I'm going to hand in a proposal about what I want to write my master thesis about in a month. The problem is that I don't really know what I want write about. So I'm looking for interesting suggestions about what to do. I have studied CS/IT for almost four years and I have studied AI for about half a year (fulltime) and have studied things like GOFAI, neural networks, evolutionary algorithms, robotics and so on. I also prefer programming in C/C++/C#/Java.

(not being sarcastic or rude here, honest:)

How about a forum answering robot, that attempts to provide the best answer when someone asks a common question?

The bot would need some good NLP skills, and would then probably try to answer by using a variety of techniques, including looking in the forum history to see how the question has been answered before, looking in similar forums, and doing a basic web search, in an AskJeves style.

You'd be amazed at the number of Java questions that show up on my favorite java forum that 1) have been answered, word-for-word, at least 50 times before, 2) have been answered on every single other java forum at least 200 times before, 3) have been answered on java tutorial sites across the web.


... but of course you may not be into NLP. That said, technologies like NLP still have so far to go, that you can do really interesting research on them. My master's thesis was on a "new" technique that can be used for GAs, i.e. a really narrow problem set, the solution of which only differs a tiny bit form dozens of other techniques. Likewise, any work on ANNs will probably be just as narrow and uninteresting.

When i started my PhD i was initially looking at studying the evolution of language using AI - over a few months this kind of changed due to its practical application and the fact that I wasn't a biologist and my studies were leading me too deep into a biological route.

Anyway, back in 2002 there was a project running called the talking heads project (http://talking-heads.csl.sony.fr/InfoIndex.html) that consisted of robots that looked at objects on a board and then began to play naming/guessing games which ultimately led to the formation of a pseudo language. A lot of interesting discoveries were made with ragards to language, such as if the meaning of the same word could hardly ever be confused due to its context, then it will keep that label until confusion does occur. For example, in the english language the word 'bank' can refer to both a financial institution as well as a river bank. So it did produce some interesting results which could help explain fundamental trends in language.

Anyway, not soon after 2002, Sony pulled the plug on the project (although it is still explored by the head of research Luc Steels). I assume it wasn't making them a lot of money.

So with regards to a potential project. It would be cool to start to reimplement this project, so people can interact with the talking heads, add human input, by playing the guessing game with the agents to see how the language emergence works. They did have this setup where anyone online could login and play the naming game with the agents to change the lexicon and eventually a whole new language evolved. All the papers and research are openly available and Luc Steels also wrote a book which provides a good all round summary of how it worked. I don't know how far you would get in a year, but i'm sure you could get a fair way through (i.e. create the agent in its most basic form, or get a couple of agents playing the naming and guessing games). Tutors always seem to like 'ongoing' projects. By the end it will no doubt provide you with a good understanding of some common AI techniques, plus some more advanced ones as well and there will be plenty of coding to do.

P.S. They used real robots in the experiments, but you could still recreate the experiment as a computer simulation (they tended to use their simulation as proof as concept, although Steels did have his reasons for using robots in the final implementation).

Quote: Original post by Nocklas
I have studied CS/IT for almost four years and I have studied AI for about half a year (fulltime) and have studied things like GOFAI, neural networks, evolutionary algorithms, robotics and so on.

Could you list specific things you learned during all of your studies that really piqued your interest in them? What was something that when you first heard about it you thought "WOW"?

It's far more likely that you will complete your project and do well in it if it's on a topic that interests you. It doesn't have to be earth-shattering research at Master's level... merely something that shows your competency at applying research skills and communicating a research result to other people.

Cheers,

Timkin

Do machine vision research, but use an artificially generated 3D world.

Like a artificial pilot/flight sim that only uses the "pilots eyes" to fly by.