Day Two at AAAS Conference…
I’m taking Sebastian Thrun’s Programming a Robotic Car class starting this week, so I was particularly excited for Peter Stone’s talk “Intersections of the Future: Using Fully Autonomous Vehicles” and boy oh boy did it not disappoint. Behold the future:
What I found particularly interesting about his system is that it depends on the human passanger not overriding the autodrive. Most proposed regulation of autonomous cars assumes human intervention is the failsafe instead of instant death, as it would be here.
It’s looking more and more likely I’ll never need to learn to drive!
There were some concerns raised in this talk about whether the cheapness of computing power lets people buy their way out of problems that it used to take serious thinking to solve. I don’t think I buy into this fear. Knowing how to brute force a solution takes a lot of serious abstract thinking.
The more interesting facet of the talk was the possible mismatch of skills a scientist needs to have to work on big science today. She needs to be a dedicated creative researcher but also needs to be an administrator and grant writer. A problem best summarized (in a different field) by Sondheim, natch.
This was all contrasted with the old patron system (or the gentleman of leisure as his own patron variant), but I wonder if that kind of structure might be on its way back. Google X, the Singularity Institute and other groups seem to recruit brilliant people for passionate work fueled by eccentric, geeky billionaires.
Web Surveillance: Fighting Terrorism and Infectious Diseases
(moderated by Vint Cerf, squee!)
Ok, I’ll admit I enjoyed a lot of this panel in a fairly technical way, so I’ll just share three fun methodology notes:
- One of the presenters was using news articles to predict epidemics faster than the clinical data came in (similar in spirit to Google Flu Trends). To put it very simply, their algorithm parses news articles from all over the world and plots the frequency of flu mentions. Once the frequency passes a critical value, it sends an alert to epidemiologists that an epidemic is in progress. Here was the surprise: the trigger value isn’t a constant, it increases over time, recognizing, I guess, that news coverage can be a feedback look, amplifying the flu news even if cases aren’t increasing. To trigger an alert, the news has to essentially outrace its own normal frenzy.
- One paper depended on content analysis of the kinds of links Google and other search engines produced when they were fed certain keywords. I wonder how salient this kind of study will be as, more and more, there’s no canonical list of search results. Google and others are trying to specialize and personalize. In their ideal world, no one sees a list of results that’s not shaped by your previous behavior. Maybe researchers will go on Mechanical Turk and try and get a sampling of personalized results.
- One study of how ideas move across social networks explained they were estimating “the hazard of becoming interested using survival analysis” and I found I became dangerously enrapt.
This one is getting its own post when I get home. I had some strong opinions.