AlphaGo, Watson, Deepblue, and Pantonium

By March 11, 2016 No Comments

This week another milestone in computer science history is being set as Google’s AlphaGo AI faces off against Lee Se-do a world champion Go (Baduk) player in a best of five series. Everyone remembers the moment in 1997 when Gary Kasparov was defeated in chess by Deep Blue, or more recently when IBM’s Watson AI crushed veteran Jeopardy players in 2011. But the game of Go is a whole other beast, it was thought to be too complicated, requiring too much human intuition, for an AI to beat a competent human. According to the team that made AlphaGo there are more possible positions in the game than there are atoms in the universe. That means the game cannot currently be solved through brute force computing. However, as of writing this post, AlphaGo is up two games to zero against Mr. Lee. This is quite the upset in the competitive Go world, Mr. Lee had predicted he would sweep the AI, or at least win 4-1. One can only imagine how two successive losses must feel. AI systems are not just here to humiliate humans in board games; they are here to solve real world problems. That is what Pantonium does, our algorithm might not play Jeopardy, but it does function in much the same way as it’s flashy cousins: AlphaGo, Watson and Deep Blue.

 15711446335_6274e44039_kThe primary function of Pantonium’s algorithm is to solve one of the hardest problems in computational mathematics, the traveling salesman problem (TSP). The basic premise of the problem is this: “given a list of cities various distances away from each other, find the shortest path for a salesperson to visit every city exactly once and return to the origin city.” There is a subset of this problem type: ‘the vehicle routing problem’ which is what Pantonium is specifically designed to tackle. Much like Go or chess, there is ballooning of possibilities with the vehicle routing problem. Take a small fleet, doing one day’s worth of trips, with only 50 drop off points, and there will be around one trillion possible routes. Most realistic types of these problems have hundreds if not thousands of drop off points. Now you could try to brute force your way through the problem, but the sun would run out of hydrogen before you solve it.

The tool that is used to overcome “unsolvable” problems like the TSP is called a heuristic, basically we make an algorithm that guesses the best solution instead of checking every single possible one.  And then “tell” the algorithm how you want to solve the problem by setting goals constraints and penalties. Do you care about getting trips done on time above all else? You can set that as the highest priority and the system will heave with all its computational might to make a route that gets your vehicles there on time. The reason why Pantonium and other heuristic algorithms makes guesses is their speed. It is all about churning through the trillions of possibilities as fast as possible to produce the near best solution. That is how Pantonium can manage a transportation system in real time and why how Watson was able to find an answer fast enough to hit the buzzer in Jeopardy.

The re12935316785_eb85f43860_ksult of this speed is your fleet gets a whole lot easier to manage on the day to day. When things change like will calls or cancellations,you must be able to rapidly change the schedule in response to new demands without jeopardizing the efficiency of the plan and without interrupting service. Moments like that are where algorithms outperform even the most experienced humans. The more dynamic, complicate, dense and busy your transportation service is the more powerful the Pantonium system becomes. If you have hundreds of vehicles operating in a single city, with a lot of riders who are sharing seats, you can tell the system to try to reduce the number of vehicles that it takes to transport those people and Pantonium will be able to think through a solution to take as many vehicles off the road as mathematically possible. This reduces the number of vehicles it takes to transport the same amount of people and that saves wages, gas, and maintenance. And it is all due to the work of an algorithm. They can beat us in chess, Go and Jeopardy and they can certainly beat us in routing vehicles. What will they do next, we don’t know, but I hope it doesn’t involve blogging.

If you have any questions about Pantonium, our algorithm, or how we can tailor a solution for your specific problem let us know in the comments or at


Feature Image: REUTERS/Peter Morgan Middle Image: Felipe Maza/Flickr Bottom Image: Atomic Taco/Flickr

Leave a Reply