Thursday, August 9th, marked the first public release of the Opex Analytics Beer Game. Previously, Larry Snyder provided a great backdrop on the classic game in his Brief History of the Beer Game blog.
In truth, the classic beer game works just fine, whether it’s played sitting around a table using chips and cards, or played on the web. Because of this, we realize you may be wondering:
“Why did Opex decide to invest their time and money into making the Opex Analytics version of the game?”
Honestly, the project has been both extremely challenging and very fun. As I think back to the beginning of that decision, I realized there are five main reasons we felt we had to create our own version.
1. We didn’t want the researchers working on machine learning for video or board games to have all the fun.
It’s been enjoyable to watch the developments in deep learning (DL) and reinforcement learning (RL) over the last several years. Recently, we witnessed DL and RL players beating the world’s best human player in the game Go. Prior to this, it was long thought computers would have trouble matching the best players in the world. This success made headlines around the world as researchers proved the ability to bring machine learning algorithms on par with top-ranked human players.
At the same time, researchers were teaching RL agents to play old Atari games. The video below shows how they trained the agent to play the game Pong.
After just a few training games, the agent was terrible (A). But, after hundreds of games, the algorithms figured out that the best strategy was to build an opening up the side and then let the ball bounce around up top, knocking out lots of blocks without effort (B).
In general, this work looked like a lot of fun. As Larry Snyder recounts, he and his students thought it would be fun to take this same technology to the Beer Game and see whether it could beat a human player. From there, the research behind the Opex Analytics Beer Game was born.
2. We wanted to educate business leaders on the value of reinforcement learning.
Go and Atari games are interesting, but it’s difficult to relate those games to business problems. The Opex Analytics Beer Game allows people to see reinforcement learning in a business setting. And, the classic beer game is already taught in many MBA and Supply Chain programs. We figured that this would be a good business game to build on that many in operations were already familiar with.
3. We like to teach complicated topics with simple examples.
The beer game teaches people how variability can propagate up the supply chain and shows how information can help improve the situation. It’s easier to teach this concept with a game rather than equations.
Then, by adding reinforcement learning to the game, we get to teach yet another concept that people are interested in.
4. We wanted to showcase our capabilities at Opex.
Reinforcement Learning is on the cutting edge of technology. And, while we have many other projects that are also on the cutting edge of technology, those are often the projects our clients want to keep private. The Opex Analytics Beer Game gave us a way to show some of the advanced AI capabilities that are out there without giving away any trade secrets.
5. We wanted to inspire business leaders to come up with creative uses of AI.
We are just at the start of having AI reinvent how businesses are run. By adding AI to the Opex Analytics Beer Game, we wanted to get people thinking about new ways to apply this type of technology. This is just one example. We think we have a lot of good uses for AI, but we know the best ideas are going to come from the people we work with who are closer to the day-to-day problems and opportunities. Our hope is that this game inspires people to think more seriously about where and how AI can best work for them.
So, there you have it. From a fun concept, to educating, to showcasing and most importantly inspiring businesses to seriously consider the possibilities of AI in their own operational decision making, the Opex Analytics Beer Game is something we felt strongly warranted our time and efforts. Don’t just take our word for it though, play the Opex Analytics Beer Game and let us know what you think!
Source: Deep Learning on Medium