One TRILLION human brain equivalents added as early as 2033! It’s pretty ridiculous for anyone to try to figure out what will happen when that amount of calculating power is added every year.

But will it run on Linux?

One thing I’ve never done myself is to take the data I retrieved from the Internet on the number of personal computers built each year, and combine it with Ray Kurzweil’s data and predictions for the number of calculations per second per $1000.

I did that during lunch today, and posted the results on my own blog. Specifically, I calculated an “Annual addition of computer power, in Human Brain Equivalents (HBEs),” using the data on production I got from the Internet, Kurzweil’s data and predictions for calculations per second per $1000, and Kurzweil’s estimate of the calculating speed of a human brain (2 x 10^16 calculations per second):

Annual addition of computer power…or why economic growth will be spectacular

The results are in Figure 1 on my blog post, but can be summarized as follows:

1) Only ONE human brain equivalent (HBE) added in 1995,

2) One million HBEs added in 2015,

3) One billion HBEs added in 2025,

4) One TRILLION HBEs added in 2033, and…

5) One SEPTILLION (i.e. 1 x 10^24, or 1,000,000,000,000,000,000,000,000) HBEs added in 2057.

One TRILLION human brain equivalents added as early as 2033! It’s pretty ridiculous for anyone to try to figure out what will happen when that amount of calculating power is added every year. But it should be pretty easy to hit 5% per year per-capita GDP growth. After all, we’ve been able to get above 3% per year the last couple of years…and we’re currently only adding about 1000 human brain equivalents per year.

It is a rather poor analogy as the underlying theory for both was established prior to attempting those projects. IOW, they already had a pretty good idea of how to get from point A (theory) to point B (engineering realization). This is not the case for computer technology mimicking human intelligence.

That was NOT the analogy I was making. The analogy I was making was between the Manhattan and Apollo projects versus space elevators and hydrogen-boron fusion.

The Manhattan and Apollo projects had a large number of human brains dedicated to bringing them to fruition. At present, space elevators and hydrogen-boron fusion have relatively few human brain equivalents dedicated to them. But circa 2049, if a computer with the calculating capacity of all the biological human brains on earth can be purchased for $1000, then essentially trillions of human brain equivalents can be brought to bear on the problems related to development of space elevators and hydrogen-boron fusion. (That’s in the somewhat-unlikely event that space elevators and hydrogen-boron fusion aren’t *already* commercially available in 2049.)

The doubling, referred to a Moors(sic) Law, is clearly coming to an end.

As Ray Kurzweil notes, the specifics of Moore’s Law is not a particularly good way of tracking progress in computers. A better way is the number of calculations per second that can be performed by $1000 worth of computer.

As Ray Kurzweil points out, the number of calculations per second that can be performed by $1000 worth of computer is increasing not merely exponentially (which would be a doubling every X number of years) but “double exponentially” (i.e., the doubling period is actually getting shorter and shorter).

If you go to the curve titled, “Exponential Growth of Computing” (under the heading, “The Exponential Growth of Computation Revisited,” you’ll see that Kurzweil predicts not just exponential growth, which would be a straight line on the exponential graph, but “double exponential” growth, in which the doubling time shrinks (rather than remaining the same, as would occur with “only” exponential growth).

Ray Kurzweil on the “Law of Accelerating Returns.”

But the answer to the question Willis asked about a hypothetical worldwide factor of 100 increase in computer speed is still the the same. I think such an increase would have no discernable effect on the annual per-capita GDP increase. This can be determined from the fact that I predict that a factor of 1 million increase from 2010 to 2030 coincides*** with “only” an increase from 3% per year to 4.5% per year.

***NOTE: The word “coincides” is much more appropriate than “causes.”. As I’ve noted before, the increases in per-capita GDP growth rate that I project for the 21st century include increases expected from ALL technology gains, e.g., increased computer speed, increased hard drive (or holographic) memory, increased wired/wireless data transfer rates, etc. etc.

A good analogy might be the Manhattan and Apollo projects.

It is a rather poor analogy as the underlying theory for both was established prior to attempting those projects. IOW, they already had a pretty good idea of how to get from point A (theory) to point B (engineering realization). This is not the case for computer technology mimicking human intelligence. There is no underlying theory that explains how intelligence works and therefore no point A as a starting point.

RE:27

But the capability of personal computers is doubling every…let’s say 2-5 years. And 10 doubling periods is a factor of 1000. Therefore, even with the capability doubling every 5 years, a personal computer will have the capability of a human brain circa 2050…and will be 1000 times more powerful than a human brain circa 2100.

The doubling, referred to a Moors Law, is clearly coming to an end. The doubling in speed is proportional to our ability to shrink the feature size. What your assuming is that computers will be able to increase speed by 1000 from a gigahertz to a terahertz. Computers based on silicon will never reach clock speeds in the terahertz region (10^12). This would require feature sizes on the order of one atom, which is clearly impossible due to numerous other physical restraints. We might get a couple of orders of magnitude increase, but that’s it, were done. To get any faster were going to need a new technology that has not been invented yet. This brings us right back to going from point A to point B without point A.

So in 2025 computers will be doing the same number of calculations per second as the human brain … so what? You seem to think that is important. It is not.

To see why, try redoing your same calculations for an elephant. An elephant has a much larger brain than a human today, so it must be doing more calculations per second, and yet somehow I don’t notice too many elephants writing novels and winning poetry contests …

Oh, brother. This is the logical equivalent of Swiss cheese. Here are just SOME of the holes in this “logic”:

1) Writing novels and winning poetry contests isn’t even being discussed. What’s being discussed is world GDP growth rate.

2) Elephants don’t contribute much to GDP growth rate because (among many reasons):

a) There are very few of them around (only small numbers in Africa and India),

b) The very few elephants in Africa and India don’t contribute much to world GDP growth rate for the same reason that the very many humans in Africa and India don’t contribute much to world GDP growth rate…the various governments’ policies are so screwed up that the elephants and humans spend all their time just trying to survive.

3) The fact that a brain is large does NOT necessarily mean it is conducting more calculations per second…there could be larger neurons, or fewer connections, or slower connections.

4) Many of the calculations of an elephant’s brain are probably dedicated to using the muscles to be able to control its amazing nose, or being able to very sensitively sense the earth’s tremors through its feet, or sensing sounds with its large ears, or any other of its bodily functions that are much improved over humans.

Suppose you invented a magic beneficial computer virus that would run on any computer, and its only effect was to instantly make that computer work one hundred times faster. You release the virus, and POOF … within a week, virus goes wild, you’ve tripled the speed of every computer in the world, you get awards and go on talk shows …

How much would that tripling of computer speed affect the GDP?

First off, 100 times is not “tripling”…100 times is 100 times.

But to answer your question, simply go to my predictions for GDP growth in the 21st century. From 2000 to 2020, I predict 3% per year. For 2020 to 2040, I predict 4.5 percent per year. The midpoint of those values is the year 2010 and 2030. Per Ray Kurzweil, the improvement in calculations per second per $1000 from 2010 to 2030 will be approximately a factor of 1 QUADRILLION.

So obviously I don’t think an overnight increase by a factor of 100 will have any impact on the economic growth rate.

As I pointed out, none has even passed the Turing Test, and to do that, a computer doesn’t have to actually be intelligent, it only has to mimic intelligence.

The Turing Test does NOT require a computer to mimic “intelligence.” Passing the Turing Test requires a computer to mimic HUMANITY. There is a huge difference.

The Turing Test allows the questioner to ask a large number of questions that have absolutely nothing to do with intelligence, but have everything to do with human experience. For example, the Turing Test questioner may ask the computer about the computer’s experiences in kindergarten or high school. But the computer has no experience in kindergarten or high school. That has nothing to do with intelligence, it has to do with humanity.

In fact, the Turing Test was apparently inspired by Turing’s familiarity with a game where males and females pretend to be of the others’ sex, and questioners (receiving typed responses in another room) attempt to guess whether the respondent is truly of the sex being represented:

http://en.wikipedia.org/wiki/Turing_test

By the way…we have IQ tests that are used to measure human intelligence. If a computer took such a test, and scored a 140+, would you accept that it was “intelligent?” If not, why not?

I agree with you that when the day comes that we can “put millions of minds to work” on a problem, we will solve problems more quickly. Your naive faith that that day is just around the corner is touching, but does not accord well with the history of computers.

Heh, heh, heh! My “naàƒÆ’à‚⮶e faith,” eh? Your condescension aside, let’s get the facts straight: My reference to putting “millions of minds to work” referred to the 2025 to 2050 time frame. Your characterization of the period from 20 to 45 years from now as “just around the corner” “does not accord well with” most humans’ conception of time. If you have to so blatantly misrepresent my position, is it perhaps because you can’t otherwise refute it?

Further, if anyone has made any statement of “naàƒÆ’à‚⮶e faith,” it is YOUR statement of October 25: “So in 2025 computers will be doing the same number of calculations per second as the human brain … so what? You seem to think that is important. It is not.”

Considering that not even ONE multi-million-dollar supercomputer can currently perform 20 quadrillion calculations per second, your assertion that it is not “important” if $1000 could buy an equivalent computer in 2025 sure seems like a statement of “naive faith” to me.

I have quoted Ray Kurzweil’s predictions regarding the cost and number of calculations per second for computers:

a) One human brain capability (2 x 10^16 cps) for $1000 around the year 2023.

b) One human brain capability for one cent around the year 2037.

c) Entire biological human race capability (2 x 10^26 cps) for $1000 around the year 2049.

d) Entire biological human race capability for one cent around the year 2059.

Do you (and Ian Castles) agree with those calculation speed/cost predictions, or do you disagree? And if you disagree, by what amount do you disagree (e.g. a factor of 1,000 slower speed, or a factor of 1,000,000, or what)?

Per my Long Bets #194, some approximate average annual per-capita GDP increases for the 2023, 2037, 2049, and 2059 time frames would be:

a) 2023: 3.5%,

b) 2037: 5.5%

c) 2049: 7.5%

d) 2059: 9.5%

Obviously, you and Ian Castles don’t agree with those predictions. But neither of you has answered my question: If one could indeed purchase a computer that can do the same number of calculations per second as the entire biological human race for $1000 in 2049, wouldn’t you both expect that to dramatically improve economic growth? If not, why not?

While I’m at it, let me add that, for example, cuurent disk drives are immensely smarter, than old disk drives, about squeezing more date within a given amount of disk surface.

Generally, recent computer technologies have become smarter about distinguishing signals from noise, and using smaller, and faster, signals than older computer technologies.

In some ways, computers have gotten smarter, largely in service of getting faster, and/or more accurate.

They are much smarter about how to send, and receive, signals over a piece of copper wire called a telephone line. They are smarter about detecting, and often correcting, internal errors.

The addition of cache memory make a computer smarter about getting pertinent data to the cpu faster.

Such smarts can be important to the functioning of computers, but they would seem relatively trivial as compared to the kinds of intelligence you may have in mind in your discussion.

Now this is an increase of hundreds of thousands of times the computer power. But has my productivity increased by a factor or hundreds of thousands? Not.

To get the results you speak of will require, not just an increase in power and speed, but in intelligence. How much more intelligent is my computer today than my computer in 1963?

Not a bit. Not a jot. Not a tittle. Not a millismart, not a microalbert. (one microalbert is one millionth of the intelligence of Albert Einstein.

Yes, when computers get intelligent we will be able to, as you say, “put millions of minds to work” on a problem. But computers are not intelligent, and show no signs of becoming so. As I pointed out, none has even passed the Turing Test, and to do that, a computer doesn’t have to actually be intelligent, it only has to mimic intelligence.

I agree with you that when the day comes that we can “put millions of minds to work” on a problem, we will solve problems more quickly. Your naive faith that that day is just around the corner is touching, but does not accord well with the history of computers.

Historically, computers have gotten larger very quickly. And computers have gotten faster very quickly.

But they have gotten smarter very, very, very slowly, if at all. My Macintosh is not any smarter than the 1963 Algol I started on. Not even the slightest bit smarter. It’s a whole lot faster and larger, but it’s still dumb as a post, just like the first one.

Now you claim this will all change, and computers will suddenly become intelligent. You say you know this because they are still getting faster and bigger.

But the last fifty years have already proven that faster and bigger computers does not mean smarter computers … so since you are claiming that this will suddenly change, you need to show some proof, or at least some theory, that explains how we’ll get from no useful progress at all on intelligent computers for the last 50 years, to having an independent machine intelligence, in only a decade or two.

w.