But AI is not robots and robots are not AI. In fact what distinguishes recent AI is it’s ability to “learn” which is why much of today’s AI is known as “machine learning”. There are many types of robots that have no ability to learn. There are also many machine learning applications that have no physical embodiment, humanoid or otherwise.
The Google-owned AI application, “DeepMind”, that defeated the human Go world champion, Lee Sedol, three games to one, had no physical capabilities. In fact DeepMind’s AlphaGo depended on AJA Huang, pictured here on the left, to be it’s hands. Dr. Huang placed stones on the Go board for AlphaGo after it “decided” it’s next move. The “mind” behind Dr. Huang was AlphaGo.
If DeepMind were embodied within Honda’s ASIMO humanoid robot it could have decided on it’s moves AND physically executed them as well. However, all of these conversations belittle the real opportunity for AI which is to enable human beings, not robots, to be smarter than we might otherwise be just like AlphaGo enabled Dr. Huang to appear to be a better Go player than he would have been without AlphaGo.
AlphaGo Zero has beaten it’s predecessor, AlphaGo, 100–0 after training for just a fraction of the time AlphaGo needed to beat it’s human competitor and it didn’t learn from observing humans playing against each other — unlike AlphaGo. Instead AlphaGo Zero’s neural network relies on an old technique in reinforcement learning: self-play.
“AlphaGo Zero “picked up Go from scratch”, without studying any human games at all. AlphaGo Zero took a mere three days to reach the point where it was pitted against an older version of itself and won 100 games to zero.”
Essentially, AlphaGo Zero trains by playing against itself. During training, “it virtually sits” on each side of the table: two instances of the same software face off against each other.
Perhaps the best example of how AI can make humans smarter is described in a recent paper by MIT researchers, in which they describe how neural networks, a type of machine learning, can be built much more economically than previously understood using the “Lottery Ticket Hypothesis”. ( https://openreview.net/pdf?id=rJl-b3RcF7 )
“A randomly-initialized, dense neural network contains a subnet-work that is initialized such that — when trained in isolation — it can match the test accuracy of the original network after training for at most the same number of iterations.”
The old AlphaGo relied on a computationally intensive Monte Carlo tree search to play through Go scenarios. The nodes and branches created a much larger tree than AlphaGo practically needed to play. AlphaGo Zero started from scratch with no experts guiding it and it is much more efficient. It only uses a single computer and four of Google’s custom TPU1 ( tensor processing unit (TPU) ) chips to using Google’s machine learning platform, Tensorflow , to play matches, compared to AlphaGo’s several machines and 48 TPUs.
Since Zero didn’t rely on human gameplay, and a smaller number of matches, its Monte Carlo tree search is smaller. Google claims it’s TPU has more than 10 times the capabilities of any existing chip of comparable power consumption. That’s three generations of Moore’s Law, or nearly what would have been seven years of advancement with conventional computing.
Google is already offering TensorFlow models which are data structures that contain the logic and knowledge of a machine learning network trained to solve a particular problem. “TensorFlow Lite” provides tools needed to convert and run TensorFlow models for Android and Apple mobile and Information of Things (IoT) devices.
This means machine learning applications can be developed for much less expensive machines and all kinds of problem solving applications. These developments are similar to ones that occurred when software applications became available for Nintendo game machines while also being used with massive mainframe applications like those used by NASA to plot a flight path to the Moon or process Census data for the entire United States.
There’s likely to be a market for thousands of AI applications. AI applications are rapidly going down the development learning curve. As they do they’ll make each of us individually, “smarter” than we ever thought we might be. Much smarter than Gort even with his intimidating lazer eye trying to terrorize humanity!