Monday, June 25, 2007

Fee fie Foo done

I just got back from Foo Camp, Tim O'Reilly's "Friends Of O'Reilly" conference.

It was an interesting event, pretty much as described, a self-organized, somewhat chaotic blend of "people who're doing interesting works in fields such as web services, data visualization and search, open source programming, computer security, hardware hacking, GPS, alternative energy, and all manner of emerging technologies" who sat down to chat, debate, "share their works-in-progress, show off the latest tech toys and hardware hacks, and tackle challenging problems together."

Most incredible was the diverse group of attendees, ranging from university professors to tech gurus to venture capitalists to goofy little startups. In addition to the various tech celebrities -- Larry Page, Caterina Fake, Paul Graham, Ray Ozzie, Kevin Rose, to name just a few -- there were even folks such as Wes Boyd, founder of

Part of the experience is the opportunity to bump into random people and explore various ideas. Joe McCarthy and I discussed innovation at startups versus innovation in research groups. I talked to Paul Kedrosky about trying to use information on the web for hedge funds. I had an extended discussion with Wes Boyd about the future of journalism. Steve Yegge convinced me not to hate Javascript quite so much. Udi Manber and I discussed his departure from A9. Nat Torkington and I talked about environmental and energy policy. Mark Atwood and I argued about the short-term prospects for utility computing. And, there were many more casual conversations on many more topics.

It was good to see old friends and colleagues. Amazon and folks included Russell Dicker, Kim Rachmeler, H.B. Siegel, Shel Kaphan, Udi Manber, DeWitt Clinton, Peter Vosshall, Chris Brown, and Steve Yegge. I finally got a chance to meet Paul Kedrosky, Garrett Camp, Doug Cutting, Bradley Horowitz, Greg Stein, Marti Hearst, Om Malik, Stewart Butterfield, Matt Cutts, Nat Torkington, Artur Bergman, Luis von Ahn, and Don MacAskill face-to-face. It also was good to see Danny Sullivan, Tim O'Reilly, Peter Norvig, John Battelle, Niall Kennedy, Mez Naam, Jed Harris, and Brian Aker again.

The many talks, most of which take the form of a discussion rather than a lecture, were remarkable as well. Sadly, there were often three or four talks in the same time slot I wanted to attend -- so much to see, so little time -- but I was able to attend and enjoy many.

For example, Mez Naam gave a fun, SciFi-like talk on what happens as 3D printers become cheaper, smaller, higher quality, and widely adopted for manufacturing. In the near term, we may see some goods reduced to information -- all you need is the blueprint for what to print to make your very own iPhone -- which could cause serious disruptions in some industries and much confusion for intellectual property laws. In the much more speculative longer term, Mez asked, what might happen if people can create drugs, even pathogens, at their desktop with cheap hardware?

Researchers Marti Hearst, Martin Wattenberg, Fernanda Viegas, Jeffrey Heer talked about data visualization, focusing on demos of Many Eyes and The talk explored how easy data visualization and sharing tools help people collaborate and learn from data. A very cool idea was the ability not only to comment on the graphs, but also draw on the graphs and refer to other graphs, facilitating discussion and exploration. Marti Hearst also briefly discussed tag clouds, ending with the thought-provoking conclusion that tag clouds are intended not as a particularly useful method of conveying and summarizing information, but as a means of socializing among people.

Researcher Andrea Thomaz from the MIT Media Lab showed off videos of Leonardo, a robot designed with gestures that naturally appeal to and are easily interpreted by people.

Researcher Neil Halelamien from CalTech discussed how placing a rapidly fluctuating magnetic field at the back of someone's head can stimulate neurons on the surface of the brain and create some unusual (and temporary) visual effects involving replay of images just seen.

I sat in on a conversation with Stephen Hsu and several other folks working on computer security that came to the rather dismal conclusion that not only can we expect severe, large scale botnet attacks in the near future, but also we can expect a future where most computers have some low level of infection by malware (much like the human body has a continuous, low-level infection by viruses and bacteria). Some of our discussion is similar to what appeared yesterday in the NYT, "When Computers Attack", which quotes one of the Foo campers, Ross Stapleton-Gray, at one point.

There was a discussion of the book Paradox of Choice -- which argues that more choice can make it difficult to take action and that overoptimizing choices makes people unhappy -- led by H.B. Siegel and including Flickr founders Caterina Fake and Stewart Butterfield. The discussion focused mostly on personal experiences, but some interesting meta questions about the economic rationality of optimization -- cost of time for gathering information versus the cost of a (usually only moderately) sub-optimal choice -- were also raised.

Toby Segaran lead a discussion about wisdom of the crowds and hive mind that, at one point, dived into fun questions of whether hive mind communities suffer from tyranny of the majority and end up fracturing at a certain size. Digg came up several times as an example of wisdom of the crowds, tyranny of the majority, and a hive mind that might fracture.

Peter Norvig gave a great version of his talk on the advantages of big data for solving many types of machine learning problems. The machine translation examples are particularly compelling. If you want to check it out, the talk was similar, though not identical, to some of Peter's talks I linked to in an older post.

Finally, I very much enjoyed a session with Erick Wilhelm, Dennis Cramey, and Will Carter talking about location-aware gaming. The basic idea is to have the virtual game world overlap with the real world. Initially, this has taken the form of games where the real world is used for navigation -- moving in the real-world moves you in the virtual world, but the virtual world is otherwise separated from the real-world -- but there was some fun talk about how the worlds could be blended further. What I would really like to see here is a game where you are essentially someone different in the real world (e.g. a secret agent) and interact with others in the game through your device and through the real world (tasks, information drops, puzzles). It would be like the cell phone is your access into a different persona, but that persona exists both in the real and virtual world.

In all, a very interesting and unusual experience. I am still not sure how I managed to get invited, but it was great to get a chance to go.


Bradley Horowitz said...

Great meeting you Greg. Foo Camp indeed rocked.

jeremy said...

Joe McCarthy and I discussed innovation at startups versus innovation in research groups.

I'd be interested in hearing a little more about the various pros and cons of each that you two were able to enumerate.

Pater Norvig

Is this Peter's dad? ;-)

Jeffrey Heer talked about data visualization, focusing on demos of Many Eyes and

Yeah, I saw this presentation at CHI '07. Very interesting stuff. Love the notion of exploratory search.

Greg Linden said...

Doh! Thanks for the spelling correction, Jeremy. Fixed!

As for the conversation I got to have with Joe McCarthy about startups versus research groups, there were two main threads.

The first was that being unfamiliar with previous work sometimes may be an advantage. The concern is that reading previous work can force a particular way of framing a problem and a certain set of core assumptions, all unconsciously, and prevent adequate exploration. In some sense, the argument is that not knowing what you should and shouldn't be able to do may be useful for creativity. By the way, I am not sure I agree with this point of view -- it strikes me as too close to ignorance is bliss -- but I think it is a very interesting debate to have.

The second was that startups (if defined broadly to include Facebook, My Space, Digg, and even Google) have access to very large data sets that are unavailable to academic researchers. Tying into Peter Norvig's common theme, big data may result in more progress on some types of problems than better algorithms, so the lack of the data in research labs may hamper their ability to innovate in some areas.

There was also a much smaller thread at the beginning talking about different time horizons for startups and research labs and the impact that might have.

jeremy said...

w/r/t "ignorance is bliss", I think both sides are not incorrect. In my own experience doing music information retrieval for many years, I worked with a diverse team of computer scientists, digital signal processing engineers, and musicologists. Though I do have some rudimentary musical knowledge, I am not a musicologist myself, by any stretch of the imagination.

As a result, much of my ignorance about music theory led me to try music search approaches that the theorists considered "silly" and that shouldn't be attempted. Well, many of those ideas ended up working quite well. So not having those pre-conceived, domain specific barriers to my thought process was an advantage in that situation.

On the other hand, imagine a text-IR educated venture capitalist who is being approached with the idea for a startup that does "personalized stemming". Generic stemming algorithms may conflate "car" and "cars", but not "new" and "news". "Our idea," this startup might say, "is to personalize these stemming algorithms, so maybe for you, "car" and "cars" do not conflate, whereas for me, "new" and "news" do conflate. At that point, the VC should probably reject the startup's funding request.

Why? Because over 30 years of researchers, both in academia and industry, have experience showing that, on average, stemming does not improve retrieval performance. So how does the startup really expect personalized stemming to do that much better? Is there really someone out there that needs to conflate "new" and "news"?

While ignorance does not guarantee that the musically-naive algorithm designer will succeed, nor that the stemming-naive startuper will fail, what it boils down to (IMHO) is different types of ignorance. In the former case, some silly ideas had never been tried and their validity, despite sounding silly to a less naive expert, was unknown. In the latter case, some non-silly ideas had been tried over and over and found not to work. I think it is much more dangerous to be ignorant of the latter than the former.

But certainly it sounds like you had some great conversations there. Thanks for sharing!

Pascal Van Hecke said...

"some interesting meta questions about the economic rationality of optimization -- cost of time for gathering information versus the cost of a (usually only moderately) sub-optimal choice"

Sounds like optimizing optimization effort :-)

Pascal Van Hecke said...

On the Peter Norvig talk: there's a version on Youtube (Google Developer Day):