Quietly, in the background, robots are getting smarter. So much smarter that they may soon be able to accomplish a lot of our work for us. They can drive us and our stuff around, manufacture a lot of that stuff, prepare our meals and care for our elderly and infirm. This is not to say they will take over the world, but they are certainly poised to increase their population exponentially.
In the forefront should be our concerns about the social upheavals that will accompany this radical change in the economics of labor. Instead, we have sci-fi movies about robots gone wild and high-minded debates about the indifference of machine superintelligence to humanity. Neither approach is all that helpful right now as we await the robot renaissance. Martin Ford, in his highly regarded new book, Rise of the Robots, argues for the morally sound—but politically impractical—concept of a basic income. What is hard is imagining how we humans will fit into a robot-filled future.
Some recent developments at the intersection of robotics and the branch of artificial intelligence called deep learning give us some clues about how our relationship to machines is going to evolve. A report from UC Berkeley, last week, described how researchers there have been using deep learning to train robots to do everyday human tasks that require dexterity and spatial awareness. The report claims that the team has “developed algorithms that enable robots to learn motor tasks through trial and error using a process that more closely approximates the way humans learn, marking a major milestone in the field of artificial intelligence” (see video on second page of this post).
Meanwhile, a Boston company called Rethink Robotics has already developed a pair of industrial robots that can work amongst and be trained by humans on the factory floor. Baxter and Sawyer (see image above) are designed to be helpful and unthreatening to humans. We will have to see how it all works out. There will likely be cases of employees “going gadget” or exhibiting “robot rage,” but, for the most part, the relationship should be collegial.
Certainly these are very well-designed robots. They are stylish and personable without being humanoid in any uncanny-valley-crossing way. And they are, in the end, very good and dependable workers that can take a lot of the repetition out of human work. It is important to remember that humans like to get paid, but most humans harbor no love for the work they do to get paid.
Most people who are on the front lines building robots and AI systems see these technologies in terms of how they can help people do more. A human assisted by a robot or a team of robots can be 10x or 100x more productive than they are on their own. Rethink Robotics calls these devices “collaborative robots.” Indeed, humans train Baxter by grasping its arm and showing it how to move through a series of gestures required to do a task.
There is something different in kind about these new robots compared to previous industrial robots. These collaborative robots take humans into account. They need us to learn, as children do. And they are deferential to us in space, as well-trained pets are. If Baxter bumps into your arm, it immediately goes slack so as not to harm you, as a Labrador learns to have a “soft mouth” with a hunted duck. (What happens when one Baxter bumps into another, I’m not sure!)
Similarly, the trial and error process that BRETT (the Berkeley Robot for the Elimination of Tedious Tasks) uses to learn tasks is endearing. In the video below you can see the robot learn to fit plastic pieces together, first for a toy plane and then for some oversized Legos. Although any one of these tasks could be hard coded, the advantage of Berkeley’s deep learning approach is its fault tolerance. Because BRETT learns through approximation and optimization, it can make adjustments on the fly if the position of objects changes somewhat.
Professor Pieter Abbeel of UC Berkeley’s Department of Electrical Engineering and Computer Sciences explains, “The key is that when a robot is faced with something new, we won’t have to reprogram it. The exact same software, which encodes how the robot can learn, was used to allow the robot to learn all the different tasks we gave it.” This approach mirrors the way humans learn. “For all our versatility, humans are not born with a repertoire of behaviors that can be deployed like a Swiss army knife, and we do not need to be programmed,” says postdoc researcher Sergey Levine. “Instead, we learn new skills over the course of our life from experience and from other humans.”
These technologies are a long way from any sort of general intelligence, but they are incredibly practical. Not only will robots change the world of work, but they will change the way we understand what it means to be human. By externalizing the learning process and making it accessible to non-technical humans, these ubiquitous robots will also change how we teach each other.
There are many ways to frame the danger of this disruption. One revolves around the human propensity for laziness. Although the training of robots could inspire us all to make things better and better, it could also plateau into the mediocrity of “good enough.” If the quality of what robots can produce becomes the default at some sub-optimal level, we may find ourselves doubly impoverished, shorn both of meaningful employment and true material pleasure.
Let me end on a more positive note that I hope to expand upon in a forthcoming post. We can use the power and productivity of learning robots to address the limitless material, social and environmental problems that confront us. There is no end to the people that can be fed, clothed and housed. There is no end to the land that can be reclaimed from industrial toxins and transformed into production and pleasure for humans and animals. Finally, there is no end to the ways that humans can productively work with one another if they are no longer riven by the conflicts of scarcity. Perhaps we will learn to love our robots.
This article was written by Anthony Wing Kosner from Forbes and was legally licensed through the NewsCred publisher network.