Introduction The image of the ''killer robot" once belonged uniquely to the world of science fiction. This is still so, of course, but only if one thinks of human­ like mechanical contraptions scheming to conquer the planet. The latest weapons systems planned by the Pentagon, however, offer a less anthropo­ morphic example of what machines with "predatory capabilities" might be like: pilotless aircraft and unmanned tanks "intelligent" enough to be able to select and destroy their own targets. Although the existing prototypes of robotic weapons, like the PROWLER or the BRAVE 3000, are not yet truly autonomous, these new weapons do demonstrate that even if Artificial Intel­ ligence is not at present sufficiently sophisticated to create true "killer robots,'' when synthetic intelligence does make its appearance on the planet, there will already be a predatory role awaiting it. The PROWLER, for example, is a small terrestrial armed vehicle, equipped with a primitive form of "machine vision" (the capability to ana­ lyze the contents of a video frame) that allows it to maneuver around a battlefield and distinguish friends from enemies. Or at least this is the aim of the robot's designers. In reality, the PROWLER still has difficulty negoti­ ating sharp turns or maneuvering over rough terrain, and it also has poor friend/foe recognition capabilities. For these reasons it has been deployed only for very simple tasks, such as patrolling a military installation along a predefined path. We do not know whether the PROWLER has ever opened fire on an intruder without human supervision, but it is doubtful that as currently designed this robot has been authorized to kill humans on its own. More likely, the TV camera that serves as its visual sensor is connected to a human operator, and the intelligent processing capabilities of the robot are used at the "advisory" and not the "executive" level. For now, the robot simply makes the job of its human remote-controller easier by preprocessing some of the information itself, or even by making and then relaying a pre­ liminary assessment of events within its visual field. But it is precisely the distinction between advisory and executive capa­ bilities that is being blurred in other military applications of Artificial Intelligence (AI). Perhaps the best example of the fading differences between a purely advisory and an executive role for computers may be drawn from the area of war games. In the war games of the recent past computers played the role of intelligent assistants: human players made decisions affecting the movements and actions of "troops" in the game, while computers calculated the effect of a given attack, using such concepts as a weapon's "lethality index," the rate of advance of tactical units, the relative strength of a given defensive posture or the effectiveness of a specific offensive maneuver. Since their invention in the early nineteenth century, war games have allowed human participants to gain strategic insights and have given offi­ cers the opportunity to acquire "battle experience" in the absence of a real war. This function has become even more important in the case of nuclear war, a type of war that has never been fought and for which there is no other way of training. But in game after game human players have proven reluc­ tant to cross the nuclear threshold. They typically attempt every possible negotiation before pushing the fateful button. This has led war-game design­ ers to create new versions of this technology in which automata completely replace human players : SAM and IVAN, as these robots are called, do not have any problem triggering World War III. To the extent that the "insights" derived from watching automata fight simulated armageddons actually find their way into strategic doctrine and contingency plans, these "robot events" have already begun to blur the distinction between a purely advisory and an executive role for intelligent machines. Now indeed robotic intelligence will find its way into military technol­ ogy in different ways and at different speeds. Traditional computer applica­ tions to warfare (radar systems, radio networks for Control, Command and Communications, navigation and guidance devices for missiles), will become "smarter" following each breakthrough in Al. Mechanical intelligence will once again "migrate" into offensive and defensive weaponry as AI creates new ways for machines to "learn'' from experience, to plan problem-solving strategies at different levels of complexity and even to acquire some "com­ mon sense" in order to eliminate irrelevant details from consideration. But we need not imagine full-fledged, human-like robots replacing soldiers in the battlefield, or robotic commanders replacing human judgment in the plan­ ning and conducting of military operations. These two technologies (auton­ omous weapons and battle management systems) were indeed announced by the Pentagon as two key goals for military research in the 1980s and '90s. But this announcement, made in a 1984 document entitled "Strategic Com­ puting," was as much a public relations maneuver as it was an indication of the military roles that AI will one day come to play. If we disregard for a moment the fact that robotic intelligence will prob­ ably not follow the anthropomorphic line of development prepared for it by science fiction, we may without much difficulty imagine a future generation of killer robots dedicated to understanding their historical origins. We may even imagine specialized "robot historians" cqmmitted to tracing the vari- ous technological lineages that gave rise to their species. And we could further imagine that such a robot historian would write a different kind of history than would its human counterpart. While a human historian might try to understand the way people assembled clockworks, motors and other physical contraptions, a robot historian would likely place a stronger empha­ sis on the way these machines affected human evolution. The robot would stress the fact that when clockworks once represented the dominant technol­ ogy on the planet, people imagined the world around them as a similar system of cogs and wheels. The solar system, for instance, was pictured right up until the nineteenth century as just such a clockwork mechanism, that is, as a motorless system animated by God from the outside. Later, when motors came along, people began to realize that many natural systems behave more like motors : they run on an external reservoir of resources and exploit the labor performed by circulating flows of matter and energy. The robot historian of course would hardly be bothered by the fact that it was a human who put the first motor together: for the role of humans would be seen as little more than that of industrious insects pollinating an independent species of machine-flowers that simply did not possess its own reproductive organs during a segment of its evolution. Similarly, when this robot historian turned its attention to the evolution of armies in order to trace the history of its own weaponry, it would see humans as no more than pieces of a larger military-industrial machine: a war machine. The assem­ bling of these machines would have been, from this point of view, influ­ enced by certain "machinic paradigms" that were prevalent at the time . The armies of Frederick the Great, for instance, could be pictured as one gigan­ tic "clockwork" mechanism, employing mercenaries as its cogs and wheels. In a similar way, Napoleon's armies could be viewed as a "motor" running on a reservoir of populations and nationalist feelings. Nor would robot historians need to ascribe an essential role to great commanders, for these might be seen as mere catalysts for the self-assembly of war machines. Such assemblages, the robot would say, were influenced no more by particular individuals than by collective forces, such as the demo­ graphic turbulence caused by migrations, crusades and invasions. Moreover, our historian would notice that some of its "machinic ancestors," like the conoidal bullet of the nineteenth century, resisted human control for over a hundred years. It simply took that long for human commanders to integrate rifled firepower into an explicit tactical doctrine. Since then, of course, the conoidal bullet has lived a life of its own as one of the most lethal inhabi­ tants of the battlefield. In this sense technological development may be said to possess its own momentum, for clearly it is not always guided by human needs. As the simple case of the conoidal bullet illustrates, a given technol­ ogy may even force humans to redefine their needs: the accuracy of the new projectile forced commanders to give up their need to exert total control over their men by making them fight in tight formations, and to replace it with more flexible "mission-oriented" tactics, in which only the goal is spec­ ified in advance, leaving the means to attain it to the initiative of small teams of soldiers (platoons). When our robot historian switched its attention from weapons to com­ puters, it would certainly also seek to emphasize the role of non-human factors in their evolution. It would, for example, recognize that the logical structures of computer hardware were once incarnated in the human body in the form of empirical problem-solving recipes. These recipes, collectively known as "heuristics" (from the Greek work for "discovery," related to the word "eureka"), include rules of thumb and shortcuts discovered by trial and error, useful habits of mind developed through experience, and tricks of the trade passed on from one generation of problem-solvers to the next. Some of the valuable insights embodied in heuristic know-how may then be captured into a general purpose, "infallible" problem-solving recipe (known as an "algorithm"). When this happens we may say that logical structures have "migrated" from the human body to the rules that make up a logical ­ notation (the syllogism, the class calculus), and from there to electrome- chanical switches and circuits.