Three articles (1, 2, 3) based on a 40-minute interview with Ray Kurzweil are available on ComputerWorld. It talks about the singularity, pervasive computing, augmented reality, storage as a philosophical issue, exponential growth" of computational power, portable computing, virtual reality, immortality, and strong vs. narrow AI. From the first article: "Ray Kurzweil is a futurist and author whose book The Singularity Is Near: When Humans Transcend Biology (Viking Adult, 2005) predicts advances in computing technologies and biological research over the next four decades, culminating in the merger of biological and nonbiological intelligence. Kurzweil is also a prolific inventor who has developed hardware and software for optical character recognition, speech recognition and electronic music.
At the heart of your book is the idea that technology advances exponentially. Can you explain? Technology, particularly if we can measure the information content, proceeds exponentially, not linearly. And a lot of people don’t realize that, and that’s one of the reasons long-term forecasts generally fall substantially short of the ultimate reality.
If we look at information technology, we see this reflected in an exponential growth in the power of those technologies. The price/performance for computing is literally doubling every year. Information processes are revolutionizing every industry, every area of technology. And so [areas] like health and medicine, which used to be hit-or-miss, are now becoming information technologies and will be subject to what I call this law of accelerating returns.
How will hardware technologies evolve over the next 10 years? If you go out 10 years, computers are not going to be these rectangular objects we carry around. They’re going to be extremely tiny. They’re going to be everywhere. There’s going to be pervasive computing. It’s going to be embedded in the environment, in our clothing. It’s going to be self-organizing.
We’re going to solve this dilemma we have now with displays. On the one hand, people like 50-inch screens, and they’ll spend thousands of dollars on them. On the other hand, they like watching movies on a 1- or 2-inch screen, but that’s really not a satisfactory experience. We are going to solve that by putting the displays in our glasses, which will beam images to our retinas. This will create very high-resolution virtual displays that can hover in the air. And it can also completely overtake your visual field of view in three dimensions, creating full-immersion visual/auditory virtual reality.
We’ll also have augmented real reality. The computers will be watching what you watch, listening to what you’re saying, and they’ll be helping. So if you look at someone, little pop-ups will appear in your field of view, reminding you of who that is, giving you information about them, reminding you that it’s their birthday next Tuesday. If you look at buildings, it will give you information, it will help you walk around. If it hears you stumbling over some information that you can’t quite think of, it will just pop up without you having to ask.
What’s your definition of artificial intelligence? Artificial intelligence is the ability to perform a task that is normally performed by natural intelligence, particularly human natural intelligence. We have in fact artificial intelligence that can perform many tasks that used to require — and could only be done by — human intelligence. There are hundreds of examples today, and they are deeply embedded in our economic infrastructure.
All communication is governed by intelligent algorithms that route and connect the information. Programs are embedded into computer-assisted design systems. AI flies and lands airplanes, guides intelligent weapons systems, places billions of dollars of financial transactions each day.
These examples are narrow AI, in that they are performing specific tasks, very often sophisticated tasks that required human experts to perform.
What could slow down the arrival of strong AI, or of the “smarter than human” technologies you call the Singularity? There are really two areas to think about. One is hardware and one is software. There’s a strong consensus that the hardware will be available. So, the key issue is how long it will take to get the software and science. I make the case that a 20-year horizon is a conservative estimate, based on the exponential progress we’re making in reverse-engineering the human brain.
In one of your earlier books, The Age of Spiritual Machines, you have a chapter titled “2009.” And you nailed quite a few technologies pretty well. But one technology that didn’t seem to fulfill the promise that you anticipated was speech recognition. Well, first of all, this isn’t 2009 yet. We need exponential progress in computation to get linear gains in speech recognition accuracy, because we are making exponential gains in computing. And a lot of people’s impressions of speech recognition are based on having tried it three, four, five years ago. It’s actually improved a great deal.
Language translation is quite good, particularly now that we have these large Rosetta Stones of matching text in different languages, so the statistical approach of doing language translation with very large Rosetta Stone text to train on, using pattern-recognition techniques, gets very excellent results.
You have also discussed an intriguing invention that you call the “Document Image and Storage Invention,” for long-term storage of computer files. But you have concluded that it really wouldn’t work. Why? Software formats are constantly changing. Try resuscitating some information on some PDP-1 magnetic tapes. Even if you could get the hardware to work, the software formats are completely alien, and nobody is there to support these formats anymore.
I think this is fundamentally a philosophical issue. I don’t think there’s any technical solution to it. Information actually will die if you don’t continually update it."
Second article
"On Monday, Computerworld published a portion of my phone interview with Ray Kurzweil for The Grill. There wasn't enough room in the print version of Computerworld to display the complete transcript, so below I am including some of the other questions and answers from the interview.Kurzweil's additional comments to my question about the "exponential growth" of computational power:
First of all, in terms of "knee of the curve" and the exponentialist view that doesn't have any discontinuities, [it's] nonetheless explosive. If I count to 30 steps linearly, I get to 30, if I could exponentially, 2-4-8-16, I get to a billion. So it's really quite a big difference. And technology, particularly if we can measure the information content, proceeds exponentially, not linearly. And a lot of people don't realize that, and that's one of the reasons long-term forecasts generally fall substantially short of the ultimate reality.
We see this in both biological and technological evolution. An evolutionary process creates a capability, then it incorporates that capability. Therefore the next stage goes more quickly. So it took a billion years for DNA to evolve, but then evolution used DNA ever since, and the Cambrian explosion came, and it went 100 times faster, and it only took 10 million years, biological evolution kept accelerating, and homo sapiens evolved in only a few hundred thousand years, and then really the cutting edge of evolution, which is really the increase of complexity in our environment, switched to technological evolution, and the first steps of that were only tens of thousands of years ago.
Then we always use the latest tools to create the next set of tools, so that process has accelerated. So we now have paradigm shifts in just a few years' time. If you look at computers, the first computers were actually designed on pen on paper and wired with screwdrivers and took years, and today you can specify some high-level formulas to a CAD program and generate 12 layers of intermediate design automatically.
... In fact, there's a second level of exponential growth. It took three years to double the price/performance of computing in 1900. Two years in the middle of the 20th century. It was 12 months in the year 2000, it's now actually down to 11 months.
And doubling every year, even ignoring the second level of exponential growth, is multiplying by a billion in 30 years, and because of the second level of exponential growth, it’s actually 25 years. And if you look at how influential information technology is already, and you imagine multiply that by a billion over the next quarter century, you get an idea of what would be feasible.
And the last point I'll make about this is that it's not just limited to computers. People don't have to have been around very long to -- even teenagers have even noticed how rapidly computation and communication technology has advanced in a short period of time. In my cell phone, I have 1000 times as much computation as all of MIT had when I was a student there.
But it's not just a computer. It's also other areas of science and technology are now becoming information technologies. A very important one is biology, which didn't use to be an information technology. There was some information in it, but it was basically hit or miss. But now that we have collected the genome, and actually have the tools to reprogram the biologies, the way we reprogram our computers, we can turn genes off with RNA interference, we can add new genes with gene therapy and so on.
This is becoming an information technology and therefore it is subject to this doubling of price/performance every year. And these technologies will be a thousand times more powerful than they are today ten years from now.
And we saw that in the genome project. The amount of genetic data has doubled every year, very smoothly. So halfway through the 15-year project, skeptics said, "I told you this wasn't going to work. You're halfway through the project, and you're only through 1% of the project." But that in fact is the correct trajectory for an exponential, because that 1% doubled every year for the next seven years, and the project got done on time.
And we've continued the exponential growth, both with genetic data and with every other aspect of biology, and it's also true of reverse-engineering the human brain, and this ultimately will solve problems that you wouldn't think are remotely related to information, like the energy crisis, because we'll use nanotechnology, which is a form of information technology, basically reorganizing matter and energy at the molecular level, to create extremely inexpensive, efficient, solar panels that can capture enough sunlight to completely eliminate fossil fuels, which I believe we'll do within 20 years. Because we actually have 10,000 times more sunlight that falls on the earth than we actually need to meet all of our energy needs.
So ultimately, everything is going to be transformed by information technology. We're moving toward tabletop devices that can actually create three-dimensional objects. Right now you can take an information file, and turn that into a movie, or a book, or a sound recording, and those things used to be physical products, and now are just information.
Well, the same thing will be true of what we now think of as physical products. We'll be able to have an information file, and be able to turn it into any three-dimensional object that you need, such as a module for a house, a solar panel, a toaster, or even the toast, or a blouse, by basically reorganizing matter and energy from very basic input materials, which will be recycled, to create physical products, and there are a number of roadmaps to get there, and I believe we'll see those kinds devices within 20 years.
Whether or not people are receptive to his ideas:
Well, if I have enough time to explain it, I generally get a very positive response. The idea is not just one idea that people can lay on their [garbled] for understanding. There are a number of different aspects to it and it also takes a certain amount of data to see how convincing a story this is. If you look at these curves lets say for the price of a transistor, buy one transistor for a dollar in 1968, it's about 300 million today for a dollar. We've heard those fantastic numbers, but if you look at the graph on a logarithmic scale, it's a perfectly straight line with very little wobble to it.
It's remarkable how predictable these trends are. In computation, which is the classical most important example, this goes back a 110 years, with the data processing equipment used in the 1890 census. It's not just Moore's Law, because Moore's Law didn't kick in until about 60 years later in the 1960s -- 70 years later.
So when I show 20, 30, 40 of these graphs, and people see how persistent and how predictable over long periods of time, these trends are, it does make a convincing case.
I've not just been looking backwards recently. I've been making forward looking predictions based on these trends for about a quarter of a century. The projections I made in my first book, which I wrote in the 1980s, provided a very accurate roadmap in the 1990s and the first decade of the 21st century. It was considered very radical back then and now the predictions seem quite mundane, like the emergence of a worldwide communications network in the mid-1990s cause I saw the DARPAnet doubling every year and so on."
Third article
"My 40-minute interview with futurist, inventor, and author Ray Kurzweil was too long to print in its entirety in the The Grill, so I'm publishing the additional segments on my blog. Earlier I posted Kurzweil's comments about the exponential growth of computing power. Below are some of his thoughts on portable computing, virtual reality and virtual worlds, storage standards, biomedical advances, immortality, and narrow vs. strong AI:On portable computing:
Right now, if you think about, your cell phone, your PC are not nodes on a worldwide network, they're spokes into the network and then the network has nodes inside it.
But all of these different devices are going to become nodes, meaning that it's not only going to be sending and receiving your own messages, but it's also going to be cooperating by forwarding other people's messages, and being part of a pervasive sort of web of communication.
The practical implication for people using this is you're not going to be carrying around objects. Computing is going to become invisible. It's going to be woven in your belt buckle, and your clothing.
On virtual reality:
We'll be spending quite a bit of our time in virtual-reality environments. Environments like Second Life are really a crude harbinger of what is to come. We'll have these virtual-reality environments which will be quite competitive with real reality. They'll be very realistic. They'll be full immersion, just as in SL you can be someone else, you don't have to look the same in these virtual environments.
And they won't just be kind of a plaything. Second Life already has a real economy, and people do real business transactions and have real romance. And we'll be doing that in a real panoply of virtual-reality environments.
On biomedical advances, and immortality:
Ten years from now, this biotechnology revolution which I alluded to earlier will be becoming quite mature. It will be a thousand times more advanced than we are today. We collected the genome four years ago, we're actually making pretty good progress already -- although it's still early on actually reverse engineering it. Which is to say, understanding how biology actually works. But that will be at an advanced stage ten and certainly 15 years from now.
My prediction is within 15 years from now, we'll be adding more than a year every year to your remaining life expectancies. So that's kind of a tipping point. It's not a guarantee of immortality, but rather than the sands of time running out, they'll be running in. You make it through a year, you'll actually make it through another 15 months added to our life expectancy. Right now we're adding about three or four months per year to life expectancy. When that goes over a year per year, it will represent the tipping point.
We're actually learning to re-engineer our bodies. If I ask you, how long does a car last? Well, most cars don't last that long. But in fact, you can take care of a car, and there are cars that are 80 years old, even a hundred years old. If you really address what goes wrong with them, you fix it, replace different parts -- we're going to be able to do that with our bodies. The reason that analogy doesn't hold with our bodies today, is that we don't have all the plans. We don't understand all of the mechanisms. That's what's proceeding exponentially. Within 15 years, we really will have enough of the plans to be achieving this tipping point in terms of life extension.
On speech recognition:
If you look at 2009-2010 cell phone technology, I believe you will be seeing speech-to-speech translation in there. You do have large-vocabulary speech recognition on the phone today, for quite a few applications. And there are millions of people who are actually creating written documents using speech recognition. That's not quite ubiquitous, because there are hundreds of millions, if not billions, of computer users, and there are millions of people using large-vocabulary speech recognition to create text
But that's growing, and the accuracy is getting quite good, particularly with a little bit of training. Like if you train it for 10-15 minutes on your voice. So that's coming. And I think speech to speech language translation on cell phones and services like Skype, will actually be very widely used before text creation, although I say there are millions of people using text creation. That's because, when you create text, you more or less want perfect documents and casual conversation, people can make an occasional error, you can compensate for that from context.
And just the opportunity to speak to people who don't speak the same language is overcoming a big barrier.
On the potential of storage standards to extend the life of digital documents:
We do use standard formats, and the standard formats are continually changed, and the formats are not always backwards compatible. It's a nice goal, but it actually doesn't work.
I have in fact electronic information that in fact goes back through many different computer systems. Some of it now I cannot access. In theory I could, or with enough effort, find people to decipher it, but it's not readily accessible. The more backwards you go, the more of a challenge it becomes.
And despite the goal of maintaining standards, or maintaining forward compatibility, or backwards compatibility, it doesn't really work out that way. Maybe we will improve that. Hard documents are actually the easiest to access. Fairly crude technologies like microfilm or microfiche which basically has documents are very easy to access.
So ironically, the most primitive formats are the ones that are easiest.
So something like acrobat documents, which are basically trying to preserve a flat document, is actually a pretty good format, and is likely to last a pretty long time. But I am not confident that these standards will remain. I think the philosophical implication is that we have to really care about knowledge. If we care about knowledge it will be preserved. And this is true knowledge in general, because knowledge is not just information. Because each generation is preserving the knowledge it cares about and of course a lot of that knowledge is preserved from earlier times, but we have to sort of re-synthesize it and re-understand it, and appreciate it anew.
On narrow AI:
The narrowness is going to gradually get less narrow and one of the sources of human-level AI that has the breadth and generality of human intelligence is going to be understanding the human brain itself. And that's another area in which we see exponential progress. The spatial resolution of brain scanning is doubling every year, and the amount of data we are collecting is doubling every year on the brain. And most importantly is we're showing that we can actually understand this data and turn it into working models and simulations. And there's an increasing number of areas of the brain, regions, that have been modeled and simulated, and these simulations are gaining in sophistication and precision.
This includes areas of the auditory cortex, the visual cortex, the cerebellum. IBM has a project to simulate a slice of the cerebral cortex, which is arguably the most important region, where we do our abstract reasoning.
And I made the case in the book, that within 20 years we will have models and simulations of all the regions. There are many different benefits to that goal to understand how the brain works, we'll be able to fix it better. But most importantly, it will expand the AI toolkit.
And it's not my view that we would actually have to do this. I actually think we would achieve human level AI if we never looked at the brain at all. On the other hand, I think we will get there faster with more supple algorithms, by understanding the best example we have of intelligence, which is the human brain.
On strong AI:
The term "strong AI" actually originally came from John Searle who is a critic of artificial intelligence. He was actually making a different point, which is about consciousness. But the term has stuck.
A lot of terms have stuck, that we may have preferred to not be the term of choice. Like "artificial intelligence", because it makes it sound like it's not real intelligence.
"Artificial intelligence" is real intelligence. And virtual reality is a real form of reality. You and I are now engaging in a form of auditory reality but it doesn't mean it's not a real conversation. I can't say, "that's not a real agreement I made with you last night, that was virtual reality."
But anyway, "strong AI" has come to refer to human level AI, artificial intelligence that could pass the Turing Test, which is the test that Alan Turing devised half a century ago in which a human judge interviews an AI and a human, or maybe several of each, over what he called teletype lines, basically instant messaging. If after a suitably long period of time, he or she cannot tell who is the AI and who is the human, then the AI is said to have passed the test. And it has actually held up as really the only good test we have of human level or so called strong AI."