Let’s say the beer is reviewable. We are in Las Vegas at re:Invent, the world‘s largest cloud event dedicated to Amazon Web Services (Aw) technologies. More precisely, we are at the entrance to AWS Builders Faire, a place as big as the largest Italian fair dedicated to technology that showcases demos and projects created by employees and partner companies on the subject of learning systems and machines. Beer was offered by Robo Tap Room, a system made up of a mixer and a mechanical arm that taps and serves the beer. We tried it.
How did it go? The plastic cup bent and the beer was full of foam. But we were first in line and they explained to us that the system will eventually learn and improve. This applies to Robo Tap Room as well as many other solutions that use machine learning algorithms to learn. It takes time, they tell us, because machines know nothing about beer, they only know what we teach them or how we tell them to learn. As in the case of the Aws DeepRacer.
Launched a few years ago, four years to be more precise, it’s a cloud-based 3D racing simulator where you race a fully self-driving 1:18 scale racing car. It is not a racing game for developers but the application of reinforcement learning or reinforced learning. The machine learns through interaction with the environment. He misses the curve, returns to the road, in short, we proceed by trial and error. It does not receive instructions from the programmer as is the case with supervised learning. There is no one to tell him when to accelerate and when to swerve. Learn based on context and therefore without the need for labeled learning data.
There is a championship with Intel that took place in Las Vegas on the most demanding track in the history of AWS DeepRacer, 33.22 meters. There are also hands-on machine learning tutorials for schools (US only) to help you learn the basics and train reinforcement learning models. During the demonstration in front of the international press we saw two cars reversing followed by two people who put them back on the right track at every wrong turn.
The feeling of failure was the same felt with beer. But it is an error of perspective into which it is easy to fall. Machine learning is increasingly a training ground for failures. And it has to be like this, it works like this. let’s say that is also its beauty. It takes time, a long time, also because the goal isn’t to finish first or to serve the best beer in the area. But learn how to do it. And in a new way.