CI Foo - Day 2
Let me preface this post by saying that I’m fried. Having so many and such intense conversations over a two day period, overlaid with jet lag, leaves one feeling like a dishrag. That said, CI Foo was a huge success as far as I’m concerned and I got so much out of meeting so many bright, insightful people willing to share. Hopefully I was able to hold up my end of the bargain.
930am - sat in on a session discussing “bad actors” among the collective, i.e., those who put up false reviews, try to game the system, etc. Preston McAfee of Yahoo (on loan from Caltech) steered the bus for this discussion, but was heavily supported by John Riedel of University of Minnesota. Using Amazon as a benchmark, we talked about large vs. small comment forums and the likelihood and ability to game the system. The group pretty much determined that the “gaming the system function” is like a hump, with small sites and large sites less impacted by bad actors than those in between. Why? Because sheer numbers make large sites with large amounts of comments like Amazon hard to taint, while small, niche sites aren’t often the focus of bad actors. Those in the middle, however, are clearly vulnerable.
10am - Dave McClure’s big ideas around the concept of “securitizing happiness.” In essence, Dave’s goal is to try and use financial innovation as a vehicle for helping several under-served constituencies: lower-income individuals and families (to get better care in order to live longer, healthier, more productive lives); small businesses (to gain easier access to growth capital, akin to the role played by FNMA in mortgage lending) and others. While I’m not sure that some of his financial approaches are necessarily the right ones to address the issues, he is focused on critical issues that warrant additional brainstorming. And I am pleased to help him think through his noble mission.
11am - Identity, Ontology and Fraternite in the context of Wikipedia, facilitated by two smart dudes from Metaweb, Robert Cook and Kurt Bollacker and another smart dude, John Riedel (who was extremely active in countless sessions). After gaining a better understanding of the structure of Wikipedia and the underlying editorial process, we discussed the profiles and statistics of key contributors as well as bad actors and their impact on the overall corpus of data. Bottom line, the numbers are pretty small but rising. It gave me a much keener understanding of Wikipedia, its construction and the issues surrounding its growth and maintenance.
130pm - Prediction markets and elections, run by Eric Zitzewitz of Dartmouth. Eric ran though a slew of data looking at sports and financial prediction markets and compared it to election prediction markets. Bottom line, sports and financial prediction markets are highly efficient and closely track actual outcomes, while election prediction markets tend to systematically underprice favorites. This is the way it is in the US. In the UK, where election prediction markets have existed for over a century, they are much more efficient. Two explanations for the US election markets’ relative inefficiency are low repetition vs. high repetition domains (sports and financials have large numbers of contests while elections have far fewer contexts), and that bettors in the US have far less collective experience betting on elections than their UK counterparts. Interesting stuff.
130pm - Modeling surprises by Eric Horvitz of Microsoft Research. I only caught the tail end of Eric’s talk but from what we discussed following his presentation I was incredibly excited. I am all about understanding tail risk and also its converse, identifying tail opportunities. I will follow up with Eric in our post-Foo lives. This is one of the great things about Foo - you meet people whom you likely never would have met any other way.
Here is a smattering of other issues that were raised during CI Foo:
- Distributing big tasks among many
- Economies of scale in knowledge
- Parallel vs. aggregate knowledge
- Scale - need taxonomy, time and # of people
- Create an ontology of classes of CI problems
- Connecting opposite ends of the CI continuum - top down/bottom up and implicit vs. explicit
- Collective intelligence vs. collective will
- Reputation
- One person’s implicit data is another person’s explicit data
- Topology of community members
- Active vs. passive collective intelligence
- Mere parallelism vs. network effects in data
Cool stuff, no? In any event, it was a great few days and I’ve got to thank all of my fellow Foo campers for making this such an enriching experience. I know I always come out of these Silicon Valley, tech-heavy conclaves sounding like a stupid, fawning idiot, but I get so little exposure to these kinds of communities I am always amazed by the brainpower, the collaboration and culture of sharing and idea exchange. You take a New York boy, a Wall Streeter no less, stick him in Silicon Valley and if he loves technology, big ideas and deep thoughts his eyes will bug out. Like mine. Guilty as charged. Almost makes me want to work in the Valley or in Redmond. Almost.
Update:
