A neat bot would be one which automatically buys and then resells things with a markup. It would figure out the usual price an item goes for on a website like Ebay, and purchase it if the price is set far enough below.
The items are worth whatever people are willing to purchase them for in WoW and IRL. All you have to do is keep track of how various items are selling. You could implement a browser plugin to capture your hover and give you a floating window with stats from your database. About the point where I had this realization, I stopped caring about WoW auctioning and decided I would rather spend that time making real money.
A classic case of arbitrage. But it seems to me that it would be quite hard to achieve on ebay. You need to have very little overhead, mailing stuff isn't usually that cheap.
Drop shipping --send things directly from seller to buyer-- is allowed on ebay but has many restrictions.
It would be very interesting to see if you could get some heuristics that choose products that are at least moderately useful instead of screen protectors and (exploding) batteries for various mobile phones.
I have always wanted to try something like this with stocks or commodities. Make a bot watch the market, and buy and sell. I even got far enough to imagine some sort of genetic filtering, aka give the master bot 10k, then it distributes it to 10 bots, all which are slightly different, they spend it, the master takes any profit above 1k and invests it in another bot that is a combination of the best 2. Etc. That way your max loss would be 10k, and you might make some money.
Yeah, this is interesting, but there are some annoying realities:
1) If you make something like 3 intra-day trades (buy and sell on the same day) in some defined period (a week?) the SEC will classify you as a "pattern day trader". To be allowed to do that, you need an account with something like $25k USD min.
2) There are a few items that people often forget in simplified models that will completely eat any theoretical profits: brokerage fees, taxes, and currency-exchange effects (e.g. if you're a Canadian trading US stocks).
3) After all that, you're competing with the big guys who have a deep knowledge of how trading works, and fast access to more information than you do. It's an insider's game.
I should add that (market-price) trades aren't completed in an all-or-nothing fashion, unless you specify this (which brings another set of costs and risks). E.g. if you want to buy 1000 shares, it might happen in several chunks (say 200, 200, 600), with the price going up each time.
So, the process of buying or selling moves the price, which is another reason why trading with real money against competitors is different from training against historical data.
I have no experience, and little intrest in stock trading. I was however interested in the AI and automation aspects of this challenge, with the stock market being a good source of data with a definite but "unknown" solution. However due to the obstructions you brought to my attention, it seems although the stock market would not be an ideal application for this.