Johan de Kleer talks about knowledge tracking, smart matter and other new developments in AI.
Johan de Kleer is Manager of the Systems and Practices Laboratory, Palo Alto Research Center (PARC). Widely published in the areas of qualitative physics, model-based reasoning, truth maintenance systems, and knowledge representation, he has co-authored three books: "Readings in Qualitative Physics," "Readings in Model-Based Diagnosis," and "Building Problem Solvers." In the award designating him an ACM Fellow, de Kleer was praised for his "seminal contributions of effective techniques for qualitative representation and reasoning about physical systems, and leadership in building research teams that span multiple disciplines."
UBIQUITY: You've been in the artificial intelligence field for 25 years now. What changes have you seen over that period of time?
DE KLEER: Twenty-five years ago, we thought that we would have an artificial mind by now. It turned out to be harder and further beyond our reach than we ever imagined. One of the biggest changes in artificial intelligence has been the realization of how hard and how long-term this project is going to be.
UBIQUITY: What is the current hot idea in the field?
DE KLEER: AI is now a set of hot things rather than one thing. Twenty-five years ago, we thought that within 10 years we'd have a computer that would be able to understand what we said in natural language, either typed or spoken. We haven't been able to do that yet. But because this is such an important goal, at PARC, we've continued to pursue linguistics projects. Now there is a renewed surge of interest in understanding and representing the content of collections of documents. Obviously, a lot of this comes from security concerns, and the ability to analyze language information better and faster. We've recently started a number of projects that combine language analysis with knowledge representation. We are working to represent the content (meaning) of a growing collection of documents. As each new document comes in, we want to be able to identify the relation of the new material to what was previously collected -- is it redundant, is it inconsistent, does it add new facts to previously described situations. The goal of this project we call "Knowledge Tracking".
UBIQUITY: Is that related to something else called Knowledge Fusion?
DE KLEER: Knowledge Fusion is a related idea. Look at Web search. A search is only as good as the documents that are out there. Often the document that you want doesn't exist. Wouldn't it be great if the document that best answered your query would be created automatically for you by fusing pieces of the meanings other documents? These two projects embody one of the hotter topics in AI -- moving away from more statistical techniques that are 80 or 85 percent correct but can't tell the difference between an article for gun control and an article against gun control.
UBIQUITY: What are some other AI topics now getting special attention?
DE KLEER: A second hot area that I'm involved in is a combination of AI and straight computer science called Smart Matter. Traditional computer science thinks of Turing machines, a single point of computation. Now we're thinking in terms of vast grids of computation but each node has limited bandwidth. We're moving into a world where there's going to be billions of sensors each with a small bit of computation and a limited amount of energy. How can we handle such a large scale-up? This is the Internet on steroids. The Turing machine model addresses when to compute -- but I think what is more interesting is where to compute. Our understanding of the constraints of the physical world and the constraints of the computational world need to come together in a whole new way to do this truly effectively.
UBIQUITY: Do you associate Smart Matter as primarily an Internet project?
DE KLEER: It goes way beyond the mechanisms that currently run the Internet because there will be so many more nodes, and their capabilities connect to the physical world through sensing and actuation. The current layered Internet protocols that segregate applications from networking layers are completely wrong when you have connections that have limited energy and bandwidth. Here's another way to think about it. If you have trillions of nodes they can't all send their data to one center of computation. It won't work. There can never be enough capacity at one place -- and it would take too long to distribute the results. The ensemble of nodes has to reason in a distributed way to figure out what to do next. It's far beyond what we're familiar with in distributed computation -- it's a whole new wave. How do you get this constellation of nodes to know where to pay attention and where not to pay attention? That's what we're working on.
UBIQUITY: PARC has always focused on the impact of technology on people. Do you ever worry about the impact of the complexity of technology on the modern world?
DE KLEER: PARC has a history of looking at the interaction of people, their practices and the technology they are using. Much of that work is going on in my laboratory now. In addition, we want to use the increasing capabilities of the technologies to hide the difficulties inherent in the hardware. As the hardware side becomes increasingly complicated the combined system becomes harder and harder to program. You have to deal with modules that will fail and must correct for it. It's going to require new programming paradigms. The programming paradigms that we're starting to use here are built on model-based reasoning (using a model of how the hardware works to control it), and distributing the computation among all the nodes. Each node has a model of its own performance and a model of other node's performances. Program execution is really an ongoing constrained optimization problem. We're using constraint techniques in distributed ways that solve the complexity and makes programming even possible. We use nodes that can change dynamically. The whole network simply adapts to the faulty one. We do all this, and help people interact by allowing them to set goals for the system, rather than having to tell each component how to achieve the goals.
UBIQUITY: Give us a brief explanation of how it works.
DE KLEER: With a grid of a vast number of sensors tracking objects, there will always be some sensors that fail. The system has a rudimentary notion of how to adapt to situations including internal failures. Some of the cataclysmic complexity disasters that one might imagine -- such as a master root node that causes every other node to malfunction -- aren't possible in the kind of architectures that we're putting together. The basic cycle of operation is controlled by techniques for dynamic constraint satisfaction and minimization of entropy.
UBIQUITY: One of the platitudes of modern office complaints is, "I only use three percent of Microsoft Word. I've been using that same three percent for ten years and I don't understand the rest of it." Do you feel that most people don't use the technology, let alone understand it?
DE KLEER: I think that the three-percent figure is true and reasonable. The tough part is that most people use a different three percent. I have three daughters and a wife and they each operate Word in a different way.
UBIQUITY: Tell us about your three daughters. How old are they?
DE KLEER: They are Hannah, 13, Katherine, 15 and Elisabeth, 17. They're so familiar with technology that it scares me. Here's an example. When I first read the academic articles on people adopting false identities in chat rooms, etc, I got very concerned about privacy and identity in cyberspace - I want my children to be careful. Other people they meet there might not be who they say they are, right? So, I sat down with my youngest daughter, Hannah, when she was 11 to explain this complex issue of identity. I try not to give rules to my daughters. I try to explain the context so they can make the right decision on their own. I thought I was explaining something sophisticated and I only got a third of the way through when she said, "Dad, you don't get it. I'm not who I say I am either." She's not going to believe what anybody else says because she has multiple identities herself! Here I thought I was explaining something complicated and she was way ahead of me. It was one of those delightful moments as a parent when your child teaches you something. I think the next generation will have a deeper appreciation and ability to use technology because they grew up with and adapted to it all along.
UBIQUITY: We've talked about artificial intelligence changing. Has PARC changed much?
DE KLEER: Initially PARC was more focused on computers and operating systems per se, on the technology, distributed computing and Internet protocols. The biggest change that happened at PARC over time has been the shift from technology to content. That was both driven the ever-widening frontier of computer science as a discipline and by the realization that Xerox wanted to become the document company. The other big shift, which John Seely Brown helped provoke, is importance of the social in everything. Us technologists tend to be technological imperialists. The future is driven by our inventions. But how do people actually use this technology? What is their work practice? In my lab, I manage PARC's biggest social science research group. Social scientists have transformed PARC. Just their presence provokes everyone to ask the questions, "What will users really think about this? This is a cool algorithm but will people actually use it?"
Right now PARC is going through its biggest change. We've become an independent corporation -- the Palo Alto Research Center. We have the freedom to license to and work with other corporate sponsors. We are creating a completely new business model for breakthrough research. It's really exciting to be on the ground floor of creating a new kind of institution.
UBIQUITY: How do you manage your researchers?
DE KLEER: I have a very simple management principle. I manage people into their passion zone. One passionate person is worth a thousand people who are just plodding along, writing papers and doing hard work. I personally tend to be hands-off on most things. I think a lot of management is learning to leave people alone, but I have to give them a context and a passion zone.
UBIQUITY: How do you help people find their passion zone?
DE KLEER: I work to understand what drives people, what makes them tick, and I encourage them, push them and stretch them into where that passion zone is for them. Everybody's different. When we hire people in my lab, I try to interview each one of them to find out whether they have a mainspring, a passion inside of their hearts. That's where I have the experience and maturity to be able to distinguish an arrogant, aimless person from someone with deep passion that may transform the world.
UBIQUITY: What is your technique for drawing it out of them?
DE KLEER: Often I ask people to tell me about that moment they decided to become computer scientists or social scientists . They talk about their dreams or their relationship with their father, or what happened in high school. People can reveal some delightful experience that reoriented their life. At that point, as far as I'm concerned, they're hired, assuming they passed the other lab members' tests. I make sure that people have a mainspring and then I work to drive their passion.
UBIQUITY: What would be good examples of a passion zone in one of your people?
DE KLEER: One of my social scientists, an anthropologist, lived in Texas during the time of school busing. One morning she woke up, went to school, sat in the front of the bus, and noticed the rest of the kids were sitting in the back of the bus. She said to herself, "What kind of world is it that this happens in?" That's something she wanted to understand. How does the social world work? For me, the fact that she remembered this moment of awareness is symbol for her passion.
UBIQUITY: Do you have an example that shows the roots of your own passion?
DE KLEER: When I was four or five, I decided one day that I was going to write down all the numbers that ever existed. I started writing all the numbers down, and I filled up my notebook. I asked my parents for another notebook and they wouldn't buy me another one, but I went to an uncle who was more sympathetic, and he bought me another notebook. I got halfway through the second notebook when I realized, "This pattern repeats," and it would never end. It was a powerful experience for me that I'll never forget. I think of my uncle being wise and generous and just not sitting me down and saying, "Johan, don't you know there is an infinite number of numbers? Why do I waste 25 cents?" He just gave me 25 cents. It was touching that he did that for me. It's also moving that I figured out infinity by myself, and I wanted more of those discovery experiences over time. That's all it takes. A little bit of self-awareness and some passion, which I call "the mainspring," is what you need.
UBIQUITY: Is artificial intelligence all sweetness and light?
DE KLEER: Oh, no. I want to include one more story on artificial intelligence because it's scary to me. Recently, I was sitting next to a physicist at an interdisciplinary conference. He said part of the problem with social science in this country -- by problem he meant how to teach, how to communicate, how to get people aligned -- is that there needs to be some discipline to social science. He proposed starting with a model of a neuron, and then building up from that using quantum computing. Then, from that, building a theory of the mind. Then, once we have a theory of the mind, we will then know what people pay attention to, how to get people to be willing to be taught in schools, how to get people not to misbehave in inner cities, and so on. This guy is an ultimate reductionist and he's downright dangerous because the one thing I've discovered from social science is that to make any progress on the kind of questions this guy was posing, you must look at it entirely differently. It is the interactions between complex systems that determine what is going to happen, and until we understand what are the categories are of these interactions, we can't even begin to understand the phenomena that we would need to account for, no less build bottom up accounts. Chemistry developed long before physics, and we still haven't completely connected the two. Twenty-five years ago, I might have supported such a reductionist approach. Now I realize why it is ineffective.
UBIQUITY: What made you change your thinking?
DE KLEER: The biggest thing was coming to PARC and watching how people actually use technology and learning to manage and see how organizations actually function. And discovering that all learning is social. Perhaps now I'm being too social, but you have to balance the two. One without the other gets you nothing. Getting back to the physicist: the path he envisions will take far longer than he ever expects. He needs a far deeper understanding of what he is actually looking for. Pure bottom-up approaches have not created the breakthroughs in science, and I do not believe they will succeed in artificial intelligence. Remember, studying feathers and birds did not get us flight.
Ubiquity welcomes the submissions of articles from everyone interested in the future of information technology. Everything published in Ubiquity is copyrighted ©2002 by the ACM and the individual authors.
A Ubiquity symposium is an organized debate around a proposition or point of view. It is a means to explore a complex issue from multiple perspectives. An early example of a symposium on teaching computer science appeared in Communications of the ACM (December 1989).
To organize a symposium, please read our guidelines.
Ubiquity Symposium: Big Data
- Big Data, Digitization, and Social Change (Opening Statement) by Jeffrey Johnson, Peter Denning, David Sousa-Rodrigues, Kemal A. Delic
- Big Data and the Attention Economy by Bernardo A. Huberman
- Big Data for Social Science Research by Mark Birkin
- Technology and Business Challenges of Big Data in the Digital Economy by Dave Penkler
- High Performance Synthetic Information Environments: An integrating architecture in the age of pervasive data and computing By Christopher L. Barrett, Jeffery Johnson, and Madhav Marathe
- Developing an Open Source "Big Data" Cognitive Computing Platform by Michael Kowolenko and Mladen Vouk
- When Good Machine Learning Leads to Bad Cyber Security by Tegjyot Singh Sethi and Mehmed Kantardzic
- Corporate Security is a Big Data Problem by Louisa Saunier and Kemal Delic
- Big Data: Business, technology, education, and science by Jeffrey Johnson, Luca Tesei, Marco Piangerelli, Emanuela Merelli, Riccardo Paci, Nenad Stojanovic, Paulo Leitão, José Barbosa, and Marco Amador
- Big Data or Big Brother? That is the question now (Closing Statement) by Jeffrey Johnson, Peter Denning, David Sousa-Rodrigues, Kemal A. Delic