What is Artificial Intelligence? by Mike Loukides & Ben Lorica

Author:Mike Loukides & Ben Lorica
Language: eng
Format: epub
Publisher: O'Reilly Media, Inc.
Published: 2016-06-23T04:00:00+00:00

Why the Surge of Interest?

Why is AI currently such a hot topic, after having being in disrepute for a few decades of “AI winter”? Of course, AI was in the news briefly after Deep Blue, and again after Watson; but these fads didn’t last. It’s tempting to see the current rise of AI as another fad. That would ignore the changes of the past decade.

The rise of AI has depended on tremendous advances in computer hardware. It’s tedious to recite the huge advances in performance and storage technology in the 30+ years since the start of the AI winter (which Wikipedia traces to 1984). But that’s an unavoidable part of the story, particularly if you’ve seen the racks of machines that made up IBM’s Watson. AlphaGo reportedly ran on 1,920 CPUs and 280 GPUs; the machine that beat Lee Sedol may have been even larger, and used custom hardware Google has developed for building neural networks. Even if AI algorithms are too slow to be productive on a typical laptop, it’s easy and relatively inexpensive to allocate some serious computing horsepower on cloud platforms like AWS, GCE, and Azure. And machine learning was enabled, in part, by the ability to store vast amounts of data. In 1985, gigabytes were rare, and weighed hundreds of pounds; now gigabytes are commonplace, inexpensive, and tiny.

In addition to the ability to store and process data, we now have the ability to generate data. In the 80s, most photography was analog. Now it’s all digital, and a lot of it is stored online, in services like Flickr, Google Photos, Apple Photos, Facebook, and more. Many online photos are already tagged with some descriptive text, making them a great dataset for training AI systems. Many of our conversations are also online, through Facebook, Twitter, and many chat services. As are our shopping histories. So we (or more precisely, Google, Apple, Yahoo, Facebook, Amazon, and others) have the data needed to train AI systems.

We’ve also made significant advances in algorithms. Neural networks aren’t particularly new, but “deep learning” stacks up a series of networks, with feedback so the network automatically trains itself. Deep learning thus tries to solve one of the hardest human problems in machine learning: learning optimal representations and features from data. Processing a lot of data is easy, but feature learning is more of an art than a science. Deep learning automates some of that art.

Not only have we made progress in algorithms, the algorithms are implemented in widely available libraries, such as Caffe, TensorFlow, Theano, Scikit-Learn, MXNet, CNTK, and others. AI isn’t limited to CS researchers in academic settings; increasingly, anyone can take part, as Pete Warden has shown. You don’t need to know how to implement a complex algorithm and make it run reasonably well on your hardware. You just need to know how to install a library and tag training data. Just as the PC revolution itself took place when computers moved out of machine rooms and became accessible to the general public, the same process of democratization is producing a revolution in AI.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.