Increased Intelligence, Improved Life
Posted by Jeriaska on October 2nd, 2007
Peter Voss is an entrepreneur with a background in electronics, computer systems, business and technical software, as well as management. He has a keen interest in cognitive science and the inter-relationship between philosophy, psychology, ethics and computer science. Since the early 90’s he has been researching and developing artificial general intelligence, and in 2001 started Adaptive A.I. Inc., with the express goal of developing a commercially viable general-purpose AI engine.
An Extropian, he is actively involved in futurism, free-market ideas, and radical life-extension. In his view, artificial general intelligence (AGI) promises unprecedented advances not only in science and technology, but also in ethics and social systems. Peter Voss’s 2007 Singularity Summit talk explores some of these improvements, making a case for how increased intelligence leads to improved morality.
The following transcript of Peter Voss’s Singularity Summit presentation “Increased Intelligence, Improved Life” has not been approved by the author. An audio version of the talk is available at the Singularity Institute website.
Increased Intelligence, Improved Life
Good afternoon. Quite a few of the talks I give are about life extension. In a way, this is going to be another one of those. However, this is not going to be about calorie restriction, cryonics, or some other biotechnical enhancement, but rather how AGI can extend our lives and potentially improve it, and I believe improve it dramatically. My talk is going to touch on the benefits and opportunities, so I’m very happy that I’m not going to have to address the negative aspect of the technology, and I want to make it clear that I do believe there are risks and dangers involved in such powerful technology, but I’m happy that I don’t have to address those issues. I’ll leave that up to other people to elaborate on. Also, because I’m just talking after Stephen [Omohundro], I would like to make clear that my talk is not really focusing on self-improving systems, even thought the technology will ultimately lead to that, it’s not something I’m really addressing. So, I just want to switch contexts here. What I’m talking about are not at the level of self-improving systems.
So, let me start off by clarifying a little bit what I mean by AGI. I’m happy to see that the term ‘AGI’ has now managed to become pretty mainstream. There is actually a conference coming up on AGI. Six years ago, when I was involved, together with Ben Goertzel and a few other people, in coming up with a term for this technology we were working on, the term was really unknown, and now it has become mainstream. So I usually contrast artificial general intelligence to ‘conventional’ or ‘narrow’ AI. So the focus is on acquiring and learning knowledge and skills, as opposed to having or encoding the knowledge and skills. So the important distinction is that new knowledge has to be acquired. The second distinction is that it’s general, ongoing, cumulative, conceptual learning - generalized learning - as opposed to domain specific, generally fixed, rule-based abilities. Experience and self-motivated improvements versus externally initiated improvements. As the system is being used it will improve itself, improving its capabilities and knowledge by learning, as opposed to the programmers sitting down and deciding what new features need to be added to the system.
To set context for the main part of what I wanted to talk about, I just want to address the issues of pathways and some of the difficulties involved. I believe there are fundamentally two pathways at the extremes. One is ongoing incremental AI improvements, basically doing more of what is being done now. I believe that will eventually lead to AGI. Eventually, machines will become that powerful, software technology will become that powerful, and it will just fall out that we will have generally intelligent systems. I think that could take a long time. The second one is, of course, the targeted approach to developing AGI system. There are a few different ways of going about that: logical, mathematical, theory-driven systems, evolutionary, reverse-engineering the brain, directly going to adult-level engineered AGI, choosing a more developmental model, and then robotics and embedded. I don’t really have the time to expand on any of them. I think we have already heard about examples of each one of these today.
One of the major challenges to achieving AGI from my experience of having been personally involved full-time in this for at least six years, probably more like ten, is I’ve come to the conclusion that these are some of the really important things. You need a viable design. That might sound very obvious, but if you actually try to find out how many teams or how many people believe they have a viable design - even were you to ask them, “Do you believe you have a viable design? If there were a $15 million X-Prize for AGI, could you even enter the competition, could you convince someone that you actually have a hope of achieving it?” I think there are very few people who have what they even themselves would consider a viable design, and there are a lot of reasons for that. Primarily, there are not a lot of people right now who are attempting to address this problem directly. Secondly you also need big vision, because you really are going against the mainstream of AI knowledge, and you need this vision to span this credibility gap and to see what you can achieve. Clearly you need some viable business or development plan to get the support that you need. Focus, focus, focus. Focusing on solving the problem of general intelligence. It is very easy to get distracted and to try and solve some interesting theoretical problem or some specific application, and it’s really, really easy for that to happen. Of course, if there is commercial pressure, then there’s even more reason that you will lose focus and try and solve some specific problem or meet some of your investors’ requirements rather than trying to solve the problem of AGI.
Leadership is always important in any large, difficult project. And, of course, funding. How far are we from AGI? Well, I believe we might be closer to it than most people believe: almost certainly less than ten years and quite likely less than five. There are some significant shortcuts to AGI that I believe are overlooked. There are a lot of unnecessary problems being addressed and a lot of unnecessarily difficult paths are being followed. I don’t believe hardware power is a serious limitation, and I believe that out of the millions of pieces of the puzzle that are out there, the problem is to identify the ones that need to be picked and put together. The AGI winter is ending, somewhat. There is a real revival happening. As I mentioned, there is a conference happening. This conference is focusing on AGI. AI Magazine was featuring cognitive systems, and there is definitely some movement happening. Then of course, lastly, and to me, most importantly, the work of my company that we’re doing gives me a strong indication of what I think can be achieved in less than five years.
Let me get to the main part and talk about the benefits of AGI. I’ve arranged them roughly in the order in which I think they will occur, though there is obviously a lot of overlap. Now ‘benefits’ is a bit of an understatement, and I’m in a little bit of a difficult situation here because our project is a stealth project and I can’t really talk about what we’re doing, specifically, but perhaps later on, you will see through some of the other things I will talk about of just how significant the benefits can be in these different categories. Now, labor is about the biggest cost in many businesses, and clearly AGI’s could reduce labor costs, so that would be a very significant improvement. Of course the quality of designs in products and services could dramatically improve, leading to an improvement in the standard of living on an overall scale. There will be displacements undoubtedly as this happens. Reduced workers risks have been talked about for a long time. This is happening, but this trend will accelerate. The accelerating improvement will in part come about because many of the systems within companies will have fewer and fewer humans in the critical timing loops, so the things that need to happen fast and efficiently, humans are often the slowest and the least reliable link in the chain. As more and more systems are automated, that will lead to an accelerating improvement, as we truly become a paperless society, for example, which hasn’t really happened yet. At the moment, computers are used to generate more paper and more reports more efficiently, but as humans are taken out of the loop, those improvements will accelerate.
Now, next, in science and technology, many of you here are involved with, or certainly interested in, nanotechnology. I believe that AGI will significantly accelerate the development of nanotechnology, which, by itself, will lead to many significant improvements. Environmental technology - not only will we have much better models for environmental issues, but we will also have better technology to prevent environmental damage and to clean it up by having machines that are more intelligent. Human intelligence really is a big bottleneck for many of the problems that humanity faces. Now, the last thing here is computer science, and I put CAI in brackets there because I certainly don’t want to talk about that in any detail, but the clear indication is that as you have systems that approach human-level intelligence, even if they are not as generally as capable as humans, if they can think, reason, and learn like humans, at a certain point of development, clearly they will be able to improve their own design. So, it is part of the equation.
This abstract here is actually from an essay I wrote on life extension and AGI. Imagine a hundred thousand PhD level researchers focusing their total efforts on life extension and anti-aging research. Imagine them working 24-7 with no distraction from grant proposals, office politics, or attractive coworkers. Imagine the fantastic progress we would see in finding solutions to eliminating debilitating disease and reversing the effects of aging. And, if you really think through how intelligent systems having hundreds of thousands of PhD level brains working 24-7 at solving aging problems and disease, clearly we will see significant progress. These are difficult problems and we need more brainpower to address them.
Move over to education and I guess I could continue, imagine millions of highly competent teachers, etc. Intelligently aggregated and validated worldwide facts and news freely available will make a difference to us, undoubtedly. At the moment we have a lot of information available to us, but the way it’s presented to us, the way it is selected and aggregated is extremely primitive. A news story for some reason is put on the front page and people click on it, which automatically increases its popularity, so more people click on it. You end up with really stupid stories basically taking up all your bandwidth. There isn’t a lot of intelligence in the way that is happening now. An intelligent system can truly give us information from many different perspectives across the world and from different languages and intelligently analyze it. The kind of information you could find from people who are from different cultures or have different backgrounds, the kinds of insights we would gain from that, which is very difficult to get with relatively dumb systems that we have right now, I think will give us a much better understanding and will make us much more informed.
The other thing is, we could also use AGI tools to vastly improve our ability to think and to reason things through. That leads me to the most important part that I wanted to address here: the ethics and morality of mankind and how I believe AGI will help improve us and make us better people. Now, my thesis here is that a significant portion of immoral behavior is really irrational behavior, and I think we can see that when somebody does something bad, that everybody recognizes they should not have done, how often is it because they didn’t think it through properly? They acted emotionally or they didn’t get good advice. A lot of immoral behavior really falls into that category, and the kind of system I’m talking about is really a personal advisor that we will have. An AGI will get to know us, will have our background, and will become a trusted advisor to us, because it will have more information than we have and will able to think things through better than we can, and help us make decisions. I think that it will help us become better people. We will basically be wiser through it.
Now, the second part of bad behavior that often people point out are psychological problems. One cannot necessarily draw a complete distinction between the two. But, again, AGI will help in psychological behavior by helping us retrain ourselves by whispering in our ear when, you know, should we really be doing this right now? Is this just your emotion talking here? Apart from the potential benefits of having better drugs to deal with psychological issues, again I think we can see quite a few benefits from AGI technology there. Also, better law enforcement technology is something that one would also have. If people cannot get away with immoral behavior, if crime does not pay - there are many parts fo the world where crime actually pays, so what happens is that people get encouraged to be criminals, because it is really a rational kind of thing to do, under the circumstances. If society turns around, where people generally are more moral, and crime does not pay as much, then that will have a positive feedback cycle where you will be in a society where being moral is just a win-win situation. It will be in your interest to act that way.
At the social level, of course, on the political and government level, one would hope that a lot of that additional rationality will also prevent countries from going to war willy-nilly, and other irrational acts like that, and really have more foresight in seeing the consequences of their actions, and perhaps find better ways of resolving conflicts. The one thing I skipped past there that I wanted to mention is also the idea of greater transparency and more knowledge available to everybody. Two books in particular come to mind. The Truth Machine by James Halperin I can highly recommend in terms of sketching a society where everybody speaks the truth, and one can argue about how realistic some of the scenarios are, but it really pushed the whole idea of how it could change society when basically lying is no longer an option. The other one is The Transparent Society by David Brin, a fantastic book that explores how society might change, and I think AGI will play a role in making both of those scenarios happen, which I believe, again, might help improve life.
So, in summary, augmented by personal AGI’s, I believe we can become more capable, more wealthy, more healthy, and more moral, and to allow us to move up further in Maslow’s hierarchy, where we are not just fighting for survival, and we don’t have to rely on our basic instincts, which is survival and reproduction, to live life. But where we can really rise high above that and fulfill our destiny. The option of vastly extended lifespans, which is geared to many of us, becomes a real option to all of us. The intelligence of AGI’s will also give us more available energy and resources, yet with a better environment. Now, the other thing that I haven’t mentioned, clearly AGI can help us avoid some of the catastrophes that we potentially are facing. Clearly we all know that in today’s society we are facing a lot of potential problems, and I think that the intelligence and the technology of AGI could certainly help us avoid some of these catastrophes.
Opportunities. I’m out of time, so I won’t spend long on it, which suits me just fine, because, as I said, we are a stealth company, and I can’t really talk much about the specifics of business opportunities. But I really do believe we are now on what we might call ‘the fifth wave.’ The first wave being where physical power was what it was all about. The second wave was where money was what it was all about. Then information. The fourth wave, really the internet and networking technology, it’s debatable. But the fifth wave, I think that having intelligence to help us improve life really is a very fundamental shift in how humans will develop. So, on the personal improvement side, indefinite life. And on the social side, avoiding catastrophe. I believe that, correctly harnessed, improved intelligence could give improved life. Thank you.
