Here we go again!
I hope the Cornell guys have not fallen into the trap of begging the question. There is an essay/book chapter by Douglas Hofstadter about an AI that the authors claimed had made analogies between electrons going round a nucleus and planets going round the sun.
Although it sounded impressive at first, it turned out that the program was given something like : "similar(electron, planet), similar(nucleus, sun), circling(x, y)" and then produced : "similar(circling(electron, nucleus), circling(planet, sun)".
Hofstadter points out that to the AI, this could easily be "similar(circling(cat, dog), circling(plant, table))" or any other combination. Without any meaning behind the words, you get out only what you put in.
On the other hand, the experimenting-robot is quite scary. Think about it : they now have the capability to make bio-weapons. Better not turn down the robots funding requests!