Over the past few years there have been a number of proclamations on the rise of machines and artificial intelligence in the workplace. We’ve had Google testing driverless cars, IBM’s Watson beating the grand master of Jeopardy and beating a path to a hospital near you.
Probably the finest chronicle of this revolution has been The Second Machine Age by MIT academics Andrew McAfee and Erik Brynjolfsson, which suggests that there will soon come a time when machines can easily replicate everything that a human can do.
A sign of the progress being made comes via a recent DARPA backed project that is working on a new robot that can learn how to perform the intricate maneuvers required to cook a meal, all by watching a YouTube video.
The machine is capable of recognizing how utensils are operated in a range of YouTube based cookery videos, before then replicating those movements without any outside intervention.
DARPA are believed to be working on the project in the hope that the fine motor movements required in cookery will enable their robots to perform similar movements in other settings.
Existing robots are quite good at pattern recognition, but are much less skilled in visual interpretation or the performance of actions based upon that information.
The ability to therefore take visual information and translate that into action is a vital next step to take in the evolution of robotics.
Each robot was trained by watching a series of cookery videos on YouTube. The robot utilized neural networks to both recognize the objects being used in the video, and then to categorize the type of grip used when handling each object. Attempts were then made to mimic the actions used in the videos.
After a period of training, the robot was able to correctly recognize the objects used in the videos around 79 percent of the time. It was also able to recognize the grip used with an accuracy of 91 percent, whilst it was 83 percent accurate with the action used.
The robot was also quite adept at remembering many of the things that it had learned, so was capable of sharing that knowledge with other robots.
“This learning-based approach is a significant step towards developing technologies that could have benefits in areas such as military repair and logistics,” the researchers say.
It’s likely to be some time before we see robots in a kitchen near you, but this is another sizable advance in a truly fascinating field.