Join 3,501 readers in helping fund MetaFilter (Hide)


They made me learn Lisp
April 3, 2012 10:49 AM   Subscribe

What are the big stories in Artificial Intelligence research these days?

I was an AI-focused cognitive science major about 20 years ago. What are the big problems in AI today? who is working on them? What language are they using? How are they doing? What is the most important paper published in the last 20 years?

Bonus points for other interesting recent developments in the cognitive science world
posted by shothotbot to Computers & Internet (9 answers total) 8 users marked this as a favorite
 
Google's driverless cars come to mind.
posted by dfriedman at 11:04 AM on April 3, 2012


Regarding Google's driverless car, one of the lead researchers is teaching a free online course using Python as the implementation language. So Python is at least one answer to your question.
posted by mmascolino at 12:15 PM on April 3, 2012 [2 favorites]


Well natural language processing seems to be a solved problem now.
posted by Popular Ethics at 12:15 PM on April 3, 2012


I don't know much about its standing in the field, but I keep seeing references to the Singularity Institute, which has a focus on AI and is largely funded by the Thiel Foundation.

If you can believe Wikipedia the Association for the Advancement of Artificial Intelligence (AAAI) is prestigious in the field and maintains a list of current AI-related news here.
posted by Wretch729 at 1:08 PM on April 3, 2012


Regarding Google's driverless car, one of the lead researchers is teaching a free online course using Python as the implementation language. So Python is at least one answer to your question.

In this case Python was chosen for teaching purposes -- the real driverless car was done in C++.
posted by vorfeed at 1:09 PM on April 3, 2012


Watson is nice but NLP is far from solved. Sebastian mentioned this specifically in ai-class. Also, he was excited about unsupervised learning in general--specifically with respect to the self driving car.
posted by jewzilla at 4:53 PM on April 3, 2012 [1 favorite]


1) NLP is not solved, at several different levels. (This is good for me, as this is part of my field, and I like having a job.) See SIGDIAL or ACL for current work.
2) AAAI is pretty much the main journal/conference/symposia in the field of "general AI." I don't think anything else comes close in terms of sheer size. An offshoot symposium/journal is Advances in Cognitive Systems.
3) I dunno if there is one "big problem" to talk about. Over the past twenty years, there has been a pretty significant split between symbolic and statistical methods of inquiry. Some of the problems I'm aware of (I'm more of a symbolic person) are transfer learning, generalizable problem solving, perspective taking, beliefs and deception, and contextual interaction.
4) Cognitive architectures are used for some things, where you have to know Java or, weirdly, Lisp. Java and C++ are probably still the most common general languages. Python sounds good in theory, yet I can't think of anyone I know personally that uses it. (Or at least talks about using it).
5) I'd say CMU and ICT (at USC) have the most interesting AI research programs. Georgia Tech and MIT have pretty decent robotics. I should be able to think of more off of the top of my head, but I'm tired and getting ready to go to bed.

As an aside, the driverless cars barely rate a mention in most AI gatherings. In fact, I know program managers in AI that are actively pissed off at what they consider the public fawning over and glorification of the DARPA Grand Challenge, as most of the new research there is more engineering problems and less cognition problems. So...I'd be cautious about mentioning the driverless cars to an AI researcher unless they are part of one of the specific teams that is really into that stuff. I would certainly not call it one of the biggest problems in the field, even to one of those folks.
posted by wending my way at 12:25 AM on April 4, 2012 [1 favorite]


Oh, and after re-reading your question, there is some AI stuff in Cognitive Science, which is probably the main publication and conference there. I missed that you were an AI focused CogSci researcher and not a CogSci focused AI researcher.
posted by wending my way at 12:31 AM on April 4, 2012


Well natural language processing seems to be a solved problem now.
Having some, err, intimate knowledge of said solution, I can safely say, no, no it's not quite solved. :-) One of the more interesting things about various kinds of AI is (a) how various the solutions are (b) and how much very particularly focused grunt work there is.

It's pretty cool when you're done, but it's still like teaching your cat and your dog to poop in the toilet and flush. Most of the time you can't just point at the bowl. There is no magic algorithm, you have to put in the time forming a particular solution to a particular problem.
posted by smidgen at 11:15 AM on April 4, 2012 [1 favorite]


« Older What are strategies that I can...   |  MS Access Filter: Magic butto... Newer »
This thread is closed to new comments.