On May 11 1997, a famous chess match was held in New York City. The game, which lasted for 6 days, was best remembered in computer science and artificial intelligence (AI) history, because it was the first time that humans were defeated by computer systems. World chess champion, Garry Kasparov, was beat by a supercomputer designed by IBM named Deep Blue under standard chess tournament time controls. Deep Blue was specifically designed for playing chess. Developed with the best technology of the time, Deep Blue was capable of processing 1 trillion operations per second. Compared to Kasparov’s ability of examining and evaluating up to three chess positions per second, Deep Blue was able to explore up to 200 million chess steps per second. The enormous calculation ability of Deep Blue helped it win the game. Deep Blue’s success made scientists wonder what the next steps of computer development will be. Will computer systems be intelligent enough to beat human in other fields in the future?
The movie Terminator 3 describes how highly intelligent computer systems take over the world and lead to human genocide. It sounds depressing that human species are terminated by something designed and made by themselves. However, at least the terminator of human species won’t be Deep Blue, because it doesn’t have the intelligence of learning and reasoning. For example, Kasparov was able to analyze the weakness of his opponents and react to his failures or successes. He used strategies to play the game. He even used an anti-computer tactic by playing an esoteric opening that successfully confused the computer. On the contrary, Deep Blue had no “receptor” to receive information from the environment that it wasn’t able to analyze Kasparov’s performance. Deep Blue used his enormous calculation ability to play the game by coming up with every possible step. Doesn’t it sound dull that coming up with every possible step that takes a lot of time and energy without thinking strategically, though time may not be an issue for Deep Blue since it calculated 200 million steps in a second? In summary, Deep Blue was just a high-speed computer programmed with chess rules but no intelligence at all. In terms of strategy thinking and learning ability, Kasparov won the game. And in fact, he won game 1 and reached an even score with Deep Blue before the decisive loss of game 6.
Kasparov’s almost equally good performance with Deep Blue in the game indicated that computing speed wouldn’t out perform human intelligence. The ability of learning and thinking strategically makes humans special. Ridley implied a similar idea in his The Agile Gene that humans have the gene of absorbing information from the environment and respond to experience accordingly (Ridley, 2004). To move forward with the idea of developing computers that are compatible with humans, incorporating intelligence—the ability of absorbing information and reacting to environment is important. Therefore, the science of artificial intelligence (AI) was born. One major mission in AI studies is simulating learning mechanism of human brain in computer systems. The neuroscience of learning is
well explained by Donald Hebb that learning is the result of strengthening synapses between neurons through repeated firing (Ridley, 2004). Inspired by Hebb, Frank Rosenblatt built an algorithm that could learn and correct errors by adjusting the strengths within layers of the program (Ridley, 2004). Rosenblatt’s primitive learning machine was further developed by IBM’s research team in 2014 when they released their SyNapse processor chip. Though what stood for SyNapse is different from “synapse”—the connection between neurons in Hebb’s theory, SyNapse utilizes mammalian neuron circuit model. It is consisted of over one million “silicon neurons” organized into 4,096 identical blocks of 250, similar to mammalian brains, which are built out of repeating circuits of 100 to 250 neurons (Simonite, 2014). With a program similar to Rosenblatt’s learning machine, SyNapse is able to react to information. In another word it can learn information to the extent of human brain’s capability. There’s no need to program for the chip since it can learn by itself through operant conditioning.
“You don’t program it. You just say ‘Good job,’ ‘Bad job,’ and it figures out what it should be doing.” –Narayan Scrinivasa, IBM Principal Research Scientist (Simonite, 2013)
SyNapse is still being tested, but its older brother, Watson, a supercomputer equipped with a self-learning and language system which beat Ken Jennings in the popular game show Jeopardy! on TV in 2011, has shown artificial intelligence’s capability of actively engaging in daily activities. Watson won the game because of its learning ability, rather than fast computational ability seen in Deep Blue. In the preshow test, Watson failed to understand a category of questions which asked for month instead of date. However, after listening to a round of answers of the other two participants, it understood what this category asked for in answers. It automatically adapted to the new environment and answered correctly in the next round. The success of Watson in Jeopardy! made scientist think again what the future of AI will be. With strong computation and learning ability, AI may become part of our life. Having a robot like Baymax which understands us, talks to us and does something together with us won’t be a Disney movie any more.
More information of SyNapse can be learned here:
Documentary on Watson’s Jeopardy! victory can be learned here:
- Ridley, M., & Ridley, M. (2004). The agile gene: How nature turns on nurture. Toronto: HarperPerennialCanada.
- Simonite, Tom. “Thinking in Silicon.” MIT Technology Review. N.p., 16 Dec. 2013. Web. 11 Feb. 2016.
- Simonite, Tom. “IBM Chip Processes Data Similar to the Way Your Brain Does.” MIT Technology Review. N.p., 7 Aug. 2014. Web. 11 Feb. 2016.
One thought on “The Future of Artificial Intelligence: Cognitive Computing”
Finiding game steps like Tic Tac Toe or three dots is an example of simple ai but learning and percepting input from invironment is Deep Blue or watson is real ai. Machine learning is introduced when descision making becomes more vast and complex. Learning from errors and correct the neural weight accordingly.