Sunday, April 13, 2008

Asimov's laws of robotics...

Asimov's Laws of Robotics
Implications for Information Technology


Roger Clarke

Part 1:


With the death of Isaac Asimov on April 6, 1992, the world lost a prodigious imagination. Unlike fiction writers before him, who regarded robotics as something to be feared, Asimov saw a promising technological innovation to be exploited and managed. Indeed, Asimov's stories are experiments with the enormous potential of information technology.

This article examines Asimov's stories not as literature but as a gedankenexperiment - an exercise in thinking through the ramifications of a design. Asimov's intent was to devise a set of rules that would provide reliable control over semi-autonomous machines. My goal is to determine whether such an achievement is likely or even possible in the real world. In the process, I focus on practical, legal, and ethical matters that may have short- or medium-term implications for practicing information technologists.

Part 1, in this issue, reviews the origins of the robot notion and explains the laws for controlling robotic behaviour, as espoused by Asimov in 1940 and presented and refined in his writings over the following 45 years. Next month, Part 2 examines the implications of Asimov's fiction not only for real roboticists but also for information technologists in general.

Origins of robotics:

Robotics, a branch of engineering, is also a popular source of inspiration in science fiction literature; indeed, the term originated in that field. Many authors have written about robot behaviour and their interaction with humans, but in this company Isaac Asimov stands supreme. He entered the field early, and from 1940 to 1990 he dominated it. Most subsequent science fiction literature expressly or implicitly recognizes his Laws of Robotics.

Asimov described how, at the age of 20 he came to write robot stories:

"In the 1920's science fiction was becoming a popular art form for the first time ..... and one of the stock plots .... was that of the invention of a robot .... Under the influence of the well-known deeds and ultimate fate of Frankenstein and Rossum, there seemed only one change to be rung on this plot - robots were created and destroyed their creator ... I quickly grew tired of this dull hundred-times-told tale .... Knowledge has its dangers, yes, but is the response to be a retreat from knowledge? .... I began in 1940, to write robot stories of my own - but robot stories of a new variety ...... My robots were machines designed by engineers, not pseudo-men created by blasphemers"1,2

Asimov was not the first to conceive of well-engineered, non-threatening robots, but he pursued the theme with such enormous imagination and persistence that most of the ideas that have emerged in this branch of science fiction are identifiable with his stories.

To cope with the potential for robots to harm people, Asimov, in 1940, in conjunction with science fiction author and editor John W. Campbell, formulated the Laws of Robotics. 3,4 He subjected all of his fictional robots to these laws by having them incorporated within the architecture of their (fictional) "platinum-iridium positronic brains". The laws (see below) first appeared publicly in his fourth robot short story, "Runaround"5.

The 1940 Laws of Robotics:

First Law:

A robot may not injure a human being, or, through inaction, allow a human being to come to harm.

Second Law:

A robot must obey orders given it by human beings, except where such orders would conflict with the First Law.

Third Law:

A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

The laws quickly attracted - and have since retained - the attention of readers and other science fiction writers. Only two years later, another established writer, Lester Del Rey, referred to "the mandatory form that would force built-in unquestioning obedience from the robot".6

As Asimov later wrote (with his characteristic clarity and lack of modesty), "Many writers of robot stories, without actually quoting the three laws, take them for granted, and expect the readers to do the same".

Asimov's fiction even influenced the origins of robotic engineering. "Engelberger, who built the first industrial robot, called Unimate, in 1958, attributes his long-standing fascination with robots to his reading of [Asimov's] 'I, Robot' when he was a teenager", and Engelberger later invited Asimov to write the foreword to his robotics manual.

The laws are simple and straightforward, and they embrace "the essential guiding principles of a good many of the world's ethical systems"7. They also appear to ensure the continued dominion of humans over robots, and to preclude the use of robots for evil purposes. In practice, however - meaning in Asimov's numerous and highly imaginative stories - a variety of difficulties arise.

My purpose here is to determine whether or not Asimov's fiction vindicates the laws he expounded. Does he successfully demonstrate that robotic technology can be applied in a responsible manner to potentially powerful, semi-autonomous and, in some sense intelligent machines? To reach a conclusion, we must examine many issues emerging from Asimov's fiction.


The robot notion derives from two strands of thought, humanoids and automata. The notion of a humanoid (or human- like nonhuman) dates back to Pandora in The Iliad, 2,500 years ago and even further. Egyptian, Babylonian, and ultimately Sumerian legends fully 5,000 years old reflect the widespread image of the creation, with god- men breathing life into clay models. One variation on the theme is the idea of the golem, associated with the Prague ghetto of the sixteenth century. This clay model, when breathed into life, became a useful but destructive ally.

The golem was an important precursor to Mary Shelley's Frankenstein: The Modern Prometheus (1818). This story combined the notion of the humanoid with the dangers of science (as suggested by the myth of Prometheus, who stole fire from the gods to give it to mortals). In addition to establishing a literary tradition and the genre of horror stories, Frankenstein also imbued humanoids with an aura of ill fate.

Automata, the second strand of thought, are literally "self- moving things" and have long interested mankind. Early models depended on levers and wheels, or on hydraulics. Clockwork technology enabled significant advances after the thirteenth century, and later steam and electro- mechanics were also applied. The primary purpose of automata was entertainment rather than employment as useful artifacts. Although many patterns were used, the human form always excited the greatest fascination. During the twentieth century, several new technologies moved automata into the utilitarian realm. Geduld and Gottesman8 and Frude2 review the chronology of clay model, water clock, golem, homunculus, android, and cyborg that culminated in the contemporary concept of the robot.

The term robot derives from the Czech word robota, meaning forced work or compulsory service, or robotnik, meaning serf. It was first used by the Czech playwright Karel Çapek in 1918 in a short story and again in his 1921 play R. U. R., which stood for Rossum's Universal Robots. Rossum, a fictional Englishman, used biological methods to invent and mass- produce "men" to serve humans. Eventually they rebelled, became the dominant race, and wiped out humanity. The play was soon well known in English- speaking countries.


Undeterred by its somewhat chilling origins (or perhaps ignorant of them), technologists of the 1950s appropriated the term robot to refer to machines controlled by programs. A robot is "a reprogrammable multifunctional device designed to manipulate and/or transport material through variable programmed motions for the performance of a variety of tasks"9. The term robotics, which Asimov claims he coined in 194210 refers to "a science or art involving both artificial intelligence (to reason) and mechanical engineering (to perform physical acts suggested by reason)"11.

As currently defined, robots exhibit three key elements:

* programmability, implying computational or symbol- manipulative capabilities that a designer can combine as desired (a robot is a computer);
* mechanical capability, enabling it to act on its environment rather than merely function as a data processing or computational device (a robot is a machine); and
* flexibility, in that it can operate using a range of programs and manipulate and transport materials in a variety of ways.

We can conceive of a robot, therefore. as either a computer- enhanced machine or as a computer with sophisticated input/output devices. Its computing capabilities enable it to use its motor devices to respond to external stimuli, which it detects with its sensory devices. The responses are more complex than would be possible using mechanical, electromechanical, and/or electronic components alone.

With the merging of computers, telecommunications networks, robotics, and distributed systems software. and the multiorganizational application of the hybrid technology, the distinction between computers and robots may become increasingly arbitrary. In some cases it would be more convenient to conceive of a principal intelligence with dispersed sensors and effectors, each with subsidiary intelligence (a robotics- enhanced computer system). In others, it would be more realistic to think in terms of multiple devices, each with appropriate sensory, processing, and motor capabilities, all subjected to some form of coordination (an integrated multi-robot system). The key difference robotics brings is the complexity and persistence that artifact behaviour achieves, independent of human involvement.

Many industrial robots resemble humans in some ways. In science fiction, the tendency has been even more pronounced, and readers encounter humanoid robots, humaniform robots, and androids. In fiction, as in life, it appears that a robot needs to exhibit only a few human- like characteristics to be treated as if it were human. For example, the relationships between humans and robots in many of Asimov's stories seem almost intimate, and audiences worldwide reacted warmly to the "personality" of the computer HAL in 2001: A Space Odyssey, and to the gibbering rubbish- bin R2- D2 in the Star Wars series.

The tendency to conceive of robots in humankind's own image may gradually yield to utilitarian considerations, since artifacts can be readily designed to transcend humans' puny sensory and motor capabilities. Frequently the disadvantages and risks involved in incorporating sensory, processing, and motor apparatus within a single housing clearly outweigh the advantages. Many robots will therefore be anything but humanoid in form. They may increasingly comprise powerful processing capabilities and associated memories in a safe and stable location, communicating with one or more sensory and motor devices (supported by limited computing capabilities and memory) at or near the location(s) where the robot performs its functions. Science fiction literature describes such architectures.12,13


Robotics offers benefits such as high reliability, accuracy, and speed of operation. Low long- term costs of computerized machines may result in significantly higher productivity, particularly in work involving variability within a general pattern. Humans can be relieved of mundane work and exposure to dangerous workplaces. Their capabilities can be extended into hostile environments involving high pressure (deep water), low pressure (space), high temperatures (furnaces), low temperatures (ice caps and cryogenics), and high- radiation areas (near nuclear materials or occurring naturally in space).

On the other hand, deleterious consequences are possible. Robots might directly or indirectly harm humans or their property; or the damage may be economic or incorporeal (for example, to a person's reputation). The harm could be accidental or result from human instructions. Indirect harm may occur to workers, since the application of robots generally results in job redefinition and sometimes in outright job displacement. Moreover, the replacement of humans by machines may undermine the self- respect of those affected, and perhaps of people generally.

During the 1980s, the scope of information technology applications and their impact on people increased dramatically. Control systems for chemical processes and air conditioning are examples of systems that already act directly and powerfully on their environments. And consider computer- integrated manufacturing, just- in- time logistics, and automated warehousing systems. Even data processing systems have become integrated into organizations' operations and constrain the ability of operations- level staff to query a machine's decisions and conclusions. In short, many modern computer systems are arguably robotic in nature already; their impact must be managed - now.

Asimov's original laws (see above) provide that robots are to be slaves to humans (the second law). However, this role is overridden by the higher-order first law, which precludes robots from injuring a human, either by their own autonomous action or by following a human's instructions. This precludes their continuing with a programmed activity when doing so would result in human injury. It also prevents their being used as a tool or accomplice in battery, murder, self- mutilation, or suicide.

The third and lowest level law creates a robotic survival instinct. This ensures that, in the absence of conflict with a higher order law, a robot will

* seek to avoid its own destruction through natural causes or accident;
* defend itself against attack by another robot or robots; and
* defend itself against attack by any human or humans.

Being neither omniscient nor omnipotent, it may of course fail in its endeavors. Moreover, the first law ensures that the robotic survival instinct fails if self- defense would necessarily involve injury to any human. For robots to successfully defend themselves against humans, they would have to be provided with sufficient speed and dexterity so as not to impose injurious force on a human.

Under the second law, a robot appears to be required to comply with a human order to (1) not resist being destroyed or dismantled, (2) cause itself to be destroyed, or (3) (within the limits of paradox) dismantle itself.1.2 In various stories, Asimov notes that the order to self- destruct does not have to be obeyed if obedience would result in harm to a human. In addition, a robot would generally not be precluded from seeking clarification of the order. In his last full- length novel, Asimov appears to go further by envisaging that court procedures would be generally necessary before a robot could be destroyed: "I believe you should be dismantled without delay. The case is too dangerous to await the slow majesty of the law. . . . If there are legal repercussions hereafter, I shall deal with them."14

Such apparent inconsistencies attest to the laws' primary role as a literary device intended to support a series of stories about robot behavior. In this, they were very successful: "There was just enough ambiguity in the Three Laws to provide the conflicts and uncertainties required for new stories, and, to my great relief, it seemed always to be possible to think up a new angle out of the 61 words of the Three Laws."1.

As Frude says, "The Laws have an interesting status. They . . . may easily be broken, just as the laws of a country may be transgressed. But Asimov's provision for building a representation of the Laws into the positronic- brain circuitry ensures that robots are physically prevented from contravening them."2 Because the laws are intrinsic to the machine's design, it should " never even enter into a robot's mind" to break them.

Subjecting the laws to analysis may seem unfair to Asimov. However, they have attained such a currency not only among sci- fi fans but also among practicing roboticists and software developers that they influence, if only subconsciously, the course of robotics.
Asimov's experiments with the 1940 laws

Asimov's early stories are examined here not in chronological sequence or on the basis of literary devices, but by looking at clusters of related ideas.

The ambiguity and cultural dependence of terms:

Any set of "machine values" provides enormous scope for linguistic ambiguity. A robot must be able to distinguish robots from humans. It must be able to recognize an order and distinguish it from a casual request. It must "understand" the concept of its own existence, a capability that arguably has eluded mankind, although it may be simpler for robots. In one short story, for example, the vagueness of the word firmly in the order "Pull [the bar] towards you firmly" jeopardizes a vital hyperspace experiment. Because robot strength is much greater than that of humans, it pulls the bar more powerfully than the human had intended, bends it, and thereby ruins the control mechanism15.

Defining injury and harm is particularly problematic, as are the distinctions between death, mortal danger, and injury or harm that is not life-threatening. Beyond this there are psychological harm. Any robot given, or developing, an awareness of human feelings would have to evaluate injury and harm in psychological as well as physical terms: "The insurmountable First Law of Robotics states: ' A robot may not injure a human being....' and to repel a friendly gesture would do injury " 16 (emphasis added). Asimov investigated this in an early short story and later in a novel: A mind-reading robot interprets the first law as requiring him to give people not the correct answers to their questions but the answers that he knows they want to hear 14,16,17.

Another critical question is how a robot is to interpret the term human. A robot could be given any number of subtly different descriptions of a human being, based for example on skin color, height range, and/or voice characteristics such as accent. it is therefore possible for robot behaviour to be manipulated: "the Laws, even the First Law, might not be absolute then, but might be whatever those who design robots define them to be"14. Faced with this difficulty, the robots in this story conclude that ..." if different robots are subject to narrow definitions of one sort or another, there can only be measureless destruction. we define human beings as all members of the species, Homo sapiens."14

In an early story, Asimov has a humanoid robot to represent itself as a human and stand for public office. It must prevent the public from realizing that it is a robot, since public reaction would not only result in its losing the election but also in tighter constraints on other robots. A political opponent, seeking to expose the robot, discovers that it is impossible to prove it is a robot solely on the basis of its behavior, because the Laws of Robotics force any robot to perform in essentially the same manner as a good human being7.

In a later novel, a roboticist says, "If a robot is human enough, he would be accepted as a human. Do you demand proof that I am a robot? The fact that I seem human is enough"16. In another scene, a humaniform robot is sufficiently similar to a human to confuse a normal robot and slow down its reaction time14. Ultimately, two advanced robots recognize each other as "human", at least for the purposes of the laws14,18.

Defining human beings becomes more difficult with the emergence of cyborgs, which may be seen as either machine-enhanced humans or biologically enhanced machines. When a human is augmented by prostheses (artificial limbs, heart pacemakers, renal dialysis machines, artificial lungs, and someday perhaps many other devices), does the notion of a human gradually blur with that of a robot? And does a robot that attains increasingly human characteristics (for example, a knowledge-based system provided with the "know-that" and "know-how" of a human expert and the ability to learn more about a domain) gradually become confused with a human? How would a robot interpret the first and second laws once the Türing test criteria can be routinely satisfied? The key outcome of the most important of Asimov's robot novellas 12 is the tenability of the argument that the prosthetization of humans leads inevitably to the humanization of robots.

The cultural dependence of meaning reflects human differences in such matters as religion, nationality, and social status. As robots become more capable, however, cultural differences between humans and robots might also be a factor. For example, in one story19 a human suggests that some laws may be bad and their enforcement unjust, but the robot replies that an unjust law is a contradiction in terms. When the human refers to something higher than justice, for example, mercy and forgiveness, the robot merely responds. "I am not acquainted with those words".

The role of judgment in decision making:

The assumption that there is a literal meaning for any given series of signals is currently considered naive. Typically, the meaning of a term is seen to depend not only on the context in which it was originally expressed but also on the context in which it is read (see, for example, Winograd and Flores20). If this is so, then robots must exercise judgment to interpret the meanings of words and hence of orders and of new data.

A robot must even determine whether and to what extent the laws apply to a particular situation. Often in the robot stories a robot action of any kind is impossible without some degree of risk to a human. To be at all useful to its human masters, a robot must therefore be able to judge how much the laws can be breached to maintain a tolerable level of risk. for example, in Asimov's very first robot short story, "Robbie [the robot] snatched up Gloria [his young human owner], slackening his speed not one iota, and, consequently knocking every breath of air out of her."21 Robbie judged that it was less harmful for Gloria to be momentarily breathless than to be mown down by a tractor.

Similarly, conflicting orders may have to be prioritized, for example, when two humans give inconsistent instructions. Whether the conflict is overt, unintentional, or even unwitting, it nonetheless requires a resolution. Even in the absence of conflicting orders, a robot may need to recognize foolish or illegal orders and decline to implement them, or at least question them. One story asks, "Must a robot follow the orders of a child; or of an idiot; or of a criminal; or of a perfectly decent intelligent man who happens to be inexpert and therefore ignorant of the undesirable consequences of his order?"18

Numerous problems surround the valuation of individual humans. First, do all humans have equal standing in a robot's evaluation? On the one hand they do: "A robot may not judge whether a human being deserves death. It is not for him to decide. He may not harm a human - variety skunk or variety angel."7 On the other hand they might not, as when a robot tells a human, "In conflict between your safety and that of another, I must guard yours."22 In another short story, robots agree that they "must obey a human being who is fit by mind, character, and knowledge to give me that order." Ultimately, this leads the robot to "disregard shape and form in judging between human beings" and to recognize his companion robot not merely as human but as a human "more fit than the others."18 Many subtle problems can be constructed. For example. a person might try forcing a robot to comply with an instruction to harm a human (and thereby violate the first law) by threatening to kill himself unless the robot obeys.

How is a robot to judge the trade- off between a high probability of lesser harm to one person versus a low probability of more serious harm to another? Asimov's stories refer to this issue but are somewhat inconsistent with each other and with the strict wording of the first law.

More serious difficulties arise in relation to the valuation of multiple humans. The first law does not even contemplate the simple case of a single terrorist threatening many lives. In a variety of stories, however, Asimov interprets the law to recognize circumstances in which a robot may have to injure or even kill one or more humans to protect one or more others: "The Machine cannot harm a human being more than minimally, and that only to save a greater number" 23 (emphasis added). And again: "The First Law is not absolute. What if harming a human being saves the lives of two others, or three others, or even three billion others? The robot may have thought that saving the Federation took precedence over the saving of one life."24

These passages value humans exclusively on the basis of numbers. A later story includes this justification: "To expect robots to make judgments of fine points such as talent, intelligence, the general usefulness to society, has always seemed impractical. That would delay decision to the point where the robot is effectively immobilized. So we go by numbers."18

A robot's cognitive powers might be sufficient for distinguishing between attacker and attackee, but the first law alone does not provide a robot with the means to distinguish between a "good" person and a "bad" one. Hence, a robot may have to constrain a "good" attackee's self- defense to protect the "bad" attacker from harm. Similarly, disciplining children and prisoners may be difficult under the laws, which would limit robots' usefulness for supervision within nurseries and penal institutions.22 Only after many generations of self- development does a humanoid robot learn to reason that "what seemed like cruelty [to a human] might, in the long run, be kindness."12

The more subtle life- and- death cases, such as assistance in the voluntary euthanasia of a fatally ill or injured person to gain immediate access to organs that would save several other lives, might fall well outside a robot's appreciation. Thus, the first law would require a robot to protect the threatened human, unless it was able to judge the steps taken to be the least harmful strategy. The practical solution to such difficult moral questions would be to keep robots out of the operating theater.22

The problem underlying all of these issues is that most probabilities used as input to normative decision models are not objective; rather, they are estimates of probability based on human (or robot) judgment. The extent to which judgment is central to robotic behavior is summed up in the cynical rephrasing of the first law by the major (human) character in the four novels: "A robot must not hurt a human being, unless he can think of a way to prove it is for the human being's ultimate good after all."19

The sheer complexity:

To cope with the judgmental element in robot decision making, Asimov's later novels introduced a further complication: "On......[worlds other than Earth], . . . the Third Law is distinctly stronger in comparison to the Second Law. . . . An order for self- destruction would be questioned and there would have to be a truly legitimate reason for it to be carried through - a clear and present danger."16 And again, "Harm through an active deed outweighs, in general, harm through passivity - all things being reasonably equal. . . . [A robot is] always to choose truth over nontruth, if the harm is roughly equal in both directions. In general, that is."16

The laws are not absolutes, and their force varies with the individual machine's programming, the circumstances, the robot's previous instructions, and its experience. To cope with the inevitable logical complexities, a human would require not only a predisposition to rigorous reasoning, and a considerable education, but also a great deal of concentration and composure. (Alternatively, of course, the human may find it easier to defer to a robot suitably equipped for fuzzy- reasoning- based judgment.)

The strategies as well as the environmental variables involve complexity. "You must not think . . . that robotic response is a simple yes or no, up or down, in or out. ... There is the matter of speed of response."16 In some cases (for example, when a human must be physically restrained), the degree of strength to be applied must also be chosen.

The scope for dilemma and deadlock:

A deadlock problem was the key feature of the short story in which Asimov first introduced the laws. He constructed the type of stand- off commonly referred to as the "Buridan's ass" problem. It involved a balance between a strong third- law self- protection tendency, causing the robot to try to avoid a source of danger, and a weak second- law order to approach that danger. "The conflict between the various rules is [meant to be] ironed out by the different positronic potentials in the brain," but in this case the robot "follows a circle around [the source of danger], staying on the locus of all points of ... equilibrium."5

Deadlock is also possible within a single law. An example under the first law would be two humans threatened with equal danger and the robot unable to contrive a strategy to protect one without sacrificing the other. Under the second law, two humans might give contradictory orders of equivalent force. The later novels address this question with greater sophistication.

What was troubling the robot was what roboticists called an equipotential of contradiction on the second level. Obedience was the Second Law and [the robot] was suffering from two roughly equal and contradictory orders. Robot- block was what the general population called it or, more frequently, roblock for short . . . [or] `mental freeze- out.' No matter how subtle and intricate a brain might be, there is always some way of setting up a contradiction. This is a fundamental truth of mathematics.16

Clearly, robots subject to such laws need to be programmed to recognize deadlock and either choose arbitrarily among the alternative strategies or arbitrarily modify an arbitrarily chosen strategy variable (say, move a short distance in any direction) and reevaluate the situation: "If A and not- A are precisely equal misery- producers according to his judgment, he chooses one or the other in a completely unpredictable way and then follows that unquestioningly. He does not go into mental freeze- out."16

The finite time that even robot decision making requires could cause another type of deadlock. Should a robot act immediately, by "instinct," to protect a human in danger? Or should it pause long enough to more carefully analyze available data - or collect more data - perhaps thereby discovering a better solution, or detecting that other humans are in even greater danger? Such situations can be approached using the techniques of information economics, but there is inherent scope for ineffectiveness and deadlock, colloquially referred to as "paralysis by analysis."

Asimov suggested one class of deadlock that would not occur: If in a given situation a robot knew that it was powerless to prevent harm to a human, then the first law would be inoperative; the third law would become relevant, and it would not self- immolate in a vain attempt to save the human.25 It does seem, however, that the deadlock is not avoided by the laws themselves, but rather by the presumed sophistication of the robot's decision- analytical capabilities.

A special case of deadlock arises when a robot is ordered to wait. For example, "[Robot] you will not move nor speak nor hear us until I say your name again.' There was no answer. The robot sat as though it were cast out of one piece of metal, and it would stay so until it heard its name again."26 As written, the passage raises the intriguing question of whether passive hearing is possible without active listening. What if the robot's name is next used in the third person rather than the second?

In interpreting a command such as "Do absolutely nothing until I call you!" a human would use common sense and, for example, attend to bodily functions in the meantime. A human would do nothing about the relevant matter until the event occurred. In addition, a human would recognize additional terminating events, such as a change in circumstances that make it impossible for the event to ever occur. A robot is likely to be constrained to a more literal interpretation, and unless it can infer a scope delimitation to the command, it would need to place the majority of its functions in abeyance

The faculties that would need to remain in operation are the:

* sensory- perceptive subsystem needed to detect the condition;
* the recommencement triggering function;
* one or more daemons to provide a time- out mechanism (presumably the scope of the command is at least restricted to the expected remaining lifetime of the person who gave the command); and
* ability to play back the audit trail so that an overseer can discover the condition on which the robot's resuscitation depends.

Asimov does not appear to have investigated whether the behavior of a robot in wait-mode is affected by the laws. If it isn't, then it will not only fail to protect its own existence and to obey an order, but will also stand by and allow a human to be harmed. A robotic security guard could therefore be nullified by an attacker's simply putting it into a wait-state.

Audit of robot compliance:

For a fiction writer, it is sufficient to have the laws embedded in robots' positronic pathways (whatever they may be). To actually apply such a set of laws in robot design, however, it would be necessary to ensure that every robot:

* had the laws imposed in precisely the manner intended; and
* was at all times subject to them - that is, they could not be overridden or modified.

It is important to know how malprogramming and modification of the laws' implementation in a robot (whether intentional or unintentional) can he prevented, detected, and dealt with.

In an early short story, robots were "rescuing" humans whose work required short periods of relatively harmless exposure to gamma radiation. Officials obtained robots with the first law modified so that they were incapable of injuring a human but under no compulsion to prevent one from coming to harm. This clearly undermined the remaining part of the first law, since, for example, a robot could drop a heavy weight toward a human, knowing that it would be fast enough and strong enough to catch it before it harmed the person. However, once gravity had taken over, the robot would be free to ignore the danger.25 Thus, a partial implementation was shown to be risky, and the importance of robot audit underlined. Other risks include trapdoors, Trojan horses, and similar devices in the robot's programming.

A further imponderable is the effect of hostile environments and stress on the reliability and robustness of robots' performance in accordance with the laws. In one short story, it transpires that "The Machine That Won the War" had been receiving only limited and poor- quality data as a result of enemy action against its receptors and had been processing it unreliably because of a shortage of experienced maintenance staff. Each of the responsible managers had, in the interests of national morale, suppressed that information, even from one another, and had separately and independently "introduced a number of necessary biases" and "adjusted" the processing parameters in accordance with intuition. The executive director, even though unaware of the adjustments, had placed little reliance on the machine's output, preferring to carry out his responsibility to mankind by exercising his own judgment.27

A major issue in military applications generally28 is the impossibility of contriving effective compliance tests for complex systems subject to hostile and competitive environments. Asimov points out that the difficulties of assuring compliance will be compounded by the design and manufacture of robots by other robots.22

Robot autonomy:

Sometimes humans may delegate control to a robot and find themselves unable to regain it, at least in a particular context. One reason is that to avoid deadlock, a robot must be capable of making arbitrary decisions. Another is that the laws embody an explicit ability for a robot to disobey an instruction, by virtue of the overriding first law.

In an early Asimov short story, a robot "knows he can keep [the energy beam] more stable than we [humans] can, since he insists he's the superior being, so he must keep us out of the control room [in accordance with the first law]."29 The same scenario forms the basis of one of the most vivid episodes in science fiction, HAL's attempt to wrest control of the spacecraft from Bowman in 2001: A Space Odyssey. Robot autonomy is also reflected in a lighter moment in one of Asimov's later novels, when a character says to his companion, "For now I must leave you. The ship is coasting in for a landing, and I must stare intelligently at the computer that controls it, or no one will believe I am the captain."14

In extreme cases, robot behavior will involve subterfuge, as the machine determines that the human, for his or her own protection, must be tricked. In another early short story, the machines that manage Earth's economy implement a form of "artificial stupidity" by making intentional errors, thereby encouraging humans to believe that the robots are fallible and that humans still have a role to play.23

Scope for adaptation:

The normal pattern of any technology is that successive generations show increased sophistication, and it seems inconceivable that robotic technology would quickly reach a plateau and require little further development. Thus there will always be many old models in existence, models that may have inherent technical weaknesses resulting in occasional malfunctions and hence infringement on the Laws of Robotics. Asimov's short stories emphasize that robots are leased from the manufacturer, never sold, so that old models can be withdrawn after a maximum of 25 years.

Looking at the first 50 years of software maintenance, it seems clear that successive modification of existing software to perform new or enhanced functions is one or more orders of magnitude harder than creating a new artifact to perform the same function. Doubts must exist about the ability of humans (or robots) to reliably adapt existing robots. The alternative - destruction of existing robots - will be resisted in accordance with the third law, robot self- preservation.

At a more abstract level, the laws are arguably incomplete because the frame of reference is explicitly human. No recognition is given to plants, animals, or as- yet- undiscovered (for example, extraterrestrial), intelligent life forms. Moreover, some future human cultures may place great value on inanimate creation, or on holism. If, however, late twentieth- century values have meanwhile been embedded in robots, that future culture may have difficulty wresting the right to change the values of the robots it has inherited. If machines are to have value sets, there must be a mechanism for adaptation, at least through human- imposed change. The difficulty is that most such value sets will be implicit rather than explicit; their effects will be scattered across a system rather than implemented in a modular and therefore replaceable manner.

At first sight, Asimov's laws are intuitively appealing, but their application encounters difficulties. Asimov, in his fiction, detected and investigated the laws' weaknesses, which this article (Part 1 of 2) has analyzed and classified. Part 2, in the next issue of Computer, will take the analysis further by considering the effects of Asimov's 1985 revision to the laws. It will then examine the extent to which the weaknesses in these laws may in fact be endemic to any set of laws regulating robotic behavior.

Part 2:


Isaac Asimov's Laws of Robotics, first formulated in 1940, were primarily a literary device intended to support a series of stories about robot behavior. Over time, he found that the three laws included enough apparent inconsistencies, ambiguity, and uncertainty to provide the conflicts required for a great many stories. In examining the ramifications of these laws, Asimov revealed problems that might later confront real roboticists and information technologists attempting to establish rules for the behavior of intelligent machines.

With their fictional "positronic" brains imprinted with the mandate to (in order of priority) prevent harm to humans, obey their human masters, and protect themselves, Asimov's robots had to deal with great complexity. In a given situation, a robot might be unable to satisfy the demands of two equally powerful mandates and go into "mental freezeout." Semantics is also a problem. As demonstrated in Part 1 of this article (Computer, December 1993, pp. 53- 61), language is much more than a set of literal meanings and Asimov showed us that a machine trying to distinguish, for example, who or what is human may encounter many difficulties that humans themselves handle easily and intuitively. Thus, robots must have sufficient capabilities for judgment - capabilities that can cause them to frustrate the intentions of their masters when, in a robot's judgment, a higher order law applies.

As information technology evolves and machines begin to design and build other machines, the issue of human control gains greater significance. In time. human values tend to change; the rules reflecting these values, and embedded in existing robotic devices. may need to be modified. But if they are implicit rather than explicit, with their effects scattered widely across a system, they may not be easily replaceable. Asimov himself discovered many contradictions and eventually revised the Laws of Robotics.

Asimov's 1985 revised Laws of Robotics:

The Zeroth law:

After introducing the original three laws, Asimov detected. as early as 1950, a need to extend the first law, which protected individual humans, so that it would protect humanity as a whole. Thus, his calculating machines "have the good of humanity at heart through the overwhelming force of the First Law of Robotics"1 (emphasis added). In 1985 he developed this idea further by postulating a "zeroth" law that placed humanity's interests above those of any individual while retaining a high value on individual human life.2 The revised set of laws is shown in the sidebar.

Asimov pointed out that under a strict interpretation of the first law, a robot would protect a person even if the survival of humanity as a whole was placed at risk. Possible threats include annihilation by an alien or mutant human race, or by a deadly virus. Even when a robot's own powers of reasoning led it to conclude that mankind as a whole was doomed if it refused to act, it was nevertheless constrained: "I sense the oncoming of catastrophe . . . [but] I can only follow the Laws."2

In Asimov's fiction the robots are tested by circumstances and must seriously consider whether they can harm a human to save humanity. The turning point comes when the robots appreciate that the laws are indirectly modifiable by roboticists through the definitions programmed into each robot: "If the Laws of Robotics, even the First Law, are not absolutes, and if human beings can modify them, might it not be that perhaps, under proper conditions, we ourselves might modify."2 Although the robots are prevented by imminent "roblock" (robot block, or deadlock) from even completing the sentence, the groundwork has been laid.

Later, when a robot perceives a clear and urgent threat to mankind, it concludes, "Humanity as a whole is more important than a single human being. There is a law that is greater than the First Law: `A robot may not injure humanity, or through inaction, allow humanity to come to harm."2

Defining "humanity":

Modification of the laws, however, leads to additional considerations. Robots are increasingly required to deal with abstractions and philosophical issues. For example, the concept of humanity may be interpreted in different ways. It may refer to the set of individual human beings (a collective), or it may be a distinct concept (a generality, as in the notion of "the State"). Asimov invokes both ideas by referring to a tapestry (a generality) made up of individual contributions (a collective): "An individual life is one thread in the tapestry, and what is one thread compared to the whole?

.....Keep your mind fixed firmly on the tapestry and do not let the trailing off of a single thread affect you."2

A human roboticist raised a difficulty with the zeroth law immediately after the robot formulated it: "What is your `humanity' but an abstraction'? Can you point to humanity? You can injure or fail to injure a specific human being and understand the injury or lack of injury that has taken place. Can you see the injury to humanity? Can you understand it? Can you point to it?"2 The robot later responds by positing an ability to "detect the hum of the mental activity of Earth's human population, overall. . . . And, extending that, can one not imagine that in the Galaxy generally there is the hum of the mental activity of all of humanity? How, then, is humanity an abstraction? It is something you can point to." Perhaps as Asimov's robots learn to reason with abstract concepts, they will inevitably become adept at sophistry and polemic.

The increased difficulty of judgment:

One of Asimov's robot characters also points out the increasing complexity of the laws: "The First Law deals with specific individuals and certainties. Your Zeroth Law deals with vague groups and probabilities."2 At this point, as he often does, Asimov resorts to poetic license and for the moment pretends that coping with harm to individuals does not involve probabilities. However, the key point is not affected: Estimating probabilities in relation to groups of humans is far more difficult than with individual humans.

It is difficult enough, when one must choose quickly . . . ,to decide which indivi dual may suffer, or inflict, the greater harm. To choose between an individual and humanity, when you are not sure of what aspect of humanity you are dealing with, is so difficult that the very validity of Robotic Laws comes to be suspect. As soon as humanity in the abstract is introduced, the Laws of Robotics begin to merge with the Laws of Humanics which may not even exist.2

Robot paternalism:

Despite these difficulties, the robots agree to implement the zeroth law, since they judge themselves more capable than anyone else of dealing with the problems. The original laws produced robots with considerable autonomy, albeit a qualified autonomy allowed by humans. But under the 1985 laws, robots were more likely to adopt a superordinate, paternalistic attitude toward humans.

Asimov's Revised Laws of Robotics (1985):

Zeroth Law:

A robot may not injure humanity, or, through inaction, allow humanity to come to harm.

First Law:

A robot may not injure a human being, or, through inaction, allow a human being to come to harm, unless this would violate the Zeroth Law of Robotics.

Second Law:

A robot must obey orders given it by human beings, except where such orders would conflict with the Zeroth or First Law.

Third Law:

A robot must protect its own existence as long as such protection does not conflict with the Zeroth, First, or Second Law.

Asimov suggested this when he first hinted at the zeroth law, because he had his chief robotpsychologist say that "...we can no longer understand our own creations. . . . [Robots] have progressed beyond the possibility of detailed human control."1 In a more recent novella, a robot proposes to treat his form "as a canvas on which I intend to draw a man." but is told by the roboticist, "It's a puny ambition. ... You're better than a man. You've gone downhill from the moment you opted for organicism."3

In the later novels, a robot with telepathic powers manipulates humans to act in a way that will solve problems,4 although its powers are constrained by the psychological dangers of mind manipulation. Naturally, humans would be alarmed by the very idea of a mind- reading robot; therefore, under the zeroth and first laws, such a robot would be permitted to manipulate the minds of humans who learned of its abilities, making them forget the knowledge, so that they could not be harmed by it. This is reminiscent of an Asimov story in which mankind is an experimental laboratory for higher beings5 and Adams' altogether more flippant Hitchhiker's Guide to the Galaxy, in which the Earth is revealed as a large experiment in which humans are being used as laboratory animals by, of all things, white mice.6 Someday those manipulators of humans might be robots.

Asimov's The Robots of Dawn is essentially about humans, with robots as important players. In the sequel Robots and Empire, however, the story is dominated by the two robots, and the humans seem more like their playthings. It comes as little surprise, then, that the robots eventually conclude that "it is not sufficient to be able to choose [among alternative humans or classes of human] . . . ; we must be able to shape."2 Clearly, any subsequent novels in the series would have been about robots, with humans playing "bit" parts.

Robot dominance has a corollary that pervades the novels: History "grew less interesting as it went along; it became almost soporific."4 With life's challenges removed, humanity naturally regresses into peace and quietude, becoming "placid, comfortable, and unmoving" - and stagnant.

So who's in charge?

As we have seen, the term human can be variously defined, thus significantly affecting the first law. The term humanity did not appear in the original laws, only in the zeroth law, which Asimov had formulated and enunciated by a robot.2 Thus, the robots define human and humanity to refer to themselves as well as to humans, and ultimately to themselves alone. Another of the great science fiction stories, Clarke's Rendezvous with Rama,7 also assumes that an alien civilization, much older than mankind, would consist of robots alone (although in this case Clarke envisioned biological robots). Asimov's vision of a robot takeover differs from those of previous authors only in that force would be unnecessary.

Asimov does not propose that the zeroth law must inevitably result in the ceding of species dominance by humans to robots. However, some concepts may be so central to humanness that any attempt to embody them in computer processing might undermine the ability of humanity to control its own fate. Weizenbaum argues this point more fully.8

The issues discussed here, and in Part 1, have grown increasingly speculative, and some are more readily associated with metaphysics than with contemporary applications of information technology. However, they demonstrate that even an intuitively attractive extension to the original laws could have very significant ramifications. Some of the weaknesses are probably inherent in any set of laws and hence in any robotic control regime.

Asimov's laws extended:

The behavior of robots in Asimov's stories is not satisfactorily explained by the laws he enunciated. This section examines the design requirements necessary to effectively subject robotic behavior to the laws. In so doing. it becomes necessary to postulate several additional laws implicit in Asimov's fiction.

Perceptual and cognitive apparatus:

Clearly, robot design must include sophisticated sensory capabilities. However, more than signal reception is needed. Many of the difficulties Asimov dramatized arose because robots were less than omniscient. Would humans, knowing that robots cognitive capabilities are limited, be prepared to trust their judgment on life- and- death matters? For example, the fact that any single robot cannot harm a human does not protect humans from being injured or killed by robotic actions. In one story, a human tells a robot to add a chemical to a glass of milk and then tells another robot to serve the milk to a human. The result is murder by poisoning. Similarly, a robot untrained in first aid might move an accident victim and break the person's spinal cord. A human character in The Naked Sun is so incensed by these shortcomings that he accuses roboticists of perpetrating a fraud on mankind by omitting key words from the first law. In effect, it really means "A robot may do nothing that to its knowledge would injure a human being, and may not, through inaction, knowingly allow a human being to come to harm."9

Robotic architecture must be designed so that the laws can effectively control a robot's behavior. A robot requires a basic grammar and vocabulary to "understand" the laws and converse with humans. In one short story, a production accident results in a "mentally retarded" robot. This robot, defending itself against a feigned attack by a human, breaks its assailant's arm. This was not a breach of the first law, because it did not knowingly injure the human: "In brushing aside the threatening arm . . . it could not know the bone would break. In human terms, no moral blame can be attached to an individual who honestly cannot differentiate good and evil."10 In Asimov's stories, instructions sometimes must be phrased carefully to be interpreted as mandatory. Thus, some authors have considered extensions to the apparatus of robots, for example, a "button labeled `Implement Order' on the robot's chest,"11 analogous to the Enter key on a computer's keyboard.

A set of laws for robotics cannot be independent but must be conceived as part of a system. A robot must also he endowed with data collection, decision- analytical, and action processes by which it can apply the laws. Inadequate sensory, perceptual, or cognitive faculties would undermine the laws' effectiveness.

Additional implicit laws:

In his first robot short story, Asimov stated that "long before enough can go wrong to alter that First Law, a robot would be completely inoperable. It's a mathematical impossibility [for Robbie the Robot to harm a human]."12 For this to be true, robot design would have to incorporate a high- order controller (a "conscience"?) that would cause a robot to detect any potential for noncompliance with the laws and report the problem or immobilize itself. The implementation of such a meta- law ("A robot may not act unless its actions are subject to the laws of robotics") might well strain both the technology and the underlying science. (Given the meta- language problem in twentieth- century philosophy, perhaps logic itself would be strained.) This difficulty highlights the simple fact that robotic behavior cannot be entirely automated; it is dependent on design and maintenance by an external agent.

Another of Asimov's requirements is that all robots must be subject to the laws at all times. Thus, it would have to be illegal for human manufacturers to create a robot that was not subject to the laws. In a future world that makes significant use of robots, their design and manufacture would naturally be undertaken by other robots. Therefore, the Laws of Robotics must include the stipulation that no robot may commit an act that could result in any robot's not being subject to the same laws.

The words "protect its own existence" raise a semantic difficulty. In The Bicentennial Man, Asimov has a robot achieve humanness by taking its own life. Van Vogt, however, wrote that "indoctrination against suicide" was considered a fundamental requirement.13 The solution might be to interpret the word protect as applying to all threats, or to amend the wording to explicitly preclude self- inflicted harm. Having to continually instruct robot slaves would be both inefficient and tiresome. Asimov hints at a further, deep- nested law that would compel robots to perform the tasks they were trained for:

Quite aside from the Three Laws, there isn't a pathway in those brains that isn't carefully designed and fixed. We have robots planned for specific tasks, implanted with specific capabilities.'14 (Emphasis added.)

So perhaps we can extrapolate an additional, lower priority law: "A robot must perform the duties for which it has been programmed, except where that would conflict with a higher order law." Asimov's laws regulate around robots' transactions with humans and thus apply where robots have relatively little to do with one another or where there is only one robot. However, the laws fail to address the management of large numbers of robots. In several stories, a robot is assigned to oversee other robots. This would be possible only if each of the lesser robots were instructed by a human to obey the orders of its robot overseer. That would create a number of logical and practical difficulties, such as the scope of the human's order. It would seem more effective to incorporate in all subordinate robots an additional law, for example, "A robot must obey the orders given it by superordinate robots except where such orders would conflict with a higher order law." Such a law would fall between the second and third laws.

Furthermore, subordinate robots should protect their superordinate robot. This could be implemented as an extension or corollary to the third law; that is, to protect itself, a robot would have to protect another robot on which it depends. Indeed, a subordinate robot may need to be capable of sacrificing itself to protect its robot overseer. Thus, an additional law superior to the third law but inferior to orders from either a human or a robot overseer seems appropriate: "A robot must protect the existence of a superordinate robot as long as such protection does not conflict with a higher order law."

The wording of such laws should allow for nesting, since robot overseers may report to higher level robots. It would also be necessary to determine the form of the superordinate relationships:

* a tree, in which each robot has precisely one immediate overseer, whether robot or human;
* a constrained network, in which each robot may have several overseers but restrictions determine who may act as an overseer; or
* an unconstrained network, in which each robot may have any number of other robots or persons as overseers.

This issue of a command structure is far from trivial, since it is central to democratic processes that no single entity shall have ultimate authority. Rather, the most senior entity in any decision- making hierarchy must be subject to review and override by some other entity, exemplified by the balance of power in the three branches of government and the authority of the ballot box. Successful, long- lived systems involve checks and balances in a lattice rather than a mere tree structure. Of course, the structures and processes of human organizations may prove inappropriate for robotic organization. In any case, additional laws of some kind would be essential to regulate relationships among robots.

The sidebar shows an extended set of laws, one that incorporates the additional laws postulated in this section. Even this set would not alway's ensure appropriate robotic behavior. However, it does reflect the implicit laws that emerge in Asimov's fiction while demonstrating that any realistic set of design principles would have to be considerably more complex than Asimov's 1940 or 1985 laws. This additional complexity would inevitably exacerbate the problems identified earlier in this article and create new ones.

An Extended Set of the Laws of Robotics:

The Meta-Law:

A robot may not act unless its actions are subject to the Laws of Robotics.

Law Zero:

A robot may not injure humanity, or, through inaction, allow humanity to come to harm.

Law One:

A robot may not injure a human being, or, through inaction, allow a human being to come to harm, unless this would violate a higher-order Law.

Law Two:

* A robot must obey orders given it by human beings, except where such orders would conflict with a higher-order Law
* A robot must obey orders given it by superordinate robots, except where such orders would conflict with a higher-order Law

Law Three:

* A robot must protect the existence of a superordinate robot as long as such protection does not conflict with a higher-order Law
* A robot must protect its own existence as long as such protection does not conflict with a higher-order Law

Law Four:

A robot must perform the duties for which it has been programmed, except where that would conflict with a higher-order law.

The Procreation Law:

A robot may not take any part in the design or manufacture of a robot unless the new robot's actions are subject to the Laws of Robotics.

While additional laws may be trivially simple to extract and formulate, the need for them serves as a warning. The 1940 laws' intuitive attractiveness and simplicity were progressively lost in complexity, legalisms, and semantic richness. Clearly then, formulating an actual set of laws as a basis for engineering design would result in similar difficulties and require a much more formal approach. Such laws would have to be based in ethics and human morality, not just in mathematics and engineering. Such a political process would probably result in a document couched in fuzzy generalities rather than constituting an operational- level, programmable specification.
Implications for information technologists

Many facets of Asimov's fiction are clearly inapplicable to real information technology or too far in the future to be relevant to contemporary applications. Some matters, however, deserve our consideration. For example, Asimov's fiction could help us assess the practicability of embedding some appropriate set of general laws into robotic designs. Alternatively, the substantive content of the laws could be used as a set of guidelines to be applied during the conception, design, development, testing, implementation, use, and maintenance of robotic systems. This section explores the second approach.
Recognition of stakeholder interests

The Laws of Robotics designate no particular class of humans (not even a robot's owner) as more deserving of protection or obedience than another. A human might establish such a relationship by command, but the laws give such a command no special status: another human could therefore countermand it. In short, the laws reflect the humanistic and egalitarian principles that theoretically underlie most democratic nations.

The laws therefore stand in stark contrast to our conventional notions about an information technology artifact, whose owner is implicitly assumed to be its primary beneficiary. An organization shapes an application's design and use for its own benefit. Admittedly, during the last decade users have been given greater consideration in terms of both the human- machine interface and participation in system development. But that trend has been justified by the better returns the organization can get from its information technology investment rather than by any recognition that users are stakeholders with a legitimate voice in decision making. The interests of other affected parties are even less likely to be reflected.

In this era of powerful information technology, professional bodies of information technologists need to consider:

* identification of stakeholders and how they are affected;
* prior consultation with stakeholders;
* quality assurance standards for design, manufacture, use, and maintenance;
* liability for harm resulting from either malfunction or use in conformance with the designer's intentions; and
* complaint- handling and dispute- resolution procedures.

Once any resulting standards reach a degree of maturity, legislatures in the many hundreds of legal jurisdictions throughout the world would probably have to devise enforcement procedures.

The interests of people affected by modern information technology applications have been gaining recognition. For example, consumer representatives are now being involved in the statement of user requirements and the establishment of the regulatory environment for consumer electronic- funds- transfer systems. This participation may extend to the logical design of such systems. Other examples are trade- union negotiations with employers regarding technology- enforced change, and the publication of software quality- assurance standards.

For large- scale applications of information technology, governments have been called upon to apply procedures like those commonly used in major industrial and social projects. Thus, commitment might have to be deferred pending dissemination and public discussion of independent environmental or social impact statements. Although organizations that use information technology might see this as interventionism, decision making and approval for major information technology applications may nevertheless become more widely representative.

Closed- system versus open- system thinking:

Computer- based systems no longer comprise independent machines each serving a single location. The marriage of computing with telecommunications has produced multicomponent systems designed to support all elements of a widely dispersed organization. Integration hasn't been simply geographic, however. The practice of information systems has matured since the early years when existing manual systems were automated largely without procedural change. Developers now seek payback via the rationalization of existing systems and varying degrees of integration among previously separate functions. With the advent of strategic and interorganizational systems, economies are being sought at the level of industry sectors, and functional integration increasingly occurs across corporate boundaries.

Although programmers can no longer regard the machine as an almost entirely closed system with tightly circumscribed sensory and motor capabilities, many habits of closed- system thinking remain. When systems have multiple components, linkages to other systems, and sophisticated sensory and motor capabilities, the scope needed for understanding and resolving problems is much broader than for a mere hardware/software machine. Human activities in particular must be perceived as part of the system. This applies to manual procedures within systems (such as reading dials on control panels), human activities on the fringes of systems (such as decision making based on computer- collated and - displayed information), and the security of the user's environment (automated teller machines, for example). The focus must broaden from mere technology to technology in use.

General systems thinking leads information technologists to recognize that relativity and change must he accommodated. Today, an artifact may be applied in multiple cultures where language, religion, laws, and customs differ. Over time, the original context may change. For example, models for a criminal justice system - one based on punishment and another based on redemption - may alternately dominate social thinking. Therefore, complex systems must be capable of adaptation.

Blind acceptance of technological and other imperatives:

Contemporary utilitarian society seldom challenges the presumption that what can be done should be done. Although this technological imperative is less pervasive than people generally think, societies nevertheless tend to follow where their technological capabilities lead. Related tendencies include the economic imperative (what can be done more efficiently should be) and the marketing imperative (any effective demand should be met). An additional tendency might be called the "information imperative," the dominance of administrative efficiency, information richness, and rational decision making. However, the collection of personal data has become so pervasive that citizens and employees have begun to object.

The greater a technology's potential to promote change, the more carefully a society should consider the desirability of each application. Complementary measures that may be needed to ameliorate its negative effects should also be considered. This is a major theme of Asimov's stories, as he explores the hidden effects of technology. The potential impact of information technology is so great that it would be inexcusable for professionals to succumb blindly to the economic, marketing, information, technological, and other imperatives. Application software professionals can no longer treat the implications of information technology as someone else's problem but must consider them as part of the project.15

Human acceptance of robots:

In Asimov's stories, humans develop affection for robots, particularly humaniform robots. In his very first short story, a little girl is too closely attached to Robbie the Robot for her parents' liking.'12 In another early story, a woman starved for affection from her husband and sensitively assisted by a humanoid robot to increase her self confidence entertains thoughts approaching love toward it/him.16

Nonhumaniforms, such as conventional industrial robots and large, highly dispersed robotic systems (such as warehouse managers. ATMs, and EFT/POS systems) seem less likely to elicit such warmth. Yet several studies have found a surprising degree of identification by humans with computers.17,18 Thus, some hitherto exclusively human characteristics are being associated with computer systems that don't even exhibit typical robotic capabilities.

Users must be continually reminded that the capabilities of hardware/software components are limited:

* they contain many inherent assumptions;
* they are not flexible enough to cope with all of the manifold exceptions that inevitably arise; and
* they do not adapt to changes in their environment;
* authority is not vested in hardware/ software components but rather in the individuals who use them.

Educational institutions and staff training programs must identify these limitations; yet even this is not sufficient: The human- machine interface must reflect them. Systems must be designed so that users are required to continually exercise their own expertise, and system output should not be phrased in a way that implies unwarranted authority. These objectives challenge the conventional outlook of system designers.

Human opposition to robots:

Robots are agents of change and therefore potentially upsetting to those with vested interests. Of all the machines so far invented or conceived of, robots represent the most direct challenge to humans. Vociferous and even violent campaigns against robotics should not be surprising. Beyond concerns of self interest is the possibility that some humans could be revulsed by robots, particularly those with humanoid characteristics. Some opponents may be mollified as robotic behavior becomes more tactful. Another tenable argument is that by creating and deploying artifacts that are in some ways superior. humans degrade themselves.

System designers must anticipate a variety of negative reactions against their creations from different groups of stakeholders. Much will depend on the number and power of the people who feel threatened - and on the scope of the change they anticipate. If, as Asimov speculates,9 a robot- based economy develops without equitable adjustments, the backlash could be considerable.

Such a rejection could involve powerful institutions as well as individuals. In one Asimov story, the US Department of Defense suppresses a project intended to produce the perfect robot- soldier. It reasons that the degree of discretion and autonomy needed for battlefield performance would tend to make robots rebellious in other circumstances (particularly during peace time) and unprepared to suffer their commanders' foolish decisions.19 At a more basic level, product lines and markets might be threatened, and hence the profits and even the survival of corporations. Although even very powerful cartels might not be able to impede robotics for very long, its development could nevertheless be delayed or altered. Information technologists need to recognize the negative perceptions of various stakeholders and manage both system design and project politics accordingly.

The structuredness of decision making:

For five decades there has been little doubt that computers hold significant computational advantages over humans. However, the merits of machine decision making remain in dispute. Some decision processes are highly structured and can be resolved using known algorithms operating on defined data items with defined interrelationships. Most structured decisions are candidates for automation, subject, of course, to economic constraints. The advantages of machines must also be balanced against risks. The choice to automate must be made carefully because the automated decision process (algorithm, problem description. problem- domain description, or analysis of empirical data) may later prove to be inappropriate for a particular type of decision. Also, humans involved as data providers, data communicators, or decision implementers may not perform rationally because of poor training, poor performance under pressure, or willfulness.

Unstructured decision making remains the preserve of humans for one or more of the following reasons:

* humans have not yet worked out a suitable way to program (or teach) a machine how to make that class of decision;
* some relevant data cannot be communicated to the machine;
* "fuzzy" or "open- textured" concepts or constructs are involved;
* such decisions involve judgments that system participants feel should not be made by machines on behalf of humans.

One important type of unstructured decision is problem diagnosis. As Asimov described the problem, "How..... can we send a robot to find a flaw in a mechanism when we cannot possibly give precise orders, since we know nothing about the flaw ourselves'? `Find out what's wrong' is not an order you can give to a robot; only to a man."20 Knowledge- based technology has since been applied to problem diagnosis, but Asimov's insight retains its validity: A problem may be linguistic rather than technical, requiring common sense, not domain knowledge. Elsewhere, Asimov calls robots "logical but not reasonable" and tells of household robots removing important evidence from a murder scene because a human did not think to order them to preserve it.9

The literature of decision support systems recognizes an intermediate case, semistructured decision making. Humans are assigned the decision task, and systems are designed to provide support for gathering and structuring potentially relevant data and for modeling and experimenting with alternative strategies. Through continual progress in science and technology, previously unstructured decisions are reduced to semistructured or structured decisions. The choice of which decisions to automate is therefore provisional, pending further advances in the relevant area of knowledge. Conversely, because of environmental or cultural change, structured decisions may not remain so. For example, a family of viruses might mutate so rapidly that the reference data within diagnostic support systems is outstripped and even the logic becomes dangerously inadequate.

Delegating to a machine any kind of decision that is less than fully structured invites errors and mishaps. Of course. human decision- makers routinely make mistakes too. One reason for humans' retaining responsibility for unstructured decision making is rational: Appropriately educated and trained humans may make more right decisions and/or fewer seriously wrong decisions than a machine. Using common sense, humans can recognize when conventional approaches and criteria do not apply, and they can introduce conscious value judgments. Perhaps a more important reason is the arational preference of humans to submit to the judgments of their peers rather than of machines: If someone is going to make a mistake costly to me, better for it to be an understandably incompetent human like myself than a mysteriously incompetent machine.8

Because robot and human capabilities differ, for the foreseeable future at least, each will have specific comparative advantages. Information technologists must delineate the relationship between robots and people by applying the concept of decision structuredness to blend computer- based and human elements advantageously. The goal should be to achieve complementary intelligence rather than to continue pursuing the chimera of unneeded artificial intelligence. As Wyndham put it in 1932: "Surely man and machine are natural complements: They assist one another."21

Risk management:

Whether or not subjected to intrinsic laws or design guidelines, robotics embodies risks to property as well as to humans. These risks must be managed; appropriate forms of risk avoidance and diminution need to be applied, and regimes for fallback, recovery, and retribution must be established.

Controls are needed to ensure that intrinsic laws, if any, are operational at all times and that guidelines for design, development, testing, use, and maintenance are applied. Second- order control mechanisms are needed to audit first- order control mechanisms. Furthermore, those bearing legal responsibility for harm arising from the use of robotics must be clearly identified. Courtroom litigation may determine the actual amount of liability, but assigning legal responsibilities in advance will ensure that participants take due care.

In most of Asimov's robot stories, robots are owned by the manufacturer even while in the possession of individual humans or corporations. Hence legal responsibility for harm arising from robot noncompliance with the laws can be assigned with relative ease. In most real- world jurisdictions, however, there are enormous uncertainties, substantial gaps in protective coverage, high costs, and long delays.

Each jurisdiction, consistent with its own product liability philosophy, needs to determine who should bear the various risks. The law must be sufficiently clear so that debilitating legal battles do not leave injured parties without recourse or sap the industry of its energy. Information technologists need to communicate to legislators the importance of revising and extending the laws that assign liability for harm arising from the use of information technology.

Enhancements to codes of ethics:

Associations of information technology professionals, such as the lEEE Computer Society, the Association for Computing Machinery, the British Computer Society, and the Australian Computer Society, are concerned with professional standards, and these standards almost always include a code of ethics. Such codes aren't intended so much to establish standards as to express standards that already exist informally. Nonetheless, they provide guidance concerning how professionals should perform their work, and there is significant literature in the area.

The issues raised in this article suggest that existing codes of ethics need to be reexamined in the light of developing technology. Codes generally fail to reflect the potential effects of computer- enhanced machines and the inadequacy of existing managerial, institutional, and legal processes for coping with inherent risks. Information technology professionals need to stimulate and inform debate on the issues. Along with robotics. many other technologies deserve consideration. Such an endeavor would mean reassessing professionalism in the light of fundamental works on ethical aspects of technology.

Asimov's Laws of Robotics have been a very successful literary device. Perhaps ironically, or perhaps because it was artistically appropriate, the sum of Asimov's stories disprove the contention that he began with: It is not possible to reliably constrain the behavior of robots by devising and applying a set of rules.

The freedom of fiction enabled Asimov to project the laws into many future scenarios; in so doing, he uncovered issues that will probably arise someday in real- world situations. Many aspects of the laws discussed in this article are likely to be weaknesses in any robotic code of conduct. Contemporary applications of information technology such as CAD/CAM, EFT/POS, warehousing systems, and traffic control are already exhibiting robotic characteristics. The difficulties identified are therefore directly and immediately relevant to information technology professionals.

Increased complexity means new sources of risk, since each activity depends directly on the effective interaction of many artifacts. Complex systems are prone to component failures and malfunctions, and to intermodule inconsistencies and misunderstandings. Thus, new forms of backup, problem diagnosis, interim operation, and recovery are needed. Tolerance and flexibility in design must replace the primacy of short- term objectives such as programming productivity. If information technologists do not respond to the challenges posed by robotic systems. as investigated in Asimov's stories, information technology artifacts will be poorly suited for real- world applications. They may be used in ways not intended by their designers, or simply be rejected as incompatible with the individuals and organizations they were meant to serve.
Isaac Asimov, 1920-1992

Born near Smolensk in Russia, Isaac Asimov came to the United States with his parents three years later. He grew up in Brooklyn, becoming a US citizen at the age of eight. He earned bachelor's, master's, and doctoral degrees in chemistry from Columbia University and qualified as an instructor in biochemistry at Boston University School of medicine, where he taught for many years and performed research in nucleic acid.

As a child, Asimov had begun reading the science fiction stories on the racks in his family's candy store, and those early years of vicarious visits to strange worlds had filled him with an undying desire to write his own adventure tales. He sold his first short story in 1938 and after wartime service as a chemist and a short hitch in the Army, he focused increasingly on his writing.

Asimov was among the most prolific of authors, publishing hundreds of books on various subjects and dozens of short stories. His Laws of Robotics underlie four of his full-length novels as well as many of his short stories. The World Science Fiction Convention bestowed Hugo Awards on Asimov in nearly every category of science fiction, and his short story "Nightfall" is often referred to as the best science fiction story ever written. The scientific authority behind his writing gave his stories a feeling of authenticity, and his work undoubtedly did much to popularize science for the reading public.

References to Part 1:

1. I. Asimov, The Rest of the Robots (a collection of short stories originally published between 1941 and 1957), Grafton Books, London, 1968.

2. N. Frude, The Robot Heritage, Century Publishing. London. 1984.

3. I. Asimov, I, Robot (a collection of short stories originally published between 1940 and 1950), Grafton Books, London, 1968.

4. I. Asimov, P.S. Warrick, and M.H. Greenberg, eds., Machines That Think, Holt, Rinehart. and Wilson, London. 1983.

5. I. Asimov, "Runaround" (originally published in 1942), reprinted in Reference 3, pp. 33- 51.

6. L. Del Rey, "Though Dreamers Die" (originally published in 1944). reprinted in Reference 4, pp. 153- 174.

7. I. Asimov, "Evidence" (originally published in 1946). reprinted in Reference 3. pp. 159- 182.

8. H.M. Geduld and R. Gottesman. eds.. Robots, Robots, Robots, New York Graphic Soc., Boston. 1978.

9. P.B. Scott. The Robotics Revolution: The Complete Guide. Blackwell, Oxford. 1984.

10. I. Asimov, Robot Dreams (a collection of short stories originally published between 1947 and 1986), Victor Gollancz, London. 1989.

11. A. Chandor, ed., The Penguin Dictionary of Computers, 3rd ed.. Penguin, London, 1985.

12. I. Asimov, "The Bicentennial Man" (originally published in 1976), reprinted in Reference 4, pp. 519- 561. Expanded into I. Asimov and R. Silverberg. The Positronic Man, Victor Gollancz, London, 1992.

13. A.C. Clarke and S. Kubrick, 2001: A Space Odyssey, Grafton Books. London, 1968.

14. I. Asimov, Robots and Empire, Grafton Books, London, 1985.

15. I. Asimov, "Risk" (originally published in 1955), reprinted in Reference 1. pp.122- 155.

16. I. Asimov, The Robots of Dawn, Grafton Books, London, 1983.

17. I. Asimov, "Liar!" (originally published in 1941), reprinted in Reference 3, pp.92- 109.

18. I. Asimov, "That Thou Art Mindful of Him" (originally published in 1974), reprinted in The Bicentennial Man, Panther Books, London, 1978, pp. 79- 107.

19. I. Asimov. The Caves of Steel (originally published in 1954). Grafton Books. London. 1958.

20. T. Winograd and F. Flores. Understanding Computers and Cognition. Ablex. Norwood, N.J., 1986.

21. I. Asimov. "Robbie" (originally published as "Strange Playfellow" in 1940). reprinted in Reference 3. pp. 13-32.

22. I. Asimov, The Naked Sun (originally published in 1957), Grafton Books. London, 1960.

23. I. Asimov. "The Evitable Conflict" (originally published in 1950). reprinted in Reference 3, pp. 183- 706.

24. I. Asimov. "The Tercentenary Incident" (originally published in 1976). reprinted in The Bicentennial Man, Panther Books, London, 1978, pp. 229- 247.

25. I. Asimov, "Little Lost Robot" (originally published in 1947). reprinted in Reference 3, pp. 110- 136.

26. I. Asimov, "Robot Dreams," first published in Reference 10, pp. 51- 58.

27. I. Asimov, "The Machine That Won the War" (originally published in 1961), reprinted in Reference 10. pp. 191- 197.

28. D. Bellin and G. Chapman. eds., Computers in Battle: Will They Work? Harcourt Brace Jovanovich, Boston, 1987.

29. I. Asimov, "Reason" (originally published in 1941), reprinted in Reference 3, pp. 52- 70.
References to Part 2

1. I. Asimov, "The Evitable Conflict" (originally' published in 1950), reprinted in I. Asimov, I Robot, Grafton Books. London. 1968. pp. l83- 206.

2. I. Asimov, Robots and Empire, Grafton Books. London. 1985.

3. I. Asimov, "The Bicentennial Man" (originally published in 1976). reprinted in I. Asimov, P.S. Warrick, and M.H. Greenberg, eds., Machines That Think. Holt. Rinehart, and Wilson, 1983, pp 519- 561.

4. I. Asimov, The Robots of Dawn, Grafton Books. London, 1983.

5. I. Asimov, "Jokester" (originally' published in 1956), reprinted in I. Asimov, Robot Dreams, Victor Gollancz, London. 1989 pp 278- ~94.

6. D. Adams. The Hitchhikers Guide to the Galaxy, Harmony Books. New York. 1979.

7. A.C. Clarke. Rendezvous with Rama, Victor Gollancz, London. 1973.

8. J. Weizenbaum. Computer Power and Human Reason, W.H. Freeman. San Francisco, 1976.

9. I. Asimov, The Naked Sun, (originally' published in 1957). Grafton Books. London. 1960.

10. I. Asimov, "Lenny" (originally published in 1958), reprinted in I. Asimov. The Rest of the Robots. Grafton Books, London. 1968, pp. 158- 177.

11. H. Harrison. "War With the Robots" (originally published in 1962), reprinted in I, Asimov, P.S. Warrick, and M.H. Greenberg, eds., Machines That Think, Holt, Rinehart, and Wilson, 1983, pp.357- 379.

12. I. Asimov, "Robbie" (originally published as "Strange Playfellow" in 1940). reprinted in I. Asimov, I, Robot. Grafton Books. London, 1968, pp. 13- 32.

13. A.E. Van Vogt, "Fulfillment" (originally published in 1951). reprinted in I. Asimov, P.S. Warrick, and M.H. Greenberg. eds., Machines That Think, Holt, Rinehart, and Wilson, 1983, pp.175- 205.

14. I. Asimov. "Feminine Intuition" (originally published in 1969), reprinted in I. Asimov, The Bicentennial Man, Panther Books, London, 1978, pp. 15- 41.

15. R.A. Clarke, "Economic, Legal, and Social Implications of Information Technology," MIS Quarterly, Vol 17 No. 4, Dec. 1988, pp. 517- 519.

16. I. Asimov, "Satisfaction Guaranteed" (originally published in 1951), reprinted in I. Asimov, The Rest of the Robots, Grafton Books, London, 1968, pp.102- 120.

17. J. Weizenbaum, "Eliza," Comm. ACM, Vol. 9, No. 1, Jan. 1966, pp. 36- 45.

18. S. Turkle, The Second Self' Computers and the Human Spirit, Simon & Schuster, New York, 1984.

19. A. Budrys, "First to Serve" (originally published in 1954), reprinted in I. Asimov, M.H. Greenberg, and C.G. Waugh, eds., Robots, Signet, New York, 1989, pp. 227- 244.

20. I. Asimov, "Risk" (originally published in 1955), reprinted in I. Asimov, The Rest of the Robots, Grafton Books, London, 1968, pp. 122- 155.

21. J. Wyndham, "The Lost Machine" (originally published in1932), reprinted in A. Wells, ed., The Best of John Wyndham, Sphere Books, London, 1973, pp. 13- 36, and in I. Asimov, P.S. Warrick, and M.H. Greenberg, eds.,Machines That Think, Holt, Rinehart, and Wilson, 1983, pp. 29- 49.

Return To Essay