How do you define intelligence? You might consider it to be the ability to:
- Learn, understand or to deal with new or trying situations (Merriam Webster)
- Acquire and apply knowledge and skills (Oxford)
- Understand and learn well, and to form judgments and opinions based on reason (Cambridge)
Since the invention of the computer, people have strived to create powerful machines that can imitate human intelligence. Computers generally do what we tell them to do, but a machine that can make its own decisions is incredibly powerful. One way to approach building such machines is to “teach” them using virtual games (like chess) since they have defined rules and strategy, which the computer can understand and evaluate.
The history of Artificial Intelligence and game theory starts with IBM’s Deep Blue, a chess-playing machine developed in 1997. It was incredibly powerful and could process 200 million moves a second. Deep Blue gained notoriety by beating Gary Kasparov, one of the greatest human chess players of all time, in a 6 match tournament.
IBM built this machine by programming the rules of chess into its system and gave it some basic guidelines, such as the weight of each piece (eg. a rook is more valuable than a pawn). It also had a pre-programmed memory of millions of chess positions, such as opening and closing strategies. However, Deep Blue was not intelligent. Blue could compute the optimal move given any board setup, searching its own library of millions of position, and its game was always reactionary. It didn't out-think Kasparov, it simply out-calculated him.
To illustrate: even though a sports car or jet plane can move much faster than any human, we do not consider it to be athletic. It simply has the correct parts that allow it to achieve such speed. Similarly, chess-playing machines relied on raw power rather than decision making. They did not understand the game or learn from their mistakes.
There have been other chess playing machines since Deep Blue, all of which approached the game in a similar way. However, a revolutionary new machine appeared last year that took a new approach. DeepMind, a company partnered with Google, created a machine called AlphaZero. This machine used deep neural networks to process its moves, rather than a data bank. It was only taught the basic rules of the game (piece movement, check/checkmate rules), but it wasn't given information on the weight of the pieces or any pre-programmed moveset.
Instead, the machine played millions of games against itself. After every game, it would decide which moves where beneficial and contributed to victory. This process, known as reinforcement learning, allowed AlphaZero to develop its own standards for move evaluation and strategy. After a few short hours, it had uncovered many of the most popular strategies that have taken years for humans to discover.
AlphaZero was matched up with Stockfish, a chess machine similar to DeepBlue (albeit exponentially more powerful). Out of 1000 games, AlphaZero won 155 and lost only 6. It was also able to beat other machines playing Shogi (Japanese chess) and Go. in short, AlphaZero destroyed its competition.
effectiveness of AlphaZero, saying that it “prioritizes piece activity over the material, preferring positions that to my eye looked risky and aggressive.” He also noted that the style that AlphaZero plays with mirrored his own, calling it “dynamic and open.”
AlphaZero could mark the beginning of intelligent machines, using programs that can learn over time to become more effective. There is still much to be discovered in regards to the usefulness and shortcomings of these machines. As they are integrated into our everyday lives, we will have to decide if computers are truly intelligent, and if so, how we should use them responsibly.
Questions:
- Do you consider AlphaZero to be “intelligent”?
- What other functions do you think artificially intelligent machines could perform?
1. In my opinion, AlphaZero is intelligent and far superior to its competition because it's come to form an intimidating play style. While some see it as risky it's also a very intelligent way of playing and different from similar AI's because it's thinking more human-like being risky because it has a better understanding of the mechanics rather than just playing the best case scenario as I'm sure a lot of these machines are doing. I like this topic(apologize kind of a ramble not sure if relevant but wanted to share) because I used to be a competitive gamer and I had a very similar play-style when I played.... even though my playing style was very risky I very well knew the mechanics of the game and would that help me to beat well-calculated players because I was making risky moves that someone who was thinking very logically wouldn't predict me to do because they assumed I was playing a probability game like them and was gonna make the most logical choice. Instead, I would weigh out my choices and would make moves on more risky choices because I believed I knew what my enemy would logically think I would do. Again no clue if any relevance but made me think of my experience and why I believe its intelligent
ReplyDelete2. Artificial intelligence in my opinion could perform anything with the right programming and training. In the future I believe they will have a growing presence in major world decisions as they already being used to make future predictions.It's only a matter of time before we highly consider AI opinions in a large portion of decisions since their ability to be fed large amounts of information is something only a computer could do. But why even stop there eventually AI's will be driving all our vehicles, filling service positions and basically doing anything we program to do they're limitless.
1. I do consider AlphaZero to be intelligent. What made AlphaZero different with other similar chess-playing machines, like Deep Blue, is it was only taught with basic rules of the game and the rest was self-taught. It was able to understand, learn, and apply its knowledge and skills with reinforcement learning by playing millions of games against itself. The fact that AlphaZero was able to uncover strategies that took humans years to discover within hours is amazing.
ReplyDelete2. Artificially intelligent machines could perform any function with the training. What interests me is the possibility of using AI machines in medicine and how that would affect us as a society. Could doctors, surgeons, or other medical professionals eventually be replaced by AI machines? It could be possible that AI machines can be taught on how to perform surgeries and with some form of reinforcement learning, AI machines may be able to perform more flawlessly than a human surgeon. The functions of AI machines are endless. I’m curious to see how much we allow these machines into the medical field.