still looking for problems to fit some solutions. my solution set currently involves fuzzy logic, hidden markov models, and natural language processing. neural nets are purposefully excluded from the solution set, because i dont want it to become my AI silver bullet.
fuzzy logic 1st. i can see how it is well fit to control systems and rule based systems. but being able to write logic in english-like terms just is not that big of a deal to me. the operators of my language of choice are quite adequate. maybe some 'else if' or 'switch' statements to really spice things up.
hidden markov 2nd. typically used by speech recognition for recognizing phonemes and words. the other standard is to train it on past works and then have it recreate new material based on the statistics of the past work. that is novel, but not very useful. for kicks i applied it to the text to speech program. all it did was randomly pick a starting phoneme, and then chose the following phonemes statistically to form words and sentences of gibberish. something like that might be useful for games like the SIMS, where the characters make speech like sounds.
natural language processing 3rd. i dont know enough to actually code this yet. the first step is to break up a sentence into the parts of speech. this would be difficult for me because i suck at grammar. you should know that if you've read this far ... second step is to determine what the sentence actually means. thats when things get interesting. still trying to formulate ideas from that point on. need to do more research to come up with something meaningful. regardless, this would probably take a significant amount of time to put something together.
so i'm slowly building an AI toolkit and figuring out what technique fits which class of problems. now my new problem is coming up with an application to code applying that technique in a way that solves some meaningful problem. i.e. pairing up AI to application coding is harder than i expected.