Recent Achievements in AI Research (Sparrho AI Feature: Part 1 of 2)
I’m bringing you a special two-part Sparrho Blog Series about Artificial Intelligence (AI) this coming fortnight. The first blog will be about the milestones that AI research has recently achieved, while the second will cover future trends in AI research.
In 1996, IBM challenged Garry Kasparov, a world chess champion, to a game of chess with their AI “Deep Blue”. Kasparov won the first match but lost in a rematch a few years later. For comparison, the computing power of the supercomputer Deep Blue was 11.38 GFLOPS (Giga Floating Point Operations Per Second) while a modern Galaxy S5 smartphone has 142 GFLOPs, more than 11 times more powerful [ 1 ]. Since this seminal victory for AI in the late 20th century, technology has moved on a great deal since then.
In 2011, the IBM Computer named Watson beat won Jeopardy! against two former winners [ 2 ]. Built by the same division of IBM that created Deep Blue, Watson had over four terabytes of disk storage, which gave it 200 million pages of information, including the entirety of Wikipedia. Recently, IBM has invested $1 billion into the Watson project, employing over 2,000 people, but more about that in our next blog [ 3 ].
Google gained publicity in January this year by announcing that they had created “AlphaGo”, a computer that was able to beat a professional Go player for the first time [ 4 5 ]. Although not the world’s best, Fan Hui was the European champion and unfortunately now, the first professional Go player to lose to AI, losing 5–0. Go was only recently won due to the complexity of the game, for example the typical chess move has 30 legal plays while Go has around 200. The number of possibilities in Go is roughly 1, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, 000, or 10¹⁷¹. So the AI needed to employ self-learning and deep neural networks from its cloud network [ 6 ]. However, this AI was not designed purely for the game of Go — the computing AI behind Alpha Go is able to be teach itself by simply watching game, such as Mario or Pacman, being played, and then can go onto master it [ 7 ].
So these are the recent major milestones in AI research, an extremely fast and exciting industry growing almost exponentially every year. Next week we’ll discuss both the what we can do with AI (such as employing Watson in healthcare) along with future trends in research.
Originally published at blog.sparrho.com.
