Google's Robotic Arms To Be Used To Improve Machine Learning

Googles Robotic Arms

Normally, when looking into an acquisition that will bring a new company or service into Google’s fold, Larry Page says something must pass the ‘toothbrush test’. If the product or service will be used daily by millions of people and improve their lives, much like toothbrushes, it’s a go. Otherwise, it is unlikely to make it. With Google’s culture as gung-ho as it is about internal entrepreneurship and the concept of moonshots, some may wonder what happens when a product or service developed internally doesn’t pass the toothbrush test. In the case of specially developed robotic arms, an offshoot of the somewhat meandering Boston Dynamics, that can lift about 10 pounds, the product is put to work inside Google for another purpose. In this case, the robotic arms will be used to analyze and enhance machine learning, Google’s longtime obsession.

The robotic arms, about 50 of them in all, were manufactured for testing purposes, but when engineers started to teach them simple tasks via AI, they noticed something; the robots, if properly set up and networked, could learn from and bounce off of one another. In essence, what had been developed was a sort of active neural network. Robotic arms that had been taught to open a door, for instance, could analyze a door that they had never seen before and figure out how to open it based on common traits it shared with the door that they were familiar with. The kicker is that the process was shorter and less intensive than such things usually are, because of the robots compiling each others’ learnings, a feat that Google is calling “collective learning”.

Google plans to continue using the robots to study and perfect collective learning. The happy accident fits Google’s agenda perfectly, with neural networks and machine learning powering the futuristic AI that is making it into a number of Google products and services, such as Google Assistant and self-driving cars. For now, given the robots’ simple knowledge base and low strength, they aren’t particularly helpful to anybody, corporate or consumer; should that change in the future as they continue to learn, Google may reassess their marketability or base future products off of them, though for now, it’s purely a wait-and-see affair. According to a Google spokesperson, like anything else with Google’s fledgling fascinations, not much can be said because we’re still in the “early days“.