You may remember when Deep Blue defeated Chess grandmaster Garry Kasparov (though that win was due to a computer bug), or when Watson appeared on Jeopardy, and won.
However, did you catch the news a few weeks ago that a computer has mastered the game of Go?
Why is this a big deal? What is Artificial Intelligence? Can a machine actually learn? What impact does this have on your daily life?
Artificial Intelligence (AI) is a bit difficult to define, as even practitioners continue to debate if a machine can even think. What do I say? While I haven’t formally studied the field in a long time, I define it as giving a computer the ability to develop the ability to answer questions and solve problems that it doesn’t have the direct solution to. AI is not one single topic, it is as complex as human intelligence.
Another term you may often hear bandied about is machine learning. It’s the ability of the program (or computer) to add to its corpus of knowledge and apply that to new situations through supervised interactions. Perhaps a decade ago you used voice recognition software instead of typing documents. As you taught it to (hopefully) understand you as long as you enunciated clearly and didn’t have a cold, that was machine learning.
There’s a new field that’s emerged in recent years, deep learning. It should not be confused with machine learning. It’s different in that its learning is mostly unsupervised. This is the new process by which many of the voice recognition systems learn today — and the success rate has risen drastically.
So what does this all mean to Deep Blue, Watson, and AlphaGo?
When Deep Blue played chess, it took a brute force approach to each move. It was still faster than a human because the processor could analyze solution after solution until the right answer was found. With chess, there are a finite (yet very large) number of possible legal moves, about 10^50 (see more about this at Shannon number). While the brute force approach isn’t considered an elegant or efficient solution, with enough computing power it is possible.
Watson by contrast began to look at all the possible answers to a question, analyze them with other answers that either supported that possible choice or refuted it. That would then rank different potential answers and the highest ranking choice was the one it chose.
So what did AlphaGo do when it bested European Go champion Fan Hui?
Go is a complex game, it’s simple to learn (even I have), but it’s difficult to master (I haven’t). There are 10^170 legal positions and that mindboggling complexity (which is more than the estimated known number of observable atoms in the universe) excites computer scientists and makes a brute force approach impractical. AlphaGo created a neural network and performed deep learning on it. We’ll find out later this month how AlphaGo does against the Go World Champion Lee Sedol.
So why is this significant?
I’m going to simplify drastically, but it’s now possible for a (very sophisticated and specialized) computer to have the ability to adapt and play by any rules.
While I don’t think we should worry about HAL taking over; to me the biggest impact, is that AI is now part of the general conversation. Like the human mind (and body), AI is proving as difficult to explain. Should we be worried and doomsayers that the end of life as we know it is here? No. Nor should we embrace these technological advances without stopping to discuss their effects, both positive and negative on society.
