Artificial Intelligence
Posted on September 22, 2008
Filed Under Business and Investing, Cool Technology, Insurance, Programming, Risk Management |
I thought about artificial intelligence today while the pathfinding algorithm in my brain directed the car. This was on my drive home from the office, naturally.
Here is a short list of interesting AI and AI-like applications I have worked on, or just come across:
1. A Behavioral application:
Take a look at these behaviors first modeled in Java by Craig Reynolds. The behavior algorithms have since been converted to a C++ library titled OpenSteer, with applications in disaster/evacuation simulations, traffic patterns, games, and probably much more. I generalize this type of behavioral AI into the term pathfinding… steering is also a good word. Pathfinding/steering has great value for simulating locomotive behaviors, and it is sometimes used to good effect in computer games where other AI elements can be applied to the same actor.
2. A business/market/riskmanagement application:
Greg and I developed a program for an insurance agency that decided which insurance markets a “risk” (this is insurance for “client”) should be sent out to. Because the task required a lot of knowledge about the requirements and writing history of many different insurance companies, we found that the software could make a decision much faster than the employees; in fact, it could process the entire agency’s renewal list before an employee could evaluate a single renewal. Some of the factors that influenced the decision of which companies to send out to were construction type of any buildings, distance to water, year of construction, type of business, year incorporated, etc. These were compared to company criteria and writing history, and an output was given to the human operators in a ready-to-submit format. Insurance rating and submission software essentially investigates the client, looks for sinkholes in the available databases, looks up census data and property appraiser’s info for their properties, identifies the property within 1000 feet or so on a map (using the census data, generally), identifies the closest “navigable water” and measures the distance, and checks the rating tables against all this information to tell us what to charge, or whether or not we can insure this person. Most people in business just call this “automation,” but it is a complex decision tree than we would normally assume is handled by humans. If you want to spin it as AI, you could comfortably call it deduction.
3. Another business/market/riskmanagement application:
Similar to the insurance need, I suppose, is the need for stock market investors/traders/pros to predict and model complex human buy-sell behaviors. Greg probably knows more about this than I do, with his background. Most of the work that I know of in this arena can be categorized as either prediction or trend following, where the predictions are used to help traders, advisers, investors make decisions, and the trend following is used to simulate trading activity as if people were making the trades. This is an area that I would be interested in knowing more about, so if anyone has a reference let me know.
This is a “note to self” more than anything, but I may add some more first hand examples as I come across them (or remember them).
I went to my doctor when I was a kid and he gave me a pill for artificial intelligence. At least that’s what my mom said.