AlphaGo, a computer program designed by Google’s DeepMind specifically for the purpose of playing “Go” – a board game invented in China more than 2000 years ago – first beat one of the world’s best players at the game in 2016 and after that also beat the world’s best Go player Ke Jie in May of this year. The AlphaGo program was built and taught using the “Supervised Learning” technique of Machine Learning. Supervised Learning basically means that you give the program a lot of examples of the correct solution that correspond to a certain problem – a given input X corresponds to a given output Y – so that it can learn from it (Brynjolfsson & McAfee, 2017).
Supervised Learning is a very useful technique as it can endow programs with the ability to master tasks at a level beyond what humans are capable of achieving. This technique however, has its limits. Programs that use supervised learning rely on data sets that have been pre-labeled by humans, meaning that the program will be limited to following a path predetermined by humans. In a way this can be seen as the machine’s approach to learning being inspired/guided by humans.
However, maybe in the future this can be exactly the other way around! As many of you may already know, the counterpart to supervised learning is unsupervised learning. Continuing the story of AlphaGo, another version of this program was later built called AlphaGo Zero, which learned using the Unsupervised Learning technique. Using this technique, the program is only given a set up inputs (X) without having these correspond to any predetermined output (Y). In the case of AlphaGo Zero this means that the program was only given the rules of the game and a reward system for when it gets points, without seeing any examples of how the game is played. What this resulted in is the program coming up with strategies that humans did not even think of during the entire existence of the game! The program got so smart that it also kept beating the original AlphaGo trained using the Supervised ML technique. After all this happened, the world champion in Go studied the gameplay of AlphaGo Zero to make himself better. So now a human’s approach to learning has been inspired/guided by a machine!
Now just think about phenomenon being applied in other contexts. For example, in the field of marketing, a particular program given with the objective of identifying consumer buying behavior patterns coming up with some interesting insights that humans never thought of before and that would radically change the way we think about how to market products. This will certainly be also true for endless other fields especially in science, leading to disruption not only in these fields but also in the thinking patterns of humans!
Do you agree that Unsupervised AI can inspire humans in the future down paths that we never thought of before, or are you one of the people that think that AI will replace us completely? Do you have more examples of how humans can be inspired by AI? Please share your thoughts in the comments.
References
Brynjolfsson, E. and McAfee, A. 2017. The Business of Artificial Intelligence. Harvard Business Review.
The Economist (2017). The latest AI can work things out without being taught. Retrieved October 21, 2017 from: https://www.economist.com/news/science-and-technology/21730391-learning-play-go-only-start-latest-ai-can-work-things-out-without?fsrc=scn/fb/te/bl/ed/thelatestaicanworkthingsoutwithoutbeingtaught
Hi Ruenswin,
I think AI Will change a lot, however have you also considered the down sides of AI? Like the bias they create, or the fact that they lack verifiability? I see potential problems in the use of AI, but they also have a great potential. Yes, I do think they will replace some jobs in the future, but they will also create new job opportunities! Interesting subject!