Computers are much better than human beings in many areas. Computers excel in efforts that require delicate precision like surgery. Computers are better at making complex calculations and faster at order execution, such as running a stock market. Google’s DeepMind artificial intelligence software has proven computers can be better at thinking.
Chess, Jeopardy, and the Game of Go
IBM started to prove this when their Deep Blue computer beat the chess world champion, Garry Kasparov, in 1997. Then, in 2011, IBM’s Watson computer beat the best human players for the $1 million prize in the game of Jeopardy. Now, Google showed that computers can play even more complex games like Go.
Google’s DeepMind used an artificial intelligence program that they developed called AlphaGo to take on the world’s best human players. The New York Times reported that in March 2016, AlphaGo beat the Go Master from South Korea, Lee Se-dol. In March 2017, that AlphaGo beat the 19-year-old Chinese whiz-kid, Ke Jie.
Beating a human player at the game of Go is not possible by calculating all the combination of moves and then selecting the best one. There are simply too many combinations of moves to calculate all of them, even with lightening-fast computer speeds. This means in order to play Go well enough to beat a human player, the computer program had to think and then make its best guess for the next move.
AlphaGo learns like a human being by playing Go. It plays all the simulations of games of the game that were recorded as played by human masters. It also tries to beat itself over a number of times. Over the last year, it has been learning very fast.
DeepMind Used for Breast Cancer Detection
Games are fine for a proof of concept; however, what about something more serious? Forbes reports that DeepMind’s artificial intelligence can read mammograms more accurately than human doctors. So far, there are over 40,000 mammograms taken of real women in the dataset used to teach the artificial intelligence program to recognize breast cancer.
Learning How to See Better than People
DeepMind is working on artificial intelligence programming that teaches robots to “see.” To practice and learn, it can take any photo and create a 3-D model of it. It can show the 3D model from different points of view.
The Google scientists are trying to teach the program to make good guesses about the real world and to have some imagination about what is “seen” in order to determine perspective, lighting, and dimensions. This learning happens without any human engineering. The artificial intelligence software gets better by itself. Next, it tries to write music.
The DeepMind software is getting better, and not yet perfect. There is still the problem of misidentification, such as when the software thinks a dog is a wolf. That is OK for a failed test; however, no one wants a real-world armed security system to make a bad decision that is based on faulty artificial intelligence to shoot the wrong person. With AI, we still may have a ways to go.