Some weeks back, I was talking with my manager about AI and how it is such a bogus field. My manager replied that in a few years we will see applications that use AI in our daily life. However, I was quite skeptical - and I refused to agree to this. He then gave an overview of neural net and how they can learn to solve the problems. Here I pointed out that Bayesian filters can also be considered a form of AI, as they can learn from their previous data and they can make decisions, but Bayesian filtering is mathematics and not AI. At this he replied that most of AI is mathematics and only some part of it is hocus-pocus and hand waving.
This brings me what I have been thinking for a long time.Joel write
A very senior Microsoft developer who moved to Google told me that Google works and thinks at a higher level of abstraction than Microsoft. "Google uses Bayesian filtering the way Microsoft uses the if statement," he said.. I had always suspected this and had also felt that this was the way to go. A few months back, we had a presentation by a researcher (not a Yahoo! employee), who was working on extraction and summarization of documents. He had a formula that he was applying on the sentences of the documents to find the weight of the whole sentence and then finally if the weight of the sentence was above some limit, it showed up in the summary. I was skeptical about this approach - my belief is that the Bayesian approach can be used to classify documents. Luckily, there is a project that seems to provide a framework on which things can be built further.
Defined tags for this entry: geek stuff, programming
Comments
You can reduce any piece of AI down to a mathematical function however, for some systems (Dynamic Neural Networks, for example) the function shape is so complex that it ceases to be useful.
If you're talking about strong AI, where machines learn to think and act like humans, then this is a very difficult subject to argue. Most arguments boil down to the fact that a human is a poor judge of what is alive and what is not. Furthermore, classifying intelligence is more than just a dictionary entry, it's actually a thorny problem that hinges around subjectivity of the debators.
Finally, if you're going to discuss a topic, it's not wise to expand the topic to make your argument heard. By saying that Bayesian Filters is part of AI is redefining the problem, not responding to the statement of your manager. Some people still think that Expert Systems (if-then-else rules) are AI but they are clearly not!