Most significant present-day AI developments?

Question

What do you consider the most significant progress / breakthroughs in real world applications of present-day AI research? (including, but not limited to: machine learning, statistical data processing, and other disciplines spinned off from AI).

Please spare / do not want: ramblings about AI winters / disappointment;

Do want: links, and pointers to concrete real-world applications.

Answer 1

I think the most significant breakthrough is that real world consumer applications actually utilize AI routinely today. It has become common, and is not just mere curiosity of academic research and special applications any more, like it was ten years ago. Some examples:

• Speech and text recognition (e.g. iPhone).
• Face recognition in digital cameras.
• Search engines.
• Email spam filtering.
• Automatic gearboxes of cars.
• Games.
• etc.

It's all around us! :-)

Answer 2

I think real/strong AI has lost it way, for decades the speaking/understanding computer was going to be available 'in the next 5 years'. Then we ended up with Dragon (no connection) which doesn't understand anything, it's a clever microphone, and it's a while since I've heard anything about AI - it's just not mainstream anymore, because it is too damn hard. One thing I think has been proven beyond doubt real AI, as in thinking machine, Turing Test passing AI - is still a (very) long way away. Don't get me wrong, there's tons of good research going on, but we'll have to wait 200-500 years for a result.

My gut feel is they'll be some interesting stuff coming out of massively parallel systems, especially ones built with really simple nodes. And if I had to point at a single AI breakthrough I'd be looking at spin offs from the nano-tech field, getting really small and seeing what cells in the brain are up to - science fiction it is, but we'll crack it one day.

Answer 3

I would add autonomous robots like those in the DARPA challenges to the list. Driving through a desert or rural area, recognizing the terrain, avoiding ostacles, finding paths and so on are definitely tough AI problems.

Answer 4

Actually, AI research is having a renaissance and has been for the past 5-8 years or so.

Back when neural networks were all the rage in the 70s and 80s, they were showing such promise in solving simple tasks that people's hopes were sky-high for the whole field of AI. Then, when it turned out to be very difficult to move on from the very simple tasks to real-world problems like language acquisition, a lot of people became disillusioned. Until recently, that is.

I am not the best person to ask -- being no AI expert -- but I believe some of the most promising areas are:

1. Semantic search and data mining (including text classification)
2. Statistical machine translation
3. 'Real intelligence' HTMs (read Jeff Hawkins' On Intelligence)
4. Relevance / Recommendation engines (essentially a hybrid of data mining and network analysis)
5. Visual object recognition