More on Moore's Law
2/11/2003

Home

Future Index

Genetic Algorithms
    As I mentioned yesterday, (
2-10-2003_Technology_and _Investments) there was an article in this week's issue of "The Scientist" about genetic algorithms that are re-inventing existing devices (demonstrating that it doesn't take a human inventor to create them), and even inventing new, improved designs that haven't been considered before). So the question arises: who did invent them?
    This raises interesting philosophical questions. Presumably, whatever intelligence is involved must appear in the act of selection, since the programs make random or semi-random variations and then test them for efficacy. This hearkens back to the idea that what we would consider "information" lies in the eye of the beholder. There is as much information encoded in one set of random numbers as there is in another. It is we who select one pattern out of the myriad patterns that are possible. Of course, some patterns seem more unique to us than others. A pattern of all 1's or all 0's would seem significant to us. But since each pattern is unique, such a pattern is no more or less common than any other. So the selection criteria we choose are those which are meaningful to us.
Natural Selection
    Natural selection is the postulated basis for evolution, and it involves a set of selections that occur when selections are made sequentially in response to a relatively  simple set of criteria. It differs from monkeys on a typewriter, in that instead of generating all possible manuscripts, "warmer-or-colder" selections are made at countless steps along the way that lead it toward some end result.  Thus, the manuscript gradually "improves" .. i. e., trends toward the end result. There might have to be "learning" along the way, in the sense that the monkeys learn what is a word and what isn't a word--that is, develops a dictionary--and then learn rules of grammar. (This might be similar to the genetic code "learning" codons, and the way in which genetic machinery operates when it reproduces an organism. The genes may vary from organism to organism, and from individual to individual,  but the machinery operates in much the same way from organism to organism.)
    I guess this would be related to screening great numbers of compounds for efficacy against cancer, or some other disease.
Who Invents Inventions That Aren't Created by Inventors?
    To return to the question, "Who invents inventions that are invented by 'guided chance'"? 
    This seems analogous to chess-playing programs. Right now, they win through the evaluation of huge numbers of alternate moves and countermoves. Presumably, these chess-playing programs are being upgraded to think strategically as well as tactically... in other words, these programs are being given additional smarts.
    What's important about these chess tournaments is that they send a message that AI programs can rival humans in other spheres of endeavor. So where do we stand right now?
    One potential follow-on might be a program that understands physical systems, and invents based upon purposeful design, combined with some random variations.
    I don't know. Developments like "Deep Junior" don't leave the back room until they're ready for public display. AI has such an unfulfilled history that it's hard to know where it stands. It will be interesting to see how rapidly AI moves forward. But this kind of incremental improvement in specific subject areas shouldn't require breakthroughs in AI concepts.
    Perhaps the really big breaks will come in computer-based AI systems.
    Time will tell. 


More on Moore's Law
    Gordon Moore is going on record with the prediction that Moore's Law will prevail for at least 10 more years, taking conventional silicon circuitry down at least to the 32-nanometer level that I project for 2009. (Intel is officially forecasting 90-nanometer circuitry this year and 65-nanometer features in 2005.... see the linked article below) From there, I anticipate 45-nanometer feature in 2007, and 32-nanometer devices in 2009. But time will tell. My halving-every-other-year schedule calls for 22 nanometers in 2011, 16 nanometers in 2013, 11 nanometers in 2015, and 8 nanometers in 2017, but that's a pavement-pounding schedule, and it may not happen that fast.
    Intel is now talking 10-to-20 GHz processors in 2010 (Intel unveils ‘building blocks’ for 10-GHz processors - Silicon Strategies), backing away from their previous 2005 introduction of 10-GHz processors. Intel is continuing to predict 1,000,000,000 transistors on a chip by 2005, leading to, perhaps, 5,000,000,000 transistors on a chip by 2010. However, Intel is still standing by their plans for one-teraflops chips by 2010. Intel also presented a paper on a 5 GHz floating point multiplier-adder-accumulator.
    It will certainly be interesting to see what IBM, Sony, and Toshiba will deliver in "The Cell".
A Low-powered, Low-cost, 50-Gigops Image Recognition Chip
    Nippon Electric Corporation (NEC) has just unveiled an image recognition chip (NEC image processor runs 50.2 giga-operations per second - El. Engr. Times) that cranks out 50.2 gigops. It utilizes 128 8-bit parallel processors, that run at 100 MHz. (This implies that each processor churns out about 4 operations during every clock cycle.) It uses  a few watts of power, and is expected to be ready next year. (It's interesting to note that Matrox Corporation had a seven-board image processor capable of 100 gigops back in 1997. I don't imagine it was cheap.)
    The article makes the point that this chip is about four times as fast as a 3-GHz Pentium 4, implying that the Pentium 4 delivers 12 gigops, and is about four times as fast as its clock speed.
    Something like this might possibly hugely reduce the processing costs for Dr. Moravec's visual navigation system, although  I would imagine that other components, such as RAM and disk memories, would have to be added to a visual navigation system to arrive at realistic costs and configurations.

    Intel is allegedly planning a continuation of Moore's Law through the next 15 years.

    


     
    

 



    Now it's only a matter of time....
    (Gordon Moore today reaffirmed Moore's Law for at least 10 more years.
Moore's Law to roll on for another decade - Business Week , Gordon Moore makes predication about Moore’s Law - Silicon Strategies  , Gordon Moore Sees Another Decade for Moore's Law- Yahoo )

Scientists Replace Stem Cell Genes... - Washington Post   This article suggests that "The work is a step toward the biomedical goal of being able to rebuild or regenerate parts of the human body by transplanting either stem cells or tissues grown from stem cells into patients, scientists said. Precise genetic changes in those formative human cells might enhance their therapeutic potential or make them more compatible with patients' immune systems." The article observes that this discovery might make it unnecessary to pursue therapeutic cloning in which "in which cloned embryos would be created as a source of therapeutic tissues that match the genetic signature of the patient." B
ut the article goes on to say, "Despite years of efforts, Johns Hopkins University stem cell researcher John Gearhart said, scientists have not been able to make lines of mouse stem cells that are compatible in all mice. He suggested that therapeutic cloning research, controversial though it is, will remain important in the search for treatments for human disease."
    A part of this article that concerns me is that it inserts a gratuitous warning against "designer babies".
    The politicization of science concerns me. One has to ask, what will the politics of genetic choice become? What if we had said,
    "Horseless carriages will be terrible! They'll go at great speeds, risking the lives and limbs of their occupants, and of innocent victims walking or riding their horses along the road. They'll churn our roads into mud, and then get stuck in the mud, blocking buggy traffic. They'll fill the air with the stench of their petroleum fumes. They'll be a lot louder than the cloppings of horses' feet. Anyone who goes more than 20 miles an hour in them will die. They'll be affordable only by the rich.
    "We'd better ban horseless carriage research and manufacture, and stick with good old Dobbin."
    If you want an idea about how accurately futurists can prognosticate, read the 1950's and 1960's science fiction forecasts for the year 2000.
    What I suspect will happen s that designer babies will be prohibited in the United States for a while. But gradually, wealthy and/or influential grandparents will help their children design the baby of their choice at a clinic somewhere in the world that is most permissive about such things. After a few years, word will seep out that so-and-so has bent the rules, and then, others will follow suit. People will work their way through loopholes. Eventually, the practice will spread. Pharmaceutical companies will want to share in the pie. Countries that are rigorous in their maintenance of genetic rules will lose business (and reputation) to countries that are more relaxed about such practices.
    Or alternatively, artificial intelligence will threaten human functions, and everyone will hustle to keep humans up with machines.

The $100,000,000,000-a-Year Scam
    Tonight, there was a Merrill lynch ad on TV. You've probably seen it. It's written in invisible ink that he's making money in the stock market, thanks to Merrill Lynch, and that she's providing the homey, feminine touches as a smiling response to his stock market returns. It's a compelling ad.
    I remembered that a few months ago, Merrill Lynch was fined some millions of dollars for leading their clients over the edge of a cliff during the dot.com boom. Like B'rer Rabbit in the briar patch, Merrill Lynch probably didn't feel such a small pinch. They must have made billions off us suckers during the latter nineties.
   Last month, I took Taylor to Target, and while I was waiting for him, I entered into a conversation with an online broker in the store. I mentioned to him that the stock market only rises by about 1% a year, and that most of the returns derive from dividends, which, during the dot.com boom, shriveled to a little more than 1%.  He said he realized that, and that it was for that reason he liked Philip Morris, which is now yielding a 10% dividend yield.
    I'm shocked by the fact that he's aware of this situation in the stock market, and I'm wondering where he found out about it. It certainly isn't in any financial literature I've ever read.
    I can see why Warren Buffett considers the financial services industry to be a $100,000,000,000-a-year con!
   
    The Dow Industrials in September, 1929 hit a high of 386 The consumer price index in 1929 stood at 17.1; today it measures 184; or a factor of 10.76 above what it registered in 1929. Correcting the 1929 Dow Industrials peak for inflation, the DJIA becomes 4,153 in 2003 dollars.

    The Dow Jones Industrial Average closed today at 7,920. That means that the Dow index has less than doubled over the last 73+ years! That's a capital gains growth rate of about 0.9% per year. And yet, in early 2000, it hit 11,8005. And it was a glorious roller coaster ride all the way up! But what goes up must come down, and down it has come, back to its historical trend line. And most us haven't done as well as the DJIA.
    The other, and only significant source of investment returns is dividend yields. Over the years, they've averaged 3.5% to 4% but lately, they've been abysmal!
    So how can you make money in the stock market?
    You can't make money by timing the market.
    You can't make money by out-guessing the experts.
    You can't make it by thinking that you're somehow smarter or savvier or more insightful than the crowd. The crowd is smarter than you think. You are the crowd.
    You can try to identify promising local companies that lie below the radar screens of funds and stock brokers, and play venture capitalist. 
    You can invest in high-dividend companies when the market is down, and then plan to hold them.
    You can invest in stocks when interest rates are low (as they are now), and in bonds when interest rates are high.
    What do you do when everybody else is cleaning up in the market and you're sitting there clipping dividend coupons?
    You can rest assured that what goes up must come down, and there's very little chance they'll cash in their chips at the right time to make money.
    Or so I suppose. The fact is that I haven't followed this advice myself. If I had, I might be richer than I am.

    How do we convey this knowledge to the public?