Without reading the report they refer to, I'll declare in advance ...
What a pile of steaming doo-doo. Give the bees a hard instance of the TSP and watch them flail. Give a computer an instance that the bees "solved" and watch it solve it in less time than it takes a signal to travel from one neuron to another.
Brains are amazing things, bees are amazing critters, but hard instances of NPC problems can't be solved exactly by using these sorts of short-cuts. Soap film gives good solutions, but not necessarily optimal. Simulated annealing gives good solutions, but not necessarily optimal. Randomized shotgun hilll-descending gives good solutions, but not necessarily optimal.
Neural nets, either wetware or software, give good solutions, but not necessarily optimal.
As a Biology major, and having taking several classes with a nationally recognized neuroscientist with concentration in honeybee brains, articles like this are particularly interesting to me, and I hope to see this study when it comes out.
It is worth noting how rarely news outlets will report correctly on the findings of research - they jump on whatever might make headlines, sometimes even completely contradicting the article in question.
That being said, I think it also inappropriate to say things like What a pile of steaming doo-doo about a study that you haven't read yet. While it is human nature to be skeptical, and yes it might not meet all of the the implications that this news brief summary of an abstract makes, it is important to recognize that there so many things that nature does way better than anything man even comes close to, and that studying the efficiencies of these systems makes for better design and undestanding.
I'm sure the article will be fascinating. I'm also sure the bees don't solve the TSP as computer scientists and mathematicians define it.
As I say elsewhere, hard instances of the TSP are non-trivial to find, especially in 2D. Random instances are almost always trivial. Further, "good" solutions are trivial to find, even when the exact problem is hard.
I'll bet the bees never solved a hard instance (in the technical, algorithmic complexity sense of "hard"). The hype is going to be way overblown.
It all depends on the definition of "solved". I doubt whether the same standards of "solved" were applied to computers as well as to the bees. If one is willing to relax the standards, TSP isnt that hard to solve on computers either. One such old solution is to simulate the dynamics of an elastic band that is attracted to the cities. http://www.google.com/search?q=elastic+net+tsp
It is not guranteed to reach the absolute minimum cost, and no point of the band may pass through a city location right away. But more often than not the band stabilizes at a decent solution. Would this be considered a solution of TSP in a computer science sense ? No, it is not even an approximation algorithm. But for practical purposes it may be good enough, and I think thats what the bees care about.
Statistical physics has answers to such problems if the standards are weakened to "solves it to within an _acceptable_ tolerance _most_ of the time". However, both the weasel words, "acceptable" and "most" are necessary. These tools may be applied to other biologically motivated solutions as well.
The citation is: Lihoreau, M., L. Chittka and N. E. Raine (2010). Travel optimization by foraging bumblebees through readjustments of traplines after discovery of new feeding locations. American Naturalist: doi:10.1086/657042.
I've been foraging around trying to get my hands on the report, but to no avail. Maybe someone with better Google-fu can help?
The one that you're interested in is listed as "in press" without a PDF yet. He'll probably link it up soon.
You could also try emailing him now at the address on that site. I've had great success email researchers directly and asking for (p)reprints of specific papers.
These "early" studies kind of drive me insane. "Research uncovers that carbon-based life forms more capable than computers when it comes to building protein".
I disregard (except to post a snarky comment apparently) until the downstream studies do something to unlock whatever potential they've uncovered.
Could these bees be solving an NP problem in P time? Could they test that by increasing the number of "cities" and seeing if it takes the bees exponentially longer?
What a pile of steaming doo-doo. Give the bees a hard instance of the TSP and watch them flail. Give a computer an instance that the bees "solved" and watch it solve it in less time than it takes a signal to travel from one neuron to another.
Brains are amazing things, bees are amazing critters, but hard instances of NPC problems can't be solved exactly by using these sorts of short-cuts. Soap film gives good solutions, but not necessarily optimal. Simulated annealing gives good solutions, but not necessarily optimal. Randomized shotgun hilll-descending gives good solutions, but not necessarily optimal.
Neural nets, either wetware or software, give good solutions, but not necessarily optimal.