Among the classes I’m taking here at UCL is Artificial Intelligence and Neural Computing. It’s a split class, which means the first half, Neural Computing, was taught by one professor, and the second half, Artificial Intelligence, is taught by another professor.
While I’ve never encountered something like that in the States, it seems to be fairly common here.
Anyway, now we are well into the Artificial Intelligence half of the course. As part of it, today we started discussing Knowledge Engineering, and doing an example of Inheritance Hierarchy. I don’t really know what those things are yet.
But the example that we did today is relatively simple to understand:
flying-things
↑isa
bird
↑instance
Tweety
isa simply means that the item is contained in the item being pointed at. An instance is a particular example of the item being pointed at. So, what this means, in short, is that bird belongs to the group flying-things, and Tweety is a particular bird.
Sounds alright, right?
Except, what if:
We know an ostrich is a bird, but we also know an ostrich isn’t a flying-thing.
So what is a poor computer to do? It is left with a contradiction in what it has been told, as Tweety is both a ostrich, which, as illustrated, means it is a bird, but that means Tweety=bird=flying thing and Tweety=ostrich≠flying thing.
Unfortunately, these types of distinctions are not contained in the language itself, but rather in our interpretations of what we are told–no person would make a pen for an ostrich when asked to build a bird cage, but a robotic carpenter, given the same input but without the layers of “common” sense that filter our understanding, would be forced to do a probability calculation to guess what type of bird we are talking about.
Of course, we could just restrict the acceptible inputs to the computer to stuff that is unambiguous–i.e., refer to birds never as birds, but always as “flying birds” versus “walking birds.”
But if you have to do that, you haven’t really succeeding in making a usable artificially intelligence.
Fascinating.
Recent Comments