The Artificial-Intelligence Boom - Entrepreneur Generations

Over the past week, as many South Koreans have searched online for information about artificial intelligence as have about food. It is not that South Korea is strangely bored by food, either—there are usually 80 searches about food for every search about AI—but rather that the victories of the computer program AlphaGo over Lee Sedol, who hails from the country, have captivated public attention.

The popular interest builds upon soaring academic interest and corporate investment in artificial intelligence and a suite of related fields, like machine learning, neural networks, natural language processing, and computer vision.

AlphaGo notwithstanding, the rising interest is driven by the research, not the media. One way to see that is just by looking at attendance of research conferences in these fields. Here, for instance, are several growth measures of the Conference on Neural Information Processing Systems (NIPS), the top machine-learning conference:

The graph comes via Sebastien Bubeck. Yet Bubeck says in another post: "I think it’s fair to say that we do not understand why deep nets [one of the new methodologies in the field] work at all, and we understand even less why they are doing so much better than anything else." (Full disclosure: Bubeck was my first statistics professor.)

He is not the only one to think that. Andrej Karpathy calls a tool in the field called recurrent neural networks (RNNs) "unreasonably effective" and writes: "There's something magical about [RNNs]...I'm training RNNs all the time and I've witnessed their power and robustness many times, and yet their magical outputs still find ways of amusing me."

And here, from Tomasz Malisiewicz, is the growth in attendance at a major computer-vision conference:

This level of attention makes researchers uneasy. What they have in the back of their minds, I think, is the "AI winter," or bust, that followed in the wake of the field's last boom:

I've spent much of the last year studying machine learning, coming from my own background in econometrics. You can expect, in fact, that this blog will shift a bit in that direction in the coming months.

My own assessment from the front lines: Even though recent advances make it easy to justify this level of attention—see here for a nice summary of the progress that has been made, and here and here for two examples of the state of the art—it is frankly hard not to worry that the field has once again set itself up for another cycle of broken hearts and missed expectations, particularly if the empirical gains from the latest methods run out. 

from Evan Soltas http://ift.tt/1QWAk2K The Artificial-Intelligence Boom - Entrepreneur Generations

Posted by Entrepreneur Generations

Tweet