How J.A.R.V.I.S defeated Captain America to become my most favorite superhero?
I am ardent fan of the Marvel Universe just like most of you. Until recently my most favorite character was Captain America, along with other favorites IronMan, Wolverine, Deadpool and this list goes long. But since, I started my grad school about a year ago a little known character from this Universe has climbed up the ladder to topple Captain America and become my most favorite superhero. Any guess who this can be? Its J.A.R.V.I.S. or Just A Rather Very Intelligent System, Tony Stark’s artificially intelligent computer. J.A.R.V.I.S. has now became VISION after Age of Ultron. But why did this happen? How did Captain lost this battle? The reason is Deep Learning. Since I have read, learnt and realized potentials and possibilities of Deep Learning I see JARVIS as reality and not just a fiction character like other action heroes. Nobody has ever witnessed a Wolverine or Ironman or Hulk in real life but all of us (the Data Science community) has witnessed power of Artificial Intelligence (AI). Algorithms serve as its superpower that draw strength from “data”. This is what makes companies like Google, Facebook and Microsoft a superhero in Silicon Valley. These algorithms are becoming more powerful each day by gathering more data. So it is very likely that we will see J.A.R.V.I.S. in action soon if not Captain America.
Google’s AlphaGo was in news recently as it defeated Lee Sedol at the Google DeepMind Challenge Match by 4–1 in March this year. The match has been compared with the historic chess match between IBM’s Deep Blue and Garry Kasparov in 1997. But AlphaGo’s victory in 2016 is far more significant than Deep Blue’s victory in 1997. It not only signifies evolution of Artificial Intelligence (AI) in past twenty years but also indicate how machines have achieved near human thinking ability. Both Go and Chess are famous strategic board games played by two players and involve no random elements such as in Backgammon. These games are solely based on reasoning of players, but Go is much more complex than Chess. A typical game on chess might last 80 moves whereas Go might go for 150 moves. Similarly, at each game state Chess has nearly 35 possible moves whereas Go can have 250. This makes Go far more complex than Chess. Therefore, AlphaGo’s victory at Go is far more significant than Deep Blue’s victory at Chess. To play games such as chess and Go AI use game trees. A game tree represents a game from start to end in form of nodes (game states) and edges (game moves). A game of chess would contain such 10¹²⁰ nodes.
At each move AI player would choose a move that would minimize its loss (minimax algorithm). Deep Mind used brute force approach which combined minimax algorithm and evaluation function. If AlphaGo was to use a similar approach for Go it would have to traverse a tree with 10⁷⁶¹ nodes. This meant AlphaGo was useless. AlphaGo uses Monte Carlo Tree Search to traverse the tree which is guided with Machine Learning and Neural Networks (Deep Learning). AlphaGo has achieved this fete with massive computation power and learning algorithms that have evolved since 1997. In addition it has relied on large datasets of games played by human players before.
We have seen IBM Watson and Google’s DeepMind in action in healthcare, finance, marketing but these machines are yet to give their best. Artificial Intelligence will evolve over next five years to build J.A.R.V.I.S. and I am waiting to see my superhero in action.