Despite the quality of the article on Dr Guestrin, he's a legit computer scientist who got in the press. That's a "good thing", right? There is a large portion of the population who only uses computers for emails and web, and this article at least opens them up to the _basic_ idea that computers and computer scientists can help people.
As far as I can tell, the actual paper that this story is about is http://www.cs.cmu.edu/~guestrin/Publications/JMLR08SensorPla... - it was published months ago, and was actually submitted in 2006, meaning they probably thought of this and did the research 2-3 years ago. Given that, it's almost hard to call this 'news', although it is interesting.
Came out today, so today would be a good day for a local paper to report it. The Post-Gazette can't vet the importance of individual CMU research projects, but they trust PopSci to.
I don't get what the difference is between "machine learning" and plain-old constraint solving. From looking at the abstract of his paper, he's solving an NP-Complete problem using a new approximation algorithm. So, the real contributions of his work are 1) mapping his particular sensor network problem to the NP-Complete complexity class, and 2) an improved approximation algorithm. The newspaper article seems astounded that multiple (real-world) problems can be mapped to the same algorithm.
So it's efficient then? As opposed to every other inefficient algorithm?
I bet this is what medical researchers feel like every time they open a newspaper.