IJCAIs I Have Seen (a quick prelude to "Goodbye, Minsky")

I am intending to write a longer blogpost called "Goodbye, Minsky" probably on 7 February, but I can't now for a number of reasons, one of which is background for that post, and more importantly, relevant to a discussion I had with some of my PhD students earlier this week.  The past only matters to the extent that it informs the future, and my students' careers are part of the future, so here we go.  Ah, I forgot to say, the reason I'm postponing a Minsky post is because of the IJCAI and Cognitive Science conference deadlines, which are always at the end of January.  And this post is about IJCAI, why I think it matters, and how it's changed since I've been in AI.  This will all be background also to thinking about Marvin Minsky's career.

Aside from tutoring Kris Hammond's first course at Chicago in the fall of 1986, I first entered AI by taking a conversion masters to Knowledge Based Systems in the University of Edinburgh's Department of Artificial Intelligence in 1991-1992.  There were four possible concentrations, and I took the one recommended for those seeking to do PhDs rather than go into industry, "Foundations of AI".  Years later the department surveyed alumni to see if we minded if they changed the name of their MSc back to AI, I said "that's what's on my CV already".

My research supervisors at Edinburgh (Alan Smaill & Geraint Wiggins) were impressed enough by my dissertation to
  1. write the letters that got me into MIT, but also
  2. suggest I submit it to IJCAI.
IJCAI was the top international conference in the field, the "International Joint Conference of Artificial Intelligence".  "Joint" because it was biannual.  Every other year was meant to be regional–Pacific, European (ECAI) and American (AAAI), and every other year was meant to be global.  Recently however the Americans had defected (as they often do) and gone to an annual meeting, and stopped attending IJCAI as much, meaning sometimes AAAI was the larger meeting.  But IJCAI was still the real pinnacle as far as Edinburgh was concerned, and all my other friends from my MSc were dead impressed that I was submitting to it.  I was less sure it was a good idea, and indeed my paper didn't get in and many of theirs did get published in whatever "lesser" meetings they'd applied to.

When I got to MIT, I was told that if I wanted military funding (the stuff the AI Lab was built on) I should attend AAAI.  The military wouldn't send their own people out of the country, and objected to funding people for travel out of it either.  IJCAI was just for people who cared about AI.  Consequently, I've been to IJCAI six times, and never attended the main meeting at AAAI.

An important thing to understanding the below stories is to know that a focus of IJCAI is three awards:
  1. Computers & Thought Awards (CTA, joke name: junior achiever's award).  These are amazing people changing the world & seeking tenure.
  2. The Award for Research Excellence (ARE, joke name: lifetime achievement award). We used to think of this as history.  I guess I still do, I'm just more impressed by history now.
  3. The Distinguished Service Award (DSA, joke name: um, the poo-work award (more or less)).

Here are the IJCAI's that I've attended or that affected me.
  • 1991  Rod Brooks won the CTA.  Or at least, everyone at Edinburgh thought he had.  In fact, it was a contested decision and he shared the award with Martha Pollock.  If you read her award paper, you'll learn about the controversy happening in AI planning in 1991.  If you read Rod's paper, you will think he is the anointed one leading AI out of the ages of alchemy and into the age of reason.  All the cool kids at Edinburgh were passing around Rod's paper, so that was all I knew for some time.  In fact, when I first met Rod Brooks he was surprised I didn't have any questions I wanted to ask him, and I said "I used to have a bunch of questions, but then I read your IJCAI paper."  He said "you're making that up," but I wasn't.
  • 1993 I submitted The Subsumption Strategy Development of a Music Modelling System.  It didn't get in.
  • 1995 was my first IJCAI.  I was being mentored by Ian Horswill and Maja Mataric, two senior PhD students at MIT, and watched how they put in at least two papers each, and proof read their papers for them.  I was astonished that part of Maja's paper was an informal promise that in three months there'd be another paragraph about new results at that point.  Everyone said getting in was a crapshoot, and indeed Ian thought it was his weaker paper that got in, and tried to cram all the best results from the other paper into the one that did get in.  Then he got his new job at Northwestern and was moving to Evanston the week of IJCAI, so MIT sent me to Montreal to give his paper, Visual routines and visual search: a real-time implementation and an automata-theoretic analysis. Herbert Simon got the ARE that year, and his talk, Explaining the Ineffable: AI on the Topics of Intuition, Insight and Inspiration, was amazing.  I bought the cassette tape and still have it.  Marvin Minsky was there too, and this will come up in Goodbye Minsky.  It was also Push Singh's home town, and he will come up too.
  • 1997 I submitted something that didn't get in, I've forgotten what, probably Specialized Learning and the Design of Intelligent Agents.  But I met this extremely attractive guy that for his MSc had found an automated way to solve the problem of creating semantic nets that IJCAI people had spent decades on and told him he should be submitting as well.  Of course, he got in.  Many people wound up scraping together money and eventually Will Lowe got sent to Kyoto to present Meaning and the Mental Lexicon.
  • 2001 I finally got my own paper into IJCAI and presented it in Seattle:  Modularity and Design in Reactive Intelligence.  Some annoying person whined her way out of a terrible time slot on the last day and I got switched into it and out of the session that made sense for my paper, so hardly anyone saw my presentation.  Daphne Koller gave an amazing CTA, and Bill Gates also gave a memorable invited talk (not for any award.)
  • 2005 Edinburgh.  I decided to run Modelling Natural Action Selection: Proceedings of an International Workshop at IJCAI.  I invited Tony Prescott to join me, because I thought his work was awesome and couldn't understand why he wasn't at least as famous as Randy O'Reilly.  IJCAI required us to have three people, so Tony invited Anil Seth.  Unfortunately, that made the meeting more neural oriented than I'd intended: I'd hoped to compare a number of approaches but wound up having to shoehorn in the few ABM papers allowed (despite the fact I later learned they'd had great reviews.)  The whole thing was so tiring I wound up blowing off most of IJCAI and just enjoying Edinburgh, but I did see Donald Michie get the ARE.
  • 2007 IJCAI went to India for the first time, and I got two papers in with two PhD students: Edinburgh's Alan Bundy got the ARE (making him one of very few people to get that and the DSE) , and what was striking to me was that he did an amazing job of relating what he did (mathematical reasoning) to what he thought IJCAI did (Rod's "new AI"), but as far as I could tell I was one of about 4 people who understood the talk.  The meeting was about 1/3 Bayesian machine learning people who were utterly confused by the talk.  It was also about 1/5 MAS people lead by Mike Wooldridge – I hadn't realised those guys were doing so well.  Another striking thing was how amazing the career of the guy who got the new DSE award had been, even though I'd never heard of him and it was pretty much purely administrative.
  • Also in 2007 I decided I wanted to get serious about doing science and switched to trying to get published in science journals, not AI conferences.  But in 2011 I fed the gods of the British Research Assessment Exercise with another IJCAI publication, Just an Artifact: Why Machines are Perceived as Moral Agents, which Phil Kime and I had intended to try to get into a journal.  This meeting was in Barcelona.  Other than that, basically nothing interested me.  Everyone in my session was incredulous that a paper without data or maths had gotten into IJCAI.  Clearly, IJCAI and I had grown apart.  I sat in the huge lecture theatre where CTA recipient described his capacity to spot terrorists in airports (the topic of about 70% of the papers) and thought "I guess I'll never hit the main stage of IJCAI", and was sort of sad, even though I massively preferred the meetings I had been attending and speaking at recently, like the European Federation of Primatology and the European Human Behaviour Evolution Association.
  • 2013 I was invited relatively late to participate in a plenary panel in Beijing,    I didn't really want to go somewhere I couldn't use twitter, but I also knew women turn down talks too much, so I went.  And so I have talked on the IJCAI main stage, to a full audience.  I also sat in a number of panels where people were giving real science talks about cooperative behaviour, because it was a lot easier than getting in science journals, and computer scientists love that stuff.  Hmmm... In these talks, I also saw a lot of Americans talk about how wonderful twitter was because it created the Arab Spring, apparently oblivious to what was happening in Egypt at the time, or what their largely Chinese audience thought about governments collapsing into chaos.
  • 2016 Now IJCAI has gone annual too, and I'm on the program committee for another workshop, Ethics for Artificial Intelligence.  I have mainstream AI PhD students again, and we went for the main conference deadline but will miss it.  But I'm still hoping to submit something to Cognitive Science.  We'll see...