Last Modified (quickly) February 24, 1997

Joanna J. Bryson

Free speech online

Hi there.

My name is Joanna Bryson. I work at the Artificial Intelligence Lab at MIT. I am a candidate for a PhD (expected within the millennium), under Gill Pratt. I'm associated with The Leg Lab, but I'm currently in Edinburgh. "There Learning with his eagle eyes / Seeks Science in her coy abode." -- Robert Burns, Address to Edinburgh

Quote of the random time interval: "What we call a mind is nothing but a heap or collection of different perceptions united together by certain relations and suppos'd, tho' falsely, to be endow'd with a perfect simplicity and identity." -- Hume, quoted by Minsky


Research Interests

I'm interested in how intelligence works, in people and in animals. (I don't think you can understand the former without the latter!) For several years now I've been working with a distributed model of intelligence called reactive, behavior-based reasoning. This approach has produced the first robots that can reliably interact with the real world at realistic, animal-like speeds. However, a decade after the approach was introduced at MIT, there still aren't many robots doing things much more complicated than navigation, obstacle avoidance and pursuit of visible goals -- ameoba intelligence! For my PhD, I'd like to fix that.

I'm think I'm making progress on both understanding and solving the problem. The best / most recent summary of my work is my IJCAI97 submission. It covers the following issues, and presents (well, introduces) my solution to them.

  • One well recognized problem is that without internal representations, it's very hard to learn anything. I wrote a paper about what kind of learning you need for robots, looking at what you have on animals. It's called The Design of Learning for an Artifact. (There's also an older, longer version about Cog.) This paper explains that learning is not a single monolithic system, but must be as modular and intricate as the rest of intelligence. I'm currently working on a mapping and navigation algorithm, which uses dual rates of learning.
  • Another problem with particularly subsumption architecture (the dominant approach to behavior based programming) is its lack of goal flexibility. That's just a page of html I wrote a while back on the issue.
  • Behavior based systems also tend to lack adequate control state (you could think of it as short term memory) to allow for complex task completion, because complex activity tends to require propogating decisions through selective attention to appropriate behaviors. I've written a paper about this problem, The Use of State in Intelligent Control . It mentions a solution I've been developing, which at the time worked in blocks world. I now have it running a Nomad mobile robot, see the IJCAI paper above.
  • They are too hard to program. I'm beginning to suspect that a lot of the difficulty we're having with controlling parallel systems is unnecessary, and my aforementioned system addresses this to some degree. I'm trying to use my experience in consulting, teaching and technology transfer to develop a reasonable programming methodology other people can use.
  • Everyone should think about the ethics of what they do professionally... right? Well anyway, I wrote a paper about the ethics of AI. It's called Just Another Artifact. I wrote it with my friend Phil Kime. If you're worried about whether AI's will take over the world in the next 30 years, or even whether its ethical to unplug robots, take a look! Feel free to argue with us, just send mail.

    That paper partly came out of being associated with Cog; here are some more of my papers and words on the subject. .

    People used to think you could only do boring, reflexive "physical" type tasks with a reactive, behavior-based approach. So I previously applied it to the fairly interesting intellectual activity of music in The Reactive Accompanist.

    Here are some other related web sites I've stumbled across.


    Other Stuff

    This term's source of income... I'm doing research for Brendan McGonigle in the Intelligent Systems Laboratory at the University of Edinburgh. Sorry about the lack of links, we're working on it! I've previously TAed undergraduate AI (6.034) in case your curious what that's like at MIT.

    I have RSI, clicking there will get you to a good information server about typing injuries. I have a datahand keyboard, clicking there will show you a picture of one. If you work in MIT-NE43, you might want to join RSI mailing list that I set up. If you are in AI, just edit the alias file, otherwise you can email me.

    The Artificial Intelligence Lab / Lab of Computer Science Soccer teams The Cold Booters kick grass. I wrote the Cold Booters C League Home Page, and I'll sign up anyone who emails me to whichever league they like. We practice occassionally all year, but the IM season propper is Fall term.

    Even though hockey is evil (it takes people away from soccer!) I am also playing with Halting Problem, which I only admit because it's such a great name.

    Politics... Check out the City of Cambridge home page which has links to the US federal government and MA state government as well. The local phone numbers for MIT area Congresspeople are Ted Kennedy 565-3170, John Kerry 565-8519 and (Rep for Cambridge and Somerville) Joe Kennedy 252-0200.

    I guess everyone should have easy access to this by now, but I find it useful to keep the whole world right here. My Dad says to warn you that page is really, really big. Parents on the internet, I dunno... For some more useful stuff, check out Mike Wessler's home page, he collects pointers more diligently than me (well, he writes some of the best!)


    My Past

    In case you are trying to figure out if you know me, or if you just want to know about some really cool people and places, here's my life's history:

    I am late of Department of AI at U of Edinburgh, LaSalle Street in Chicago, The College of the University of Chicago, and Glenbard North Highschool in Carol Stream, Illinois (which is cooler than you think.) I spent the late 60's in Omaha, Nebraska (but I don't remember them,) and I was born in Milwaukee, Wisconsin, like Herbert Simon.

    Here's my resume', in case you are looking for excuses to give me money.


    Acknowledgments

    Thanks to Andrew Fitzgibbon for telling me "millennia" is plural.

    The guys at the Computer Vision Group at Trinity College Dublin first harassed me about not having a Web page, so now I do, and in return I'd like to suggest the world drop by and ask Gerry Lacey where's the best place to go for a pint.

    I must acknowledge Will Lowe, though I can't say for what in public. (Parse that in British!)


    Joanna Bryson (This pointed to my finger back in pre-spam, pre-hacking-paranoia days)
    MIT Artificial Intelligence Laboratory
    545 Technology Square, Room 006
    Cambridge, MA 02139
    (617) 253-2475
    or, for the time being
    Intelligent Systems Laboratory
    Appleton Tower, level 8
    University of Edinburgh
    Edinburgh, EH8 9LE, UK
    [44] (131) 650 4354