Ray Kurzweil
The Power of Hierarchical Thinking
Thanks for that. I see many friends in the audience. My travel schedule is not always this busy. I am frustrated that I can't take in more of the conference. I am trying to go through 101 to present very quickly, so that we can have some dialog. Aubrey just left. Okay. I think in that 2075, I'd tell him that he's too conservative about longevity escape velocity. We dialog about that issue in our movie that opened up last night in Brackenridge. He's a little more conservative than I am.
Um, yeah. I did give a presentation to the SETI folks. They have this steady assumption based on the formula, a scientific formula for assessing how many intelligent radio-capable computational-capable civilization there are in a typical galaxy like the Milky Way, there's like 15 different variables, Brackenridge is actually 10,000 feet which seems to have effected my voice. There's 15 different variables, and orders of magnitude difference of opinion on what the likelihood is that a lifeform will become intelligent, will develop radio technology, destroy itself, etc.
I have two different reasonably analysis, that can be extrapolated to a million different civilizations in the Milky Way according to SETI. I wanted to point out that if you take the standard SETI assumption that there are thousands if not millions of advanced civilizations capable of radio and computational technology, they would be spread out across cosmological time. They might be ahead of us by tens, hundreds, thousands or millions of years. A civilization could be a million years ahead of us, they might be like us, but the cell phones might be a little smaller. There's a linear assumption of the law of accelerating returns.. from 1850 we've gone from ponies being the fastest way to send a message, to 4G, we've increased trillions fold over the last century, very smoothly, smooth exponential growth for a hundred years. It would only take a few centuries at most to go from primitive radio tech to galaxy-wide engineering. Maybe it would be harder to interpret their language than say Hebrew, but it would be hard not to notice galaxy-wide engineering. If you apply the law of accelerating returns to the standard SETI assumptions, there should be thousands of civilizations doing engineering beyond the galaxy.
I kind of came up with a soft conclusion. You can't conclude that there's not a civilization out there other than ours. Yes, maybe they destroy themselves, maybe there's the Star Trek ethical standard not to disrupt the civilization and such. There's a doubt that a million civilizations are doing galaxy wide engineering, and they are all following a ethics code to not interfere with us.
I think that by the anthropic principle, with this antiquated model of physics that allows evolution to take place, and that's a reflection on SETI. I did point out that I think the project is very important, even if it results in a negative result, and so far it has. Of course, they describe their work as looking at a very small slice of the frequency and as many other frequencies and places in the galaxy that we could be searching. None the less, it would be hard with even our primitive way of looking at the univerise today, it'd be hard not to notice them.
The negative finding is significant, even more significant. We are really in the league, with these exponential technologies, we have a greater responsibility for being a steward of not only this planet, but perhaps of intelligence in our galaxy and this universe. These are the points in the story, and earlier in the movie I have dialogues with Marvin Minsky, Aubrey de Grey, etc. about the law of accelerating returns.
These are all of the points I'd like to make. Are there any questions? (laughter)
Technology is getting smaller, getting faster, getting more powerful. When I was student down the street at MIT, I went there because it was so advanced at the time that it actually had a computer. It was tens of millions of dollars, took up an entire building. You had to take a class to get to it. The computer that I carry around, and it's a million times less expensive, it's a thousand times more powerful, it's a billion fold improvement in price-performance. This is not just wild conjecture about the future. As a result of the exponential growth of information tech, the adoption rate is getting faster, partly because we communicate much more quickly. I took at the 550th of University of Basal, which was founded just a few years after Gutenberg started his press. Oh yes, they had some, just a hundred years later. So things happen much more smoothly in those days, it needed 400 years for the Gutenberg press to reach everyone, tell phones was 50 years, cell phones was 7 years, wikipedia took 3 years, the paradigm shift rate is getting faster and faster.
30 steps exponentially gets you to a billion. The linear expectation is our intuition, that's hard-wired into our brain. People's intuition includes scientists, that change will continue at its current rate. Only one percent had been collected. Uh, the skeptics told you that this wouldn't work. Only one ten thousandth would be sequenced.. only 1% is inaccurate of the genome now. Now it's just a fraction, only a few more orders of magnitude to get to 100%. That's what happened. It's continuing to double every year. In the last 7 years of the project, it was finished a year ahead of schedule. When I said it doubled, that was not an approximation. It exactly increased by 10% every year. It's really incredible how these exponential trajectories really are.
We always use the greatest tools to create the next. Today we use computers to design next generation computers, and so on. That's one of the reasons why things are going more quickly. The same phenomenon is the same of everything, even biological evolution, which did not have humans guiding it. Human capability and adoption, so over the next 80 years. The first step, life itself, DNA took a billion years, and then biological evolution used it ever since, the Cambrian explosion, 10M, and then a few million years for cognitive functions, and then a few hundred thousand years for hominids, and then the key innovation - a neocortex, the size of a table map - to do hierarchical thinking at a sufficient level to do language and technology and tools. I want to talk about neocortices later. I want to do a book on how the mind works, and share a few reflections about that. The neocortex can take a whole bunch of symbols, call that an idea, and get a new symbol, and you can take that symbol and build it up into a pattern, and call that one a symbol. We have this nested hierarchy called patterns. The neocortex is a bunch of pattern recognizers organized in a hierarchy, a lower level pattern recognizer feeds into a higher one, and so on. Interestingly, it's the same module which recognizes patterns like the cross bar in the capital A, in printed letters, we can recognize it in different forms and shapes. The same pattern recognizer is in different levels to recognize irony and humor and sexual attraction, and uh, and sentiment. And so on. And, it's basically a billion of these, you have some idea now of how it would work, and the simulation itself. I'll come back to that.
The next stage, evolution adopting homo sapiens is really the next stage, and then human technological evolution. Fire, the tool, the wheel, tens of thousands of years. We're really at an infliction point now where things get faster and faster. This graph was critized, and I chose things that gfit on a straight line, and if it didn't, I didn't use it. What did the encyclopedias think the key events were? There were some disagreements, and not much happened a billion years ago, not much happened a thousand years ago, and so on.
The adoption rate is getting faster and faster. It took decades, about a century ago, with the telephone and TV, and now adopted and in just a few years time the cell phone was adopted. Three years ago, most people didn't use blogs, tweets, etc. In 2004, when on this campus, a couple of students, these books, you used to get printed books with thumbnail books with a picture of a freshmen, there was a glee club. If you had a blind date, you'd look up your blind date to see what she looked like. So someone said, hey, we can put this online. Students can add additional pictures, they can say who their friends were, they talked about this and then implemented it on a $1000 USD laptop and now six years later they have a $50B IPO. There's $150B/day- that's really transformed our access to knowledge.
We all know that exponentials don't take off forever. That's true for specific paradigms. In information tech, there's pressure to create the next paradigm. That's happened five times for computers already. So, we noticedf that the exponential growth of the price performance of computing was going on for decades between Moore's.. Moore's paradigm was not the first paradigm. This is not just Moore's law. Moore's law is going to end, and that's not the end. In the 1950s, they were shrinking the size of vacuum tubes - that was the third paradigm. Then they were shrinking the size of vacuum tubes, smaller and smaller into tiny vacuum tubes, to a point where they couldn't shrink it any more and couldn't keep it. It's not the end of the exponential growth of computing, it just went to transistors, and then Moore's law about transistor density. When you hit 20 atoms, you can't shrink any more. The 6th paradigm is 3D molecular computing, there are prototypes of these computing power that they expect to get out there before the end of the fifth paradigm.
Every level is 100,000 times more powerful than the one before it, in terms of price point performance. It's a double exponential, straight line on a log log scale, it's exponential, slow exponential, it took 3, 2 years 1950, 12 months in 2000, 11 months.. none of that is the most remarkable thing about this graph. It's not the only graph. It really applies to anything we can measure, even biological technologies. Look at how smooth and predictable this is. This is over a century, through thick and thin, through boom times and ressessions, you don't see any impact of WW1, WW2, or the great depression, or the gulf war. Why not just sit back and relax about this? Well, then it wouldn't happen. There's a pre-requisite to this.
You need millions of people who are passionate about their projects, you need that for this kind of progress. But it's still pretty remarkable. You think that information would be the least predictable access, measuring human innovation, competition, entrepreneurship, there's things in science where you see something similar. You can model each particle in a gas that does a random walk. This includes the laws of thermodynamics, the overall properties of the gas are highly predictable. The law of large numbers works out to be very reliable. When we can measure it, when we have some basic measure of the information content. It follows this predictable trajectory.
30 years ago I realized that the key is _____. It was quite surprising how remarkably predictable this aspect is, it doesn't tell you what company is going to succeed. You needed some key insight to be able to tell, reverse engineering, and the engineering were, would be the one. But you could tell that the technologies were coming into place, for enough computation, enough speeds to do something like that. I don't want to dwell on these examples. Many examples, the price of transistor for one dollar in 1986 per dollar, then to a billion per dollar. This is not some government mandated program, it's not table top physics. The cost of the transistor in cycles comes out by half.. that's a really key point, that's a 50% deflation point in the price of electronics. Apple can do an iphone today, it's twice as good as the one four years ago for half the money. This improvement in price performance is something that we all benefit from.
Some economists have expressed concern that as more of the economy becomes information based, then by 2020s, at the 50% deflation rate, that would lead to a contraction not of the universe but of the economy. Some of them have good reasons. Anyway, that would not be a good thing. If you get twice as much stuff, in the next year, for the same price, you're not going to double your consumption. There's emerging tech, like 3D printers, these features, the scale of these features are shrinking by.. decade.. shrinking in 25 yyears, that would get us to a nanoscale, I could email you the spec for almost any 3D spec, and you just print it out, like clothing and so on. I do mean stuff, I could email you a movie, a book, or a sound recording. And, if you could get twice as much stuff for a year later, you're going to increase your consumption, but not double it. With respect to currency it might shrink.. but not really. There's been 18% in dollars in every form of information technology despite you getting twice as much every year. People did do search engines, people did do buy ipods, This is what is providing economic growth. That's not Moore's law.. different engineers, different companies, it's the law of accelerating returns. Moore's law is one example.
Biotechnology is very significant, the costs are dropping (Carlson curves). The slope of that graph is doubling each year, the amount of genetic data. How to express the proteins, simulating things like protein folding, I don't want to dwell too much on this, because it's not the topic of this talk. It's another example and a very important one, this is actually what we call bridge 2, bridge 2 is is the reprogramming of our biology, and this is what Aubrey focuses on. How long do you go when you update your software on your phone? I am walking around with software, not a metaphor, it's literally- evolved millions of years ago- I'd like to tell my blah gene, don't hold on to every calorie, I'll have food tomorrow. It was evolved thousands of years ago. Some of these genes were turned off in an experiment with an animals.. this can bebrought to the human market. That's just one of the 22,000 genes. You can use RNA interference to turn off genes. So, these are technologies that are generally staged, now that they are subject to law of accelerating returns, now that they are information technologies, they will be a million times more powerful in 20 years. MIT announced the first.. in 39 years.. Department of Biological Engineering. The first new department in 39 years at MIT was the Department of Biological Engineering.
This is the basic graph of teh amount of hits on the internet, doubling every year. I had this graph on the left. Just a few points with the internet, when it was called ARPANET. I projected it out, WWW in the 1800s, with mails, and 1990. People thought that was ridiculous. The entire defense budget, a few thousand scientists, it happened, that's the power of exponential growth. We do adjust, eventually. These things will seem very daunting, and now that we have them, it's just a part of every day reality. How did life ever exist without facebook? So, shrinking tech and exponential rates.
We'll have plenty of computation going through the 21st century. We'll simulate the human brain and ai. I'd like to spend some more time on this topic. Reverse engineering the brain. I'm writing a book now on reverse engineering the brain. The ultimate source of the templates of intelligence. Consciousness and free will. Some people dismiss this, they are not scientific, I agree, but I don't think we can dismiss them, because our whole moral and ethical system is based on that. If I cause harm to some other conscious entity, that's considered wrong- that's the one golden rule. The question is then, what's consciousness? Causing pain and suffering to consciousness, that's a crime. So, we can't ignore consciousness so easily. There's a lot of new theories, on somehow consciousness is somehow related to quantum computing, and people pointed out that neurons were large messy environments, and maybe it's the tubules, and the structures for quantum computing, and I do think that the motivation is that consciousness is mysterious, because it's not identifiable to any identifiable means. This one is conscious, this one is not, it's philosophical. Biological neural system, then operations, entity has some model of its own performance and its own decision making, it wouldn't matter what the substrate is. Different models would have different assumptions, purely objective scientific measure, exceptional gap between objective measurement in science, and subjectivity which is a synonym for consciousness. So, consciousness is somewhat mysterious. Quantum computing is somewhat mysterious, so there must be a link between the two. That was an assumption, there's a new theory of his theory now, that it's now cellular automonon, within the tubules, within the cells. I'm reading his paper, and it promises to explain the source of consciousness, and I've read a number of papers like this, and it's a very scientific paper, it describes the tubules, and the atomic mechanisms that could be like little computers, computations, mathematical analyses, what these would be capable of, we think this is the source of consciousness, and then it's just "woah, that's a leap of faith". It's no less a leap of faith than any other, like of religion, that there's some entity called God, and a soul, and that it's consciousness, at least that's attributed as a leap of faith. At least they admit it, it's not presented as a scientific paper. It looks like a scientific paper, in the paragraph it doesn't advertize itself at all, it's a little leap. What's the basis for believing that this has anything to do with consciousness? Maybe it does, but there's no demonstration, and it's very hard to imagine what that demonstration could be. There are other theories like this, algorithms having to do with consciousness, so all these different frequency ryhthtms.
But, and sometimes it's a little bit of evidence that is presented, goes away in anesthesia, and that's the test bed for consciousness, under anesthesia. There's two points to be made there, lots of things are not working under anesthesia, like the whole mechanism cerebral cortex, billion pattern recognizers, which are constantly firing and looking for firing and irony, and looking for concepts, that's not working under anesthesia, I would actually think that consciousness would have more to do with that, because that's the content of our thinking. You can't even conclude that we're not conscious under anesthesia, all that we know is that we have no memory of our experience, that doesn't mean we weren't conscious. There's definitely a difference between conscious and memory. There's lots of things that I can't remember now, that I am pretty sure I was conscious for at the time.
There's this issue, and I try to bring it up to issue to what we know now. My ultimate conclusion, the cahper is called "you got to have faith"- there is no scientific experiment that we can even imagine- that can definitively demonstrate consciousness. Turing intended his test to be a test for consciousness, it's not testing consciousness but rather whether or not an entity seems like human life. But that makes a few assumptions. My leap of faith, that's an assumption that I would make, if I was interacting with another person. If it seemed like the thing was conscious, then sure I could make that leap. And if not, I'm not so sure. We're making exponential gains in understanding these phemoemon, but you have to understand the nature of exponential growth. Halfway through the genome project, we had 1%. At that point, it gained real traction. There are 20 different regions that have been simulated, even in full or in part. And a lot of things are growing exponential, like the resolution in brain scanning is doubling every year, and we're simulating various regions, the most important one is the cerebral cortex, which is where we do our hierarchical thinking. It's not exactly unique to humans, it's unique to mammals, only mammals have a neocortex, it's about the size of a postage stamp in a mouse. It's a structure. It's capable of hierarchical thinking, we can see how one pattern recognizer and a whole bunch of them leading to another, and you can actually re-assign them, re-learn material, we can see this with brain scanners now, we can read thoughts and read our brain and see it creating new structures, part of it gets swept away with a stroke or brain damage. We can relearn material using completely other regions of the neocortex. It takes time to develop that. Blue Brain Project images.
One of my critics wrote a paper saying that we would never simulate a brain, and it would require a trillion line of codes. He looked at the vast complexity of connections, and this is just a small slice of connections. John Vorgan. So I looked up the picture, it's not a pic of the neocortex, it's a picture of the simulation of the neocortex.. so I guess his point just self-destructed. The point is that we have figured it out, and these are scaling up exponentially. I was at a panel a few months ago, and I was conservative. The head of the Blue Brain Project that he would have a few human scale simulation of the human cortex of 2018. I told him that he may have something by 2018, that has the billion pattern recognizers, but it's not going to work, it's going to take time to get it to function, so I said 2029. But there are other people who have simulated the auditory cortex, the visual cortex, the cerebellum, which has at least half the neurons of the brain. Large scale simulations. Some of them are taken over by the neocortexx, a lot of it is by the cerebellum. It has to solve several simultaneous differential equations to catch a ball in seconds, and ten year olds haven't taken calculus. The cerebellum uses something called basis functions, and that's how it solves the equations. Where are the trillion lines of code? The design of the brain is general, along w..
The genome has a lot of redundancy. Including the junk DNA to not be junk, to be meaningful, it has redundancies, sequences, 300000x times, you have lots of redundancy, lossless compression, and you can compress it to about 50 million bytes, or maybe 25 million bytes, that's maybe half a million lines of code. How could that be? There are trillions of connections in the cerebellum, and trillions in the cerebral cortex. How could you get that out of the lines of code? Well, redundancy. The cerebral cortex has modules repeated a billion times. I use this as an analogy, the uh, Mandelbrot set. Picture of which is on the .. local complexity. It's going to be .. very complicated image, and uh, it would take trillions of bits to represent it at the whatever level which is what the brain is. The design of the mandelbrot is just six lines long, it's a fractal, and if you iterate it, you get those images. The structure of the brain is a probabilistic recursive fractal. The brain isn't six letters long, it's 25M letters long or something.
This is the source of economic growth, the adoption of these technologies is exponential. People sometimes ask, how come we don't see the boom and bust of the .com. Well, that was a Washington thing. Exponential growth could be sublinear, doubling looks like tiny numbers, it could look like nothing is happening. Washington came back and looked at 2000 and said that, it didn't look like anything changed. Meanwhile, it was growing exponentially. You had Google, which was really a .com, got hundreds of millions in real revenue. Century ago.. more .. skill latter is moving up. We've increased ten-fold over the .. in investment per .. 60,000 college students in 1870, 6M today. Sometimes people talk about wealth.
Longevity tech. Aubrey talked about these, like biotech, reprogramming the information processes, which now considers information technology will progress exponentially, and nanotech, we're going to run out of resources. Biological population. We're wat.. we have 10,000 and energy we need from the sun, this happens to be in the wrong form, not in the form of electricity, it's modestly useful. We're applying information tech to solar panels, so that's going down for the moment. If you look at the lower right-hand graph, the cumulative solar energy in the world has been doubling every 2 years and has been for 20 years. It's only 8 doublings for hitting 100% of the world's energy needs. These technologies always look like fringe players. The internet wasn't going anywhere when it was just 1% of potential users, it's only 8 doublings at 2 years each for meeting 100% of the world's energy needs.
So I shared this report with the Prime Minister of Israel a couple months ago, he said great we have enough sun light. Yeah, we have 10000x we have than we need. So, the next day the president announced an israel initiative to replace the world's fossil fuels with solar energy to use these scientists. It's going to take 16 years to do it, but whatever.
Computers are starting to disappear. Images directly to retina, I've seen early products of these. A virtual screen that's a big as high resolution. carrying around a big screne, that's the emerging, virtual display, overtaking your whole visual field, you'll be immersed in games, Microsoft for example, pick up all of your movements and moving in a .. or elm-related.. or music-related interactive game, it's .. really.. we will see. These technologies are getting closer and closer to us, when I was a student down the street, I had to take.. people put them in their bodies already, putting computers inside your brain. Time is a pressing concern, people hacking into the brain, and it was not a humor piece. That is today. These technologies will be a billion times more powerful in 25 years, and a hundred times smaller, and you get some kind of idea of what would be feasible.
This is what we've done, before alpha medicine and information technology. This was when it was hit or mix, it was progressing linearly, not digitally, so that's been useful. This is going to come to height here. Any other kinds of limits, we can engineer, re-engineer those, now that health and medicine is becoming an information technology, beyond biology by introducing nanotechnology, we're going to become a hybrid of machine and biological and nanobiology heritage, my mind is not going to.. transcending our biology, not our humanity. Hang in there, and thank you for the remarkable future.