our physical and mental reach. We will merge with this technology by sending intelligent nanobots (blood-cell-sized computerized robots) into our brains through the capillaries to intimately interact with our biological neurons. If this scenario sounds very futuristic, I would point out that we already have blood-cell-sized devices that are performing sophisticated therapeutic functions in animals, such as curing Type I diabetes and identifying and destroying cancer cells. We already have a pea-sized device approved for human use that can be placed in patients' brains to replace the biological neurons destroyed by Parkinson's disease, the latest generation of which allows you to download new software to your neural implant from outside the patient.

If you consider what machines are already capable of, and apply a billion-fold increase in price-performance and capacity of computational technology over the next quarter century (while at the same time we shrink the key features of both electronic and mechanical technology by a factor of 100,000), you will get some idea of what will be feasible in 25 years.

By the mid-2040s, the nonbiological portion of the intelligence of our human-machine civilization will be about a billion times greater than the biological portion (we have about 10^26 cps among all human brains today; nonbiological intelligence in 2045 will provide about 10^35 cps). Keep in mind that, as this happens, our civilization will be become capable of performing more ambitious engineering projects. One of these projects will be to keep this exponential growth of computation going. Another will be to continually redesign the source code of our own intelligence. We cannot easily redesign human intelligence today, given that our biological intelligence is largely hard-wired. But our future--largely nonbiological--intelligence will be able to apply its own intelligence to redesign its own algorithms...


I don't buy the "by the mid-2040s" timeframe, but that's just caution based on my experience of how human events typically go. I have no actual rationale for why it probably won't be so.

But more importantly, I think that Kurzweil resembles those who, in an earlier era, thought that television would be the perfect tool for teaching the masses. While it certainly has delivered on that promise, what it teaches generally isn't all that progressive or positive.
There are many, many opportunities for people to learn actual useful and/or highly sophisticated stuff from television shows, but the people who participate in "educational TV" are a self-selecting, small minority.

So my long-standing guess is that the majority of people will use this new-found computing power for entertainment, by living most or all of their lives in virtual universes of their own creation.

But those who don't eat of the lotus will have mind-boggling tools to work with, to shape the objective universe - or at least our own solar system.

On a personal note, I'd like to think that I'd be among the world-shapers, but I suspect that I'll spend most of my time on the holo-deck. Snap.

10 Comments:

Blogger EVadvocate said...

Kurzweil resembles those who, in an earlier era, thought that television would be the perfect tool for teaching the masses. While it certainly has delivered on that promise, what it teaches generally isn't all that progressive or positive.

Like all technological innovations (e.g., telephone, camera, moving pictures, television, VCR, internet, etc.) this new and awesome computing power will no doubt be used for pornography.

February 15, 2007 7:35 AM  
Blogger Harry Eagar said...

What's wrong with pornography?

I don't see any great benefits from better computers. We don't use the ones we have now.

Don't they have computers in Detroit? We already agreed yesterday that Detroit is run by idiots.

What about climate modeling? When the Maui High Performance Computing Center opened 13 years ago, one of its tasks (called 'grand challenges') was to model climate. Last month, MHPCC inaugurated 'Jaws,' its latest array, which is 15,000 times more capable than the original.

Are our climate models 15,000 times better than they were in 1993? More to the point, is the public policy discussion of climate any better than it was in 1993?

It isn't the case that the people whose only interest in computing power is a faster xBox are the ones who are failing to take advantage of the computer. The ones who think they are so smart are really dumb, too.

And, if I am right about Islam being the only real challenge facing the United States, what good is a faster computer for that?

February 15, 2007 7:57 AM  
Blogger David said...

Why do these nirvanas always sound like Hell to me?

I have no doubt that it is possible to build a brain/computer interface so that we can all walk around Googling gigantic information caches. I'd even look forward to that. But the first use will undoubtedly be for pornography. It's easy to imagine men and women linking their interfaces together and having virtual sex while lying next to each other. I can't say I find the prospect all that attractive. (Also, I'm not entirely convinced that the actual tactile mechanics are as disposable as this scenario presupposes.)

By the way, I never agreed that the people who run Detroit are idiots. That would be entirely too convenient.

February 15, 2007 11:12 AM  
Blogger Bret said...

"By 2029, sufficient computation to simulate the entire human brain, which I estimate at about 1016 (10 million billion) calculations per second (cps), will cost about a dollar."

Kurweil and Moravec (Robot: Mere Machine to Transcendent Mind) both think that you'll be able to download your brain scan and have yourself simulated. Maybe so, but I'd bet the effect would be, without the hormonal and other physical aspects of the brain, to simulate you while in a coma.

harry eagar wrote: "I don't see any great benefits from better computers."

Really, you don't like Googling? Don't like the Internet? Don't like blogging? Don't like digital cell phones? None of these things were feasible 20 years ago.

Working on robots, I can easily use those teraflops. You'll like your personal assistant in 20 years that rescues you from being stuck in your bathtub because one day you're too feeble to get out of it anymore (happens with alarming frequency). When you're too old to drive, you'll probably quite grateful for the cars that drive themselves so you can get out the house without having to rely on your busy children. I suspect that you'll end up using all those teraflops and more without even realizing it. The smarter computers become the less they'll seem like computers.

david wrote: "I can't say I find the prospect all that attractive."

Then do it the other way. Might be fun for a change though.

February 15, 2007 12:05 PM  
Blogger Harry Eagar said...

Google, the Internet and cell phones have not made a dent in solving any problem I care about.

Some people think we will be saved through art, others religion, other silicon.

It won't be silicon, anyway. That couldn't even save Nicole what's her name?

February 15, 2007 6:55 PM  
Blogger Oroborous said...

And, if I am right about Islam being the only real challenge facing the United States, what good is a faster computer for that?

Well, for starters it'll help solve the primary problem you've pointed out with remote/autonomous warcraft - that of choosing targets.

As long as we get cars that transform into robots, I will be looking forward to this.

For that, I think that you'll also need a pocket fusion reactor. Those, sad to say, are a little more than five years off.

February 16, 2007 2:41 AM  
Blogger Harry Eagar said...

I have suggested several less drastic, incomputable ways of approaching the problem.

All the really interesting questions are nonlinear. That's why they're interesting.

February 16, 2007 7:59 AM  
Blogger Unknown said...

Flying cars were always a stupid idea. Just think of the air traffic control problems if every driver today had a flying car and could take any route he wished to go from point a to point b. Just think of how much the fatality rate would climb when even minor fender-benders involve a precipitous plunge to the ground. Plus all the additonal damage and injury suffered by those on the ground. Certain experiences just can't scale.

February 16, 2007 8:33 AM  
Blogger Mike Beversluis said...

I suspect Kurzweil is wrong about molecular electronics lying just around the corner - if they are, then so are fusion reactors.

Finding a way to make reliable electrical contact with single molecules hasn't been done. Much of the mol-e literature contradicts itself because often the measured result comes from the contacts.

February 16, 2007 4:29 PM  
Blogger Harry Eagar said...

There's also the problem that any two clean surfaces, of any materials, fuse on contact.

February 16, 2007 7:28 PM  

Post a Comment

<< Home