I got to listen to the Radio Lab program on my way home from a meeting last night on my local NPR station, WBUR. The final segment discussed the limits of human knowledge in science, and in particular the possibility that humans might reach limits of insight especially when scientific discovery is computer-aided.

As reported, a computer program called Eureqa was able to independently discover Newton’s law F=ma just by watching a double pendulum for a day. With input from a completely different field, it then discovered some rules by which simple cellular mechanisms work. These rules seem to be correct, because they accurately predict what will happen in the cell. However, as scientists look at those rules, they are at loss to understand them. They have no insight as to why the rules are correct.

You can listen to the segment and see the comments here.

At 8:05 into the segment, the hosts say that the scientists are “in this awkward position where they’ve got the answer, but they don’t have [pause] the insight.” And in that pause, all I could think was they would finish the sentence with “the question” … as in “they’ve got the answer, but they don’t have the question”. And why would I think that? Because of Deep Thought, Hitchhiker’s Guide to the Galaxy, and the answer to the meaning of life, the universe and everything. Douglas Adams fans (like me) know what I am talking about, and others can read about it here. I was gratified to see the comments on the Radio Lab website went exactly to the same thoughts.

1 comment:

Dan Greenberg said...

The same problem -- lack of insight -- has been a criticism of computational neural systems ("neural nets") for decades. You can train a neuromorphic system to do/predict all sorts of things correctly... but you can't then look at how it self-assembled and determine why it's so accurate. The same is true of the human brain (so far), a fact which offers the solace that at least the models were good!

The contrast, of course, is AI. There, it's all insight and lousy results.