YMMV on this, but I studied CS with an informal concentration on AI/natural languages. Here's some take-them-or-leave-them suggestions.
If you want to maximize the return on your time for this class, do a project which:
1) Uses one or many data sources which are publicly available but which, ideally, are not quite as simple to access as straight downloading a CSV file. A bit of practical experience with scraping, API use, or data processing doesn't hurt. Bonus points if you get a taste for working with large data sets.
2) You will not make an AI which learns to play chess in 11 weeks, or in 11 years. Just to set expectations. A more reasonable task for the same timeframe given your current skillset is e.g. "Given a large corpus of documents and a small number of them are hand-tagged, explore a few different approaches for classifying the remainder of the documents." A motivated undergrad can succeed at implementing a Bayesian classifier, but you will not advance the state of the art on chess.
3) A lot of academic projects focus on toy problems, like e.g. chess or a contrived simplification of a real system. There is no reason that you have to adopt this academic convention: consider picking a real system with consequences. There exist many websites which have information on them that actually impact decisions which people care about -- wouldn't you rather learn to do analysis on that rather than pulling out arbitrary trivia out of e.g. the British national corpus (which, I rush to mention, is an excellent tool).
4) Think about the presentation layer for findings in more detail that the typical academic paper, which spits out a sentence or two of summary stats and maybe graphs them. This might be an opportunity to have a bit of fun doing, e.g., a website which lets you search through your (voluminous) findings.
Putting it all together, you could imagine something like "I have developed a website and/or Chrome plugin which, when pointed at an Etsy item, predicts the likelihood that it will sell. Or it predicts the likelihood that a KickStarter campaign will succeed. Or it predicts the final sale value of an eBay auction -- better in some categories than others, see page 6. Or it successfully paints a red/blue map of the United States using no prior knowledge other than a geolocation database and the Twitter stream. Or it asks you ten questions about seemingly irrelevant trivia and then makes a surprisingly accurate prediction on how long it has been since you ate sushi."