Episode 122 – “Steel”

At least since the 17th century, philosophers and scientists have considered whether it is possible for machines to think. Descartes argued that it is not possible for animals (and by extension,  machines) to think. Thinking, he argued, requires the use of language, and language use is open-ended and unpredictable. While animals, which are simply mechanisms, can act “from the disposition of their organs,” humans act “from knowledge,” in a flexible manner that enables appropriate responses to any provocation. Only humans have language, and so, only humans can think.  Other beings can respond appropriately based on habit to a narrow range of inputs. Not all philosophers in the 17th and 18th centuries agreed with Descartes. Thomas Hobbes held that thinking was a kind of computation, carried out by motions in the brain and the heart. Pascal and Leibniz both worked on mechanical computing devices, with the hope of achieving a universal computer that could communicate in the language of mathematics.

In 1950, Alan Turing, one of the architects of the modern digital computer, took on Descartes’ challenge, and suggested that if a computer could be built that has the capacity to communicate in a natural language like English, in the same way that humans communicate in language, then we would have to conclude that it is intelligent. Turing proposed a test, since called The Turing Test,” where a computer replaces a human being in a parlor game called “The Imitation Game.” The Imitation Game is usually played with three human players: a male player, a female player,  and a judge. The players communicate with the judge via printed text. The judge can’t see or speak with the other players. The goal of the female player is to convince the judge that she is the female, and that the other player is the male participant. The goal of the male player is to convince the judge that he is the female and that the other player is the male. The Turing Test replaces either the male or female with a computer, who then takes on the role of the person it replaces. If the computer, for example, playing the male, can achieve the same rate of misidentification by the judge as a human player, then the computer passes the test.

Turing thought that the test was appropriate because it tests only the linguistic skill of the computer. The judge can’t see the body of the computer during the game, and won’t be likely to rule out its ability to think in advance of its exchanges with the machine. Turing reasoned that thinking is a purely intellectual activity, and doesn’t depend on the looks or other physical characteristics of the machine, beyond its ability to produce appropriate typed responses to the questions presented by the judge. Turing predicted that within fifty years, that is, by around 2000, computers would be powerful enough to pass the Turing Test, and be truly artificially intelligent.

“Steel” is about robots, embodied computers, that are not quite up to the  conversational skill level of the computers Turing envisioned,  but have been designed for a less lofty purpose. In this story, robots have replaced humans in the sport of boxing. Instead of dangerous bouts between vulnerable human players, boxing (in the 1970s!) has evolved into a sport between teams of roboticists, who develop androids who fight in front of audiences, just as the human boxers of old did. The moral objections to the sport of boxing have been accommodated. Robots damaged in a bout can be repaired or junked. No one gets hurt.

Worried that their robotic boxer will not be able to perform, due to technical difficulties, one of the promoters of the robot decides to replace the robot in the boxing ring, and attempt to fool the judges and the audience into thinking that he is the robot. In contrast to the Turing Test, where a robot takes on the role of a human, and tries to convince the judge that he is the male or female human being, here a human, nicknamed “Steel,” tries to convince the judges that he is a robot.

How difficult is it for a human to convincingly imitate a robot? Of course that depends on what the robots are like! In this story, robots look very much like humans, and they are designed to engage in a physical activity that developed originally as an activity between humans. The closer the robot is to imitating human boxers, the easier it will be for a human boxer to imitate a robotic one.

One might think that it would be easier to build a robotic boxer than a robotic thinker. In fact, both are challenges that we have yet to meet. Twenty years after the date Turing specified as the date by which the Turing Test would be passed, we do not have a computer that can engage in the kind of open-ended communication in language required to pass the test. While we are making progress with autonomous vehicles, we are not close to constructing autonomous agents who can compete in sports.

Robotics has in fact developed since the airing of this episode in the early 1960s along the lines hypothesized here. Many tasks carried out by human beings, from boxing to mining, from factory work to driving a car, are dangerous, and there are clear benefits to turning those tasks over to intelligent robots that can perform the associated actions for us.

If robots could imitate humans, and humans imitate robots, in a wide range of activities, intellectual, physical and social, would the lines be so blurred that we would have to attribute not only intelligence, but feelings, and not just feelings, but rights and obligations, to our robots? Paul Ziff raised this question in 1959, and it’s still with us today.

Further Reading:

Descartes, Rene, Discourse on the Method, 1985, in The Philosophical Writings of Descartes, Volume 1, Cottingham, John, Stoothoff, Robert, Murdoch, Dugald, trans., Cambridge University Press.

Turing, Alan, 1950, “Computing Machinery and Intelligence,” Mind Vol. 59: 433-460. (also cited in “The Lonely.”)

Ziff, Paul 1959, “The Feelings of Robots,” Analysis 19.

This entry was posted in reflections. Bookmark the permalink.

Comments are closed.