Oh I know – this is the stuff that I was introduced to at hampshire college: purely statistical neural network AI.
It’s simple weighting of probabilities. Purely statistical. It’s actually quite simple stuff at its core, but it’s very powerful as the weightings get shifted around based upon new inputs.
What’s awesome about it is: Once you set it up _properly_ (and that’s the key – it has to be set up properly – database design is crucial to get the right answers) – you just keep pumping good information into it and the weightings continue to improve.
I’ve utilized it in Excel on numerous occasions when working for Schering-Plough, – my favorite being calculating probabilities of increasing sales by ‘teaching it’ to recognize common patterns of individual sales force behavior based upon the amount of drops and successful conversions of doctors in their prescriptions from allegra or zyrtec to claritin (and later to clarinex)
Once it knows how to recognize the patterns, it can produce advice – making predictions that are actionable.
Good stuff. Very practical.
What I did was very very simple algorithms though – nothing all that complicated as the stuff coming out of Stanford and whatnot.
My more interesting case was having to teach it to recognize “mood” via natural langauge comments. I couldn’t just do word search: I had to figure out how to parse whatever text people were saying and figure out percentages of happy-with, disappointed-with… and three other metrics that I don’t remember (that was 15 years ago), based upon free-text comments collected nationwide.
It’s not difficult stuff really, although what I was working on wasn’t that difficult. That was less AI and more having a glossary of possibilities really. It wasn’t automated: I’d set my program onto the data and run through it once.
Then I’d check its findings with my own decisions.
Then I’d adjust the metrics and run it again, taking a sample and seeing if the moods it calculated corresponded to what I thought was reality.
After about 4 or 5 run throughs, it was good enough and I was able to send the sheets and code to process single sentence “mood” comments for other parts of the company to use in for other surveys. Pretty neat. Got a $100 Amex gift card for it. tongue emoticon
[They called me a “SYSTEMS ANALYST II”. I still don’t know what that meant, really. I was just a temp who got bored with copying and pasting and started learning how to automate processes. Ended up writing a complicated thing that took over the whole project. As I was starting to go back to college to finish up *some* sort of a degree, I cut my hours down to 4, let my automated things do their thing… I didn’t care about the money… and was going to school for elementary education. Anyway, they had to hire me. Didn’t want to do it but whatever. Life’s short. I figured out the max that was acceptable for the title they were giving, went through the hiring process game and got what I asked for.
Fun stuff. I always like a challenge smile emoticon ]