ad info

 custom news
 Headline News brief
 daily almanac
 CNN networks
 CNN programs
 on-air transcripts
 news quiz

CNN Websites
 video on demand
 video archive
 audio on demand
 news email services
 free email accounts
 desktop headlines

 message boards




Dr. James Hendler

A chat about the future of artificial intelligence

January 1, 2000
Web posted at: 1:44 p.m. EDT

TEST CNN Chat Calendar
How to Connect
Message Boards
Time Zone Converter
Discussion Standards
(CNN) --- Dr. James Hendler, Program Manager at the Defense Advanced Research Projects Agency (DARPA), and Professor of Computer Science at the University of Maryland, joined to chat about artificial intelligence as part of as our @2000 chat series with leading authors, historians and experts to contemplate life at the turn of the century.

Dr. Hendler was the recipient of a 1995 Fulbright Foundation Fellowship, is a member of the US Air Force Science Advisory Board, and is a Fellow of the American Association for Artificial Intelligence. He has authored over 100 technical papers in artificial intelligence, robotics, intelligent agents and high performance computing.

Dr. Hendler joined the chat from Maryland on December 16, 1999. The following is an edited transcript.

Chat Moderator: Thank you for joining us today, James Hendler, and welcome to chat!

Jim Hendler: Hello, I'm here and looking forward to our chat.

Chat Moderator: Please tell us a little bit about your background.

Jim Hendler: I am a professor at the University of Maryland where I've worked since about 1986. At the university I run a couple of research groups, one that does AI research and another that does robotics research. I grew up in New York, went to College at Yale, and then worked for a few years for Texas Instruments before returning to grad school at Brown.

Question from McKennaWA: How do you define Artificial Intelligence?

Jim Hendler: This is a much harder question than you might think. One answer I like a lot, although it isn't very precise, is "what computers cannot do yet." The problem with that is it admits a lot of things that aren't AI, so I guess the better answer is to say that there are three main schools of AI: people trying to model what humans do (sort of psychology based), people trying to make what people do easier and better (tools for humans), and people who are trying to build new tools with "far out" capabilities. I tend to be a little bit in all three of these, with a preference for the third.

Question from Beach Girl: What advances have been made in the past ten years in AI and what can we expect in the next ten years?

Jim Hendler: Past ten years is easier to answer . The "embedding" of AI in a lot of things is happening. Here are some examples you might not be aware of.

That little paperclip in Microsoft Office uses a Bayesian "belief" network. The tools that underline your misspellings in MS Word grew out of AI research in language. Tax preparation software comes from rule-based AI, and many many more. Your car probably includes some control mechanisms in its inner workings that grew out of AI labs, as does your TV and many other products.

Next ten years, I think the real focus of most AI you'll see is on the Internet, speech recognition, language translation, better search, and much more. There's also some work in robotics that you'll be seeing in your kid's toys. Furby is the tip of this iceberg.

Question from McKennaWA: Do you believe that a machine will ever be capable of un-doubtably passing the Turing test?

Jim Hendler: Quickly, for those who don't know what the Turing test is, imagine you have two rooms. You're in one. In the other is a person pretending to be someone else, or it may be a computer pretending to be a person. You're allowed to type back and forth. If you cannot eventually tell the human from the computer, it passes the test.

I believe that for limited domains, i.e. if we restrict the inputs to, say, sports, economics, etc., we're already pretty close. In open domains, we're a long way away. But I have a very different question. Why would we want a computer to pass the Turing test? If it can just be mistaken for a human , why not use a human?

Question from Beach Girl: How involved is Natural Language Processing in the field of artificial intelligence?

Jim Hendler: It depends who you ask. The field itself was/is an integral part of basic AI. Many exciting results have been achieved in language translation, test understanding, etc. At the same time, there are now a lot of people doing NL-related work using techniques that are not coming from the AI community (statistics, information retrieval) that have a lot of interesting properties. If we view NL as the union of all this work, then AI is one key component, but not the whole ball of wax.

Question from Joshuadearing: How far is AI from games like Quake 3 etc.?

Jim Hendler: An interesting question. A few years ago some undergraduates in my robotics course did an independent project . They built a program that played Quake (not yet 3) for a couple of levels. By the end of the term, it beat every human it played against, and they put it out on the net where it played a pretty good game (limited to those levels).

That sort of thing, and many more ambitious projects, has convinced the game manufacturers to look very heavily at AI. This is a booming area, and I suspect we'll see a lot of it in the future.

Interesting thought. In some of these games the problem is that the artificial players are so good that they make it boring for humans who can never win, so a research issue is how to "dumb down" these players.

Question from McKennaWA: Do you believe that machines will ever become "aware"? In any sense of the word?

Jim Hendler: This is a very deep question, and I have an answer which may seem odd. I don't think it matters. Let me put it this way, is your cat aware? Yes, in some senses, no in others. So maybe the question is will we ever think that our machines are aware. I think the answer to this is yes, we'll see more and better capabilities that we tend to attribute as awareness. That said, I don't think machines will ever have "human awareness" in the philosophical sense of the term, but they'll be awfully close some day.

Question from Davspa: How close are we to easy to use programs directed through voice input?

Question from Beach Girl: In the field of artificial intelligence, as far as Language is involved, do you still have problems wrecking nice beaches?

Jim Hendler: I'm going to answer these together, beaches first. (Beach Girl obviously knows something about this field.) "Wreck a nice beach" and "recognize speech" are two phrases that are phonetically very similar, even though they have very little meaning in common.

Anyway, the answer is that speech input is already here, commercially available, and increasing in capability at a phenomenal rate. Many telephone services use speech recognition now, and more will come. Much more exciting is that we're approaching (I'll guess 10-15 years, but I'm being conservative) a time when language translation can also happen during that speech recognition . (I'll speak in English, you'll hear it out in Japan).

I've already seen demos of this in some large corporation research groups. However, robust general speech is still a hard problem, so I think this is one where we'll see increasing competence for a long time to come.

Example: your current output screen is probably a million times better than the one I used when I first started playing with computers. And the future ones will be even better. So at what point do we say "screen technology has no more problems?" We always want more.

Question from Wei: What is your take on the usefulness of machine learning with the subject of intrusion detection ?

Jim Hendler: Let me split that in two parts. With respect to machine learning (ML), I think that it is another limited, but improving rapidly technology. I think that for what some people call "automatic intrusion detection" (protecting your computer network from unwanted users) there is a lot of potential, and need, for this technology, but it is not yet ready for prime time.

Question from Sanj: Where do you see robotics headed?

Jim Hendler: Oh, thank you, a question I love answering!! I think the exciting frontier is that robotic technology's becoming lower cost and much more widely spread. Some stuff I love -- if you go to my web page and then to my "robotics lab" you'll see a lot more about this -- is the idea of robotics for kids.

Consider that a big challenge with current technology is that to interact with a computer, and thus to the net and etc., a kid must come sit at a terminal. What a boring learning experience!! But if, as it were, the kid could come to a place where they interacted with information, or where the terminal could come to the kid, what neat ideas this opens up.

I have some amazingly creative graduate students working jointly with a group of 8-12-year-old kids, and they are doing some things that really show that the merging of information technologies and robots in education could be exciting and revolutionary.

Question from LeAnthony: Can you talk about the product of the Cyc project?

Jim Hendler: Okay. CYC was an effort starting about 15 years ago now, led by Doug Lenat, which aimed at building a big common-sense "knowledge base" (i.e. the enCYClopedia) in a way machines could use it for inference and etc.

They've somewhat changed goals, but have achieved some exciting things. In fact, one large web search engine is starting to work with them to use that knowledge to help in better searches. So if you ask the search engine to search on "mouse" it will come back with "the animal or, the input device"? And depending how you answer, it will generate a better search that will get you better answers.

Question from Infuryum: How involved is NASA in the development of AI for space-exploration oriented uses?

Jim Hendler: Very. One obvious use was the Pathfinder robot, which used a number of modern robotics techniques to be able to explore the surface of Mars. Another is the Deep Space One mission in which an "autonomous" controller, i.e. without human intervention, was able to fly a spacecraft for part of a mission , and exceeded all performance goals.

NASA is exploring this technology heavily, based on these early successes, and is a major player in some of the exciting new innovations. JPL and Ames are the leaders in this. Their web pages take you to some great sites

Question from Johnny99: So is the Robin Williams movie plausible?

Jim Hendler: I have not seen the movie (sorry to admit it), but from the commercials my answer is no. However, a better answer is "not yet." And if you joined this chat early you know that I said "AI is what computers can't do yet."

I think the science fiction writers are more in touch with where AI is going, and if any students are out there use these things as inspiration, I'd love to build "Lieutenant Commander Data" some day. And while it may be beyond my lifetime till it happens, if ever, aiming at a fun goal like this makes AI one of the most fun fields there is to be in.

Chat Moderator: Any final thoughts?

Jim Hendler: This was fun, I hope CNN will invite us back to continue. Again, if there are students out there, AI is one of the best fields in computing to be in. Go for it!

Chat Moderator: Thank you, James Hendler, for chatting with us today!

Jim Hendler: My pleasure. Good bye.

  • Go to our auditorium chat room
  • Check out the CNN Chat calendar
  • Participate in our @2000 message boards

  • In-depth: @2000
  • In-depth: The Next Millennium: Now what?
  • Hendler's bio

  • Note: Pages will open in a new browser window
    External sites are not endorsed by CNN Interactive.
    Enter keyword(s)   go    help

    Back to the top   © 2001 Cable News Network. All Rights Reserved.
    Terms under which this service is provided to you.
    Read our privacy guidelines.