November 19, 2007
Michael Lebowitz on Universe
My interest in story generation, heightened by work on my book manuscript, has brought me to correspond with the authors of a number of seminal systems. I’ve been posting what I can share publicly, leading to posts about Tale-Spin (1 2) and Minstrel (1 2). Now I’m pleased to add a post with more information about Michael Lebowitz’s Universe.
I outlined the approach taken by Universe last year, in a series of posts contrasting it with Minstrel (1 2 3 4 5). But, as with my writing about other systems, this was entirely based on reading publications about the work — I hadn’t yet heard directly from the authors. In this post I’m including some of the thoughts and memories shared by Lebowitz in our recent correspondence:
I’ve got to admit to being surprised that anyone has run into UNIVERSE after all this time. It was a pretty small piece of work but one that was fun and I rather liked. We could have done a lot more with it these days — we were quite limited by computing power.
In terms of an archive I’m dubious. I really doubt they have it at Columbia. I had ported it at one point to one of my home PCs but that was many computers ago. There’s a small (small) chance I have a version on permanent media at home. I’ll take a look.
In terms of anecdotes around it, there are a few. One of the drivers behind the project was that a friend of mine and I watched Days of Our Lives. The project provided an in to talk to the writers at the show. I made one visit to meet with the writers and got copies of some outlines and talked about their creative process (which is quite mechanical given the volume they have to produce). I was impressed I was talking to real TV writers and they seemed impressed a professor was interested in them.
A project that followed UNIVERSE was with a woman named Mary Milton at the Merkle foundation who wanted to create artistic interactive movies. We had an interesting project group of academics from film, AI, literature. We even did some experiments where we put video footage from Days of Our Lives on videodisk and sliced and diced them to make different stories. The project ultimately died for lack of funding and computer power, but it was quite ahead of its time when you see the directions videogames have gone.
Finally, OMNI magazine published a small fluffy blurb on UNIVERSE. Turns out they have a publicity department and that radio stations have a lot of time to fill So I did a bunch of radio phone interviews on plot generation and even a couple of TV spots. Vastly more publicity than any of my more serious work got.
I’m intrigued to hear about the Merkle project, which turns out to have also involved hypertext fiction pioneer Michael Joyce. Hopefully I’ll be able to track down more information about it. If I can I’ll share it here.
After the message above, I asked Lebowitz a number of questions. I was curious about several issues, including how far development progressed with Universe, what they might have done with more computing power, and so on. Here’s the key paragraph from his reply:
The project never really got very far — it was basically just me and a few undergraduate projects. What it needed was a massively larger database. Even by hand we could have put in a lot more data if we had the computing power, but the way I would have liked to take the project would be to combine the work with what I’d done on Machine Learning and have it build its own DB. Imagine reading the soap plot summaries on say TVGuide.com and automatically process them into fragments that could be reused (and real soaps certainly do reuse plot lines all the time). Ought to work.
Certainly an interesting idea. In my opinion, Universe is one of those intriguing systems that showed exciting early results — and that we could learn more from if it had been completed. What surprises would have emerged as the database expanded? What strengths and weaknesses would we see in Universe if used as a component of a larger system?
November 19th, 2007 at 12:23 pm
A couple of comments…
On one occasion the AI Lab at UCLA hosted a half-dozen or so TV script writers to talk about MINSTREL, etc. I don’t recall their exact affiliation after all this time, but “Merkle” rings a faint bell, so it might have been the same crew Michael mentions. I do recall that nothing became of it :-)
OMNI also published a fairly lengthy article on AI that mentioned some of the work from the UCLA AI Lab. IIRC, I wasn’t around on the day the reporter visited the lab, so no mention of MINSTREL. I believe that Erik Mueller’s work was highlighted.
It’s interesting that Michael notes one of the same stumbling blocks I highlighted — the lack of a very large database to fuel the creative/authoring process.
November 19th, 2007 at 1:59 pm
As a current Columbia grad student I’d be happy to poke around and see if anyone has an archived copy of UNIVERSE, if Michael can give me a place to start. It would likely be with a story generation fan, since most people’s code seems to wither away after graduation…
November 19th, 2007 at 5:09 pm
I just heard from Michael Joyce, who tells me that it’s the “Markle Foundation” (with an “a”) and reminds me that some of his work with them is discussed in chapter two of his book Of Two Minds (which I haven’t read for years and should clearly pick up again). Apparently the first funding for Storyspace came from Markle as well — and this year is the 20th anniversary of Storyspace and Michael’s groundbreaking fiction afternoon. (There will be a panel at this year’s MLA to mark the occasion: “Electronic Literature: After Afternoon.”) He also sent me some further info on which I’ll be following up.
David, I’ll drop Michael Lebowitz a note and ask if he can send you any leads. Thanks for the offer!
Scott, I think you’re absolutely right. One of the problems with story generation projects in general may be that, from a computer science standpoint, there’s no institutional reward for time spent authoring data – and it’s very time consuming. But from a media-making perspective, it’s absolutely essential. But in many cases we can’t simply bring on media makers to create content for these systems, because the authoring requires a deep understanding of how the system actually functions – and most media makers aren’t very “procedurally literate.” I understand this is part of what has lead this blog’s Michael (Mateas) to start thinking seriously about how we could create AI-supported tools for authoring content for AI-driven systems…
December 16th, 2007 at 8:08 pm
[…] Tale-Spin (1 2), Scott Turner’s Minstrel (1 2), and Michael Lebowitz’s Universe (1). Now I’m pleased to continue the series with some information from GTxA’s own Andrew […]