Wednesday, May 20, 2009

Why Worry About This Sci-Fi Stuff Now?



The term “mindclone” evokes a wide range of sci-fi images from the “Cylons” of Battlestar Galactica to the “Mr. Smiths” of The Matrix. While it is indisputable that we are creating large mindfiles, as described in Question 1, and surely there are geeks working hard on mindware, as reviewed in Question 2, how close could we be to an actual mindclone when computers can’t converse on their own much better than a two-year old kid?

Very close. Close enough to feel the bits and bytes of cyberbreath on our cheeks. To realize how close we are to cyberclone reality it is necessary to understand the exponential nature of advances in information technology.

Pattern recognition expert Ray Kurzweil has shown in his best-seller, The Age of Spiritual Machines, that information technology has been doubling its capabilities every one to two years since the early 1950s. For example, we have more computing power in our cellphones today, for about $200, than there was in the Apollo spacecraft that went to the moon in the 1960s and 1970s at a cost of $30 billion. While voice recognition technology was non-existent in the 1990s, only ten years later it was a free feature in most laptops and PDAs, and a ubiquitous customer service front end for large companies. Ray Kurzweil and others have shown that based upon the doubling rate of information technology it is reasonable to expect mindclones for about $1000 by the end of the 2020s, and sooner than that for a higher price.

The following chart compares the number of information processing circuits available at various dates with biological lifeforms that have the equivalent number of neurons. Up through 2010 we see computer programs no smarter than a mouse or a bug, and are not impressed. However, this is most misleading in two ways.


First, there are a great many software packages that are already far smarter than even very clever humans. For example, the mapping software in your cellphone or PDA can best any of us at finding addresses in unknown neighborhoods. Gaming software, from grandmaster chess to the complex Asian game of go, and including all manner of virtual world environments, far outstrip the conceptual capabilities of ordinary humans. Hence, while no software package has “put it all together” the way a human mind does, programs are popping into existence with great competence in many of the areas to which we devote our mental skills. How long could it be until some software “puts it all together?” Per Kurzweil’s chart, no longer than the 2020 timeframe in which computers will have as many processors as the human brain has neurons.

There is a second reason, born of psychology, why we find it hard to realize how near-term the mindclones are when all we see on television are robots with the brains of a rodent or a bug. This reason relates to the difference between our natural way of perceiving things, which is linear, and the way in which information technology is advancing, which is exponential. Linear things proceed the way children grow, perhaps half-a-foot or so a year until they reach a plateau height. This is the “linear” way we have evolved to perceive the relationship between changing things and the time it takes them to change forms. Things that change like people grow, linearly, change about the same amount each year or so. Hence, if there is a billion-fold deficit between the processing capability of a computer today, and that of a human mind, it is natural for us to project the arrival of mindclones as equal in years to about one billion divided by the increase in processor speed we can get on our computers from one year to the next.

For example, my one year-old computer has about 1/100,000th of the capability of a human mind (its processing speed is about that fraction of the number of human brain neural connections, although its software is in some areas pretty advanced). In other words, it has only .001% of the capability of a human mind. It’s a rodent. I could go buy a new computer today that has 2/100,000th or .002% of the capability of a human mind. At this rate, with the way my linear mind works, I would expect to be able to buy a mindclone in 99,998 more years. What, me worry! Our linear minds take our most recent experience – such as going from a 1/100,000th of a human mind computer to a 2/100,000th of a human mind computer in one year – and extrapolate it forward such that we think it will take 998 more years to get 1% of a human mind, another 1000 years to get to 2% of a human mind, another 1000 years to get to 3% of a human mind, and so on.

In fact, though, information technology does not grow linearly, but exponentially. This means, according to “Moore’s Law”, information technology doubles each 1-2 years – something very different from growing linearly. Because computer capability doubles it means next year I will get not 3/100,000th of a human brain computer, but 4/100,000th of one. Exponential growth means the year after that I will get not 5/100,000th of a human brain computer, but 8/100,000th of one. With information technology, I can expect to reach mindclone computing as rapidly as this:

Years From Now Fraction of a Mindclone
Next Year 4/100,000th
Year After 8,100,000th
Third Year 16/100,000th
Fourth Year 32/100,000th
Fifth Year 64/100,000th
Sixth Year 128/100,000th
Seventh Year 256/100,000th
Eighth Year 512/100,000th
Ninth Year 1000/100,000th
Tenth Year 2000/100,000th
Eleventh Year 4000/100,000th
Twelfth Year 8000/100,000th
Thirteenth Year 16,000/100,000th
Fourteenth Year 32,000/100,000th
Fifteenth Year 64,000/100,000th
Sixteenth Year 128,000/100,000th = MINDCLONE

Four clarifying comments are in order. First, the rounding down from 1,024 to 1,000 in the ninth year is just to make the arithmetic easier to follow. Second, while Moore’s Law says that the doubling occurs every 1-2 years, in the example given above I showed the doubling every year. The effect of making it every two years would simply be to postpone mindclones to 32 years from now instead of 16, or to 24 years from now if we use a doubling period of every 18 months. The important point is that mindclones are around the corner – not in some other millennium, or even in some other generation. This is about our lives.

The third clarifying comment is that some people question for how long Moore’s Law can continue, noting that other exponential phenomena – such as the growth of bacteria in a Petri dish – end when the room for growth runs out. In fact, because knowledge (unlike bacteria) can grow without limit, the doubling of information technology is not limited. Knowledge is the only resource that the more you exploit it, the more you have to exploit. Engineers have already designed the pathways for Moore’s Law to continue for many decades. For example, when technology limits are reached with flat integrated circuits computers will shift to three-dimensional integrated circuits.

The fourth comment is that it takes more than brain-level hardware to create a mind -- it requires human-level software. My premise is that the software will arrive no later than the hardware. I cannot prove this premise because there are no "lines of code" equivalent to the mind that one could show exponential progress against (more code is not necessarily better). While it is true that we will be able to use the computational power of our hardware to "reverse engineer" the human brain, there may be more rapid approaches born of intuitive software design.

Let me share an analogy that reveals the fallacy underlying those who believe the software challenge of reverse engineering the brain puts us centuries away from mindware. I contend that the quest for artificial consciousness is analogous to the quest for artificial flight. The two broad approaches to these quests can be called “structural replication” and “functional replication.” Humans tend to fixate on structural replication as it seems apparent that “if we could but flap wings like a bird then we could fly like a bird.” Similarly, “if we could but reverse engineer the human brain, then we could be conscious like a human brain.” While structuralists are probably correct, theirs is not the more rapid approach. Instead, we have functionally replicated flight using non-bird approaches such as propellers and jet engines married to largely fixed wings.

We will also be able to functionally replicate human consciousness using non-brain approaches such as massively cross-correlated relational databases, virtualization software and iteratively modifiable natural language-based programs. We can copy what a brain does (make a mind) rather than copy what a brain is, just as we have copied what a bird does (make a flight) rather than copy what a bird is. Of course brains do many things besides make minds, and birds do many things besides flying. But, those other things are beyond the scope of our mindware (and others’ aerodynamics) projects.

So, in summary, we delude ourselves that mindclones are in the distant future because our linear minds have great difficulty projecting exponential phenomena. In fact mindclones are as close to us in time as the birth of punk rock and Apple Computer. The very same revolution that:

➢ brought cellphones from almost no one’s hands to almost everyone’s hands in under 20 years, and
➢ brought the internet from a military toy to a universal joy in under 15 years,
➢ will bring mindclones from chatbot infancy to human simulacra adulthood in the time it will take to get a kid through school or complete a first professional career.

Not everything sci-fi is far-off. I have a wristwatch communicator (thanks to SkyTel) and daily videophone conversations (thanks to iChat). My Oakley sunglasses magically deliver stereo music and phone calls directly to my ears, without wires (thanks to Bluetooth). My eyes are fixed by lasers, my teeth are cleaned by ultrasound and my food is cooked by microwaves. Weren’t all these things supposed to be sci-fi? I don’t see any of them in classic movies or early TV shows.

Nothing in our society is advancing faster than software, and mindclones are simply that: one part mindfile software and one part mindware software. True, some good processors are needed to run that software, but Moore’s law is delivering those processors right on schedule. We need to figure out this mindclone thing right now because this is one part of the future, one aspect of sci-fi, that is banging on the front door.

There is plenty of time to dream about flying cars. There’s a century to wait for a family trip to Mars. But mindclones are in a whole different category. They are riding a wave of exponential information technology that is real, here and now. Climb outside your linear mind and check out the latest avatars. If you do so, if you think exponentially, then it is impossible to not see the first, costly mindclones in 10-20 years, and then a mass-market explosion of them in 20-30 years.

Wait, and get caught by surprise, or read on, and be poised for the prize. The future is an opportunity that benefits the prepared. These 100 answers will make you mindclone savvy. So prepped, your mindclone will help make you happy. What a deal! Read on!

No comments:

Post a Comment