Coming Back Around to Artificial Intelligence
Posted by Jeriaska on September 25th, 2007
SIAI Interview Series - Sam Adams
The following transcript of the SIAI Interview with Sam Adams has not been approved by the author. Video and audio are available at the Singularity Institute website.
“[O]ne of the areas that I think will be very fruitful is looking at some of the really old thoughts in the very first years of computers, when people did not know what they would be used for, and then applying them in these new high-powered computer systems that we have.”
“The notion of the Singularity, that it would take off and leave us behind, maybe to our detriment, I personally think we are so very far away from that. In our lifetimes, we are not going to have to worry about it.”
Coming Back Around to Artificial Intelligence
“People have been saying for fifty years that artificial intelligence is just around the corner. What’s different now?”
Well, I think one, we’ve been on a journey of learning what intelligence really means. We’re not there yet. We are a lot closer. We have learned what it is not, entirely. Things like math and logic, which were the basis of early work in AI, is obviously part of what humans do that is intelligent, but there is a lot that we do that is not math and logic-based, like emotion and creativity.
A lot of things that were easy for the early technology approaches to do really opened up things that were hard to do later. Like common sense reasoning: being able to do what a three year-old can do. A three year-old knows that when he lets go of something, it’s going to fall and hit the ground. We do not have computers that know that very well. There are a few that are starting to get that capability. But we started out jumping to the end goal. If you look at things like the Turing test, it was about trying to have a computer that a person could not tell whether it was a human or a computer. We have found that it actually is not a very good test, because it is easy to spoof. To trick a person, there are lots of people who play like they are someone else and fool people. So that turned out not to be as good a test as we thought to drive research.
We also had a lot of great ideas that the hardware and software technology were not ready for. I think one of the reasons why this is starting to come back around and people are much more enthused about making progress is that we have so much more capability. The systems that we work with now have increased in power so many fold over what we started with. In fact, one of the areas that I think will be very fruitful is looking at some of the really old thoughts in the very first years of computers, when people did not know what they would be used for, and then applying them in these new high-powered computer systems that we have.
“Can you give an example of what’s changing?”
Before we had operating systems, word processors, GUI systems and all the kinds of things we take for granted now, computers were a lot more open-ended. People could create something using the electronics, memory and processors. There were not really standards. If you were going to create an application, you had to start from scratch and put it all together. After we started building all these standard systems and platforms that we became very accustomed to, it gave us a lot of productivity, a lot of good business value, but it also kind of locked us into a certain way of thinking about how to use computers. I think one of the things that is changing is people’s willingness to look beyond current models, to look at new kinds of models.
“Can we build greater than human intelligence?”
When we start talking about computers that are smarter than people, or even developing ones that are, we don’t even know how to build systems that are as smart as people. Not even as smart as mammals much smaller scale than us. I think that when we start talking about developing intelligent systems that are general purpose, hence the “artificial general intelligence” moniker that you hear people using now, we are talking about things that would have a kind of learning ability and free learning, autonomous knowledge acquisition capabilities. So just like you walk into a new situation, you have to figure out what is going on and how it affects you. What are you going to do in that environment? How might you utilize what’s there to help you reach your goals. Building a system that is capable of that does not mean it is smarter than a human, but it could be a very useful system, and we do not have systems like that yet.
Now, when you start talking about a large part of the conversation at this conference, what would happen if we created one that was smarter than humans? The notion of the Singularity, that it would take off and leave us behind, maybe to our detriment, I personally think we are so very far away from that. In our lifetimes, we are not going to have to worry about it.
But I do believe that probably in the ten to fifteen year timeframe, we will have interesting AGI’s. We will have systems that have very interesting capabilities that if you had one today you would say it is a revolution. By the time we get there, just like lots of advances we have made in artificial intelligence, everyone says, “Oh, that’s not real AI. It’s something else.” We will develop very useful, interesting systems. There is a new wave of that happening now. I think we will see a lot of interesting advances over the next few years. But, one taking off and leaving us behind I really don’t believe in.
“How is Moore’s law playing out at IBM?”
The interesting point there is that we are actually reaching diminishing returns with classical Moore’s Law. You have noticed now that high-end laptops now have two processor cores per chip? We have gotten to the point now to where just making one processor faster and faster is starting to get where we cannot keep pushing it because we are running into physical limits. We cannot run it any hotter or we start melting the silicon. We cannot get the memory fast enough to drive the chip. So what people are looking at are chips that have many, many processors on one core. And that is a different kind of advance in computing that we really have not dealt with too much before: things like Blue Gene, a lot of the systems that we use to simulate engineering or nuclear decay for the government, things like that.
We have been doing high performance parallel computing for a long time, but not really in a general purpose way - it has been for very special purpose applications. So, even though processors are getting faster and more powerful, they are getting more powerful in a different way. It’s not just going to run the same application and keep running it twice as fast every year or two. It’s going to have to be different kinds of applications. Right now we don’t know how to make those work.
“Is IBM integrating future technologies?”
I am definitely not an expert in all that IBM has done in nanotechnology or quantum computing, we have some of the world class people in that space. We developed a lot of the field and are looking at all kinds of interesting applications there. Some of the technology, like quantum computing, is a lot farther off from large scale practical use, but we are learning interesting things about it that will help guide the way we do chip design today.
Now, nanoscale work, we use that all the time in modern chip technology. It’s a way of improving things. When you look at what we are starting to do with carbon nanotubes, for instance, in building circuits, we are doing fascinating work there that is going to keep that track of progression in increasing computing power moving. But there is so much more than just the technology growth. We have to learn how to make that accessible to large numbers of programmers. Right now, doing large scale parallel processing like we do with, say, Blue Gene, takes very specialized training to do. It is not something that everyone who can write some JAVA script and build a webpage could get at. So how do you make that technology accessible? How do you let people at the new power to create new kinds of innovation and business value? That’s the thing that we worry about a lot, not just the technology advancement.
“Is IBM working toward AGI?”
IBM has been involved in this particular field of AI as long as it has been around. Things like Deep Blue were the result of thirty years of research that IBM funded - not only internally, but we also funded a lot of work in universities and collaborated with other companies. In fact, our research model now is moving much more toward an open, collaborative model with universities and small companies as well. You obviously run into some issues there. Small companies are there in order to make their mark, and their need to make a profit off of something may not jive with what IBM wants to do with the same technology. So, there are always issues about working out how to collaborate and how to share the benefits. The same is true with universities which focus on getting value for the intellectual property that they create as part of research. But we are always interested in collaborating with people when it makes good business sense for our customers. We find that to be very common. We do lots of work with different folks, both start-ups, universities, and other large companies.
“What are your thoughts on SIAI’s Singularity Summit?”
It’s an amazing collection of people with extremely broad views. People who are philosophers who are trying to figure out where humanity is going. People who are engineers who are trying to make something work that has not been made before. Everywhere in between. People who are afraid of where technology might be taking us. People who are excited about it. A whole collection of those people and the kind of open exchange that this event has generated I think is very positive. I think for one, we are going to find some useful dialog and ways of discussing these things, even with people who are very polarized on these issues, to make good progress. I hope that we continue doing these on an annual basis. I think it’s very worthwhile.
