Tuesday, July 12, 2011

Homework and DARPA

So one of the homework questions that I had for my 'knowledge based systems' course got was about the major AI contests which kind of got me thinking a bit about why not smaller challenges. While there are some things that will only ever be answered by the large and expensive grand challenges, there is another aspect that is lost with them. It is important to note that participation in these projects like the DARPA grand challenge or DARPA urban challenge is a very exclusive thing and requires large interdisciplinary teams. While much good can come of this, I think there is also some harm in all the focus being placed on these projects.

First, a large portion of academia in unable to participate in these challenges. While my college has a graduate CS program, we lack ME and ECE programs, which would obviously be required for these projects. Additionally we are not a tier 1 university, but instead focused on professional development. However that does not mean that non of the students have anything to offer to hard problems, and by alienating so many smaller colleges and universities we loose a large portion of potential minds that could work on some of these problems.

Another problem that I see is that many of these research groups, because of the huge financial input into the problem, are likely to take risks on outrageous ideas. Sometimes these are critical for making breakthroughs, or even just seeing that another legitimate solution exists.

So why not mini challenges? What I see is that the Maker/DIY/Hacker community has incredibly sophisticated tools, everything from arduino, beagleboard, Digilent FPGA boards, Android devices, and the Kinect. As an R/C enthusiast I also know that we have quite sophisticated platforms. So it does not seem like there is a technical reason not to fund small teams of 2-5 individuals, probably university students, with a standard platform to perform some given tasks in a scaled down challenge.

In addition to the additional mind share working on the problem I think some other huge things will come out of something like this. First, we will see more people with sophisticated AI backgrounds graduating from ALL colleges and universities instead of just the elite ones. This is great for companies that might want a low cost/risk way to explore some of these ways of solving programming problems.

Additionally I think we will see one huge impact if standardized and smaller scaled hardware is used, that of massively filtered data. One of the great things that the human mind does is filter at various levels. Why can we not use the same ideas in AI to take real world problems and simplify them, but still come out with satisfactory results?

No comments: