The Sheer Difficulty of Defining What a Robot Is
Right now, nobody can agree on what a robot is.
Image: i k o/Flickr
What is a robot? You might say that a robot has arms and legs, like C-3PO. Or, maybe, a robot is more like the drones in hokey sci fi classic Silent Running—they don't need to be humanoid to be called a robot, they just need to move. Perhaps, though, a robot is really more like HAL 9000; it doesn't have a body per se, but it controls physical objects and systems. Wait, does a robot even need a body?
Ah yes, bodies. These deteriorating, decrepit, frail things we call meatbags. Our experience as humans is intimately tied to our embodiment: we feel and think with meat, and our memories are etched in sensory inputs. In the Judeo-Christian tradition, the body is a curse that carries with it the pain of childbirth and eventual death to remind us of a time, millennia ago, when we spurned God. And now, the story goes, we're trapped until we die and trade our imperfect bodies for heavenly astral forms.
Who could blame us for imagining that robots, the objects we make in our own image, must be similarly caged? But whether a philosophical conundrum or a mere bias of vocation is to blame, nobody can agree on what a robot is, in terms of their robo-bodies; not roboticists, and especially not policy makers and labour advocates. This is a problem, because to reckon with all the coming changes that robotics and artificial intelligence may bring—widespread worker displacement and killer machines, to name just a couple—we need to know what we're talking about.
"Many different existing legal regimes will be able to competently address problems posed by robots without having to settle on a standard definition," Woodrow Hartzog, an assistant law professor with a specialization in robotics at Samford University, told me. "However, robots might also present new challenges that require new rules. If those rules are going to be limited to certain technologies, the decision will have to be made as to whether a definition of what a robot is will be necessary."
As a bit of an illustrative exercise, I asked some very smart people a pretty simple question, at least on the surface: "What is a robot?" I received answers dripping in ambiguity.
Selmer Bringsjord, chair of the department of cognitive science at the Rensselaer Polytechnic Institute, told me in an email, "An [artificially intelligent agent] that has physical sensors and effectors, and whose physical parts are all physically connected, is a robot. Most AIAs are not robots. I think that one of the most interesting questions is what exactly are the creatures that are AIAs, and have physical sensors and effectors, but do not meet the second condition—i.e., their physical parts are not connected."
Here, there is a distinct line drawn between artificially intelligent software alone and software paired with hardware that can interact with its physical environment. Its physical sensors and "limbs" must also all be connected, like our arms and legs are connected to our torsos. An assembly line robot would definitely be a "robot" here, but what about an AI that controls and directs the different parts of a factory? It has physical "sensors and effectors," but its parts are not physically connected.
"I would advise we avoid trying to define the term robot"
Michael Milford, senior lecturer of electrical engineering at the Queensland University of Technology, gave a slightly broader definition. "There's no consensus but here are two to work with: A robot is a machine that senses the world around it, thinks and acts," he explained. "This isn't fully satisfactory, as an air conditioner fits [that definition]. So we extend it to: A robot is a machine that senses the world around it, thinks, and acts in some physical/mechanical way."
Okay, now we're talking. In the aforementioned factory scenario, the artificially intelligent factory overlord would be, in fact, a robot, despite not walking around and cracking wise like TARS. But when I asked Jamie Haines, a representative of Canadian labour union UNIFOR, for his opinion on robots in the workplace in an interview last month, he had—again—an entirely different idea of what a robot is.
"A robot has to be put behind fencing and all that," Haines said. "[An unmanned autonomous vehicle] is not a robot, under the definition of what a robot is. A robot has multiple actuating arms, and it's controlled by a controller."
All of these different definitions add up to a vague idea of what a robot is: it can plan, think, and interact with its environment in some way. At issue is how extensive that interaction has to be, and what embodiment means in this context. Is the real substance of a robot its computer brain? Or the casing that brain goes into? It's a question that mirrors our own, in terms of our fleshy bodies, and it's still debated by philosophers. But with robots, the question of embodiment has immediate real-world implications.
Constraining our definition of robots to bots like those in the video above could limit our ability to address the change they bring to society.
When we think of robots in labour, for example, do we think of rows of mechanical arms installed on a factory floor, piecing together components that a human used to? Or, should we also consider algorithms that control a factory more indirectly, whether by organizing and carrying out the flow of components throughout the space, or something else? What about the software that automates spreadsheets and white collar workflows, or monitors employees for unsavoury behaviour?
Holy shit, is capitalism itself a robot?
Okay, that last question was a bit tongue-in-cheek, but the idea that robots can be systems is not so far-fetched. In a largely automated system that makes demands on our bodies—where to stand, what to do, how to act—we could be cast, as Karl Marx proposed in "The Fragment on Machines," as mere "conscious linkages" in a system governed by machine calculation and control. There might be no evil humanoid robot like Futurama's Bender or Star Wars' HK-47 issuing orders to us puny meatbags, but the result is the same.
And what's the difference here? Should we think of these things as entirely different and address them separately in terms of labour's resistance to unadvantageous automation, policy-making, and regulation? The core tension is between a robot as something that has a body and as a "system" that has a more fluid relationship with embodiment.
It's worth noting here that the origin of the word "robot" comes from a 1920 Czech play called "Rossum's Universal Robots." In the play, robots are artificial humanoid workers, and the term "robot" itself is derived from the Czech word "roboti," which translates to "serf labor". So, slaves, basically. Humanoid labourers with bodies that mimic ours. Is this nearly century-old conception one we should cling to after decades of technological development?
These are all important questions. But, perhaps a systems-based approach to thinking about robotics is more instructive than a narrow one. At the very least, it's more compelling to me. For now, however, maybe we don't need a strict definition of a robot as the technology is still in its nascent stages.
"In the meantime I would advise we avoid trying to define the term robot," Hartzog suggested. "We've managed to regulate other nebulous concepts like privacy to varying degrees of success without a settled definition of the term. We might be able to do something similar with robots."
Goodbye, Meatbags is a series on Motherboard about the waning relevance of the human physical form. Follow along here.