The Oxford Handbook of Ethics of AI (Oxford Handbooks)

by Markus D. Dubber

The Ethics of the Ethics of AI

AI technologies will bring risks and rewards for individuals and societies.

When combined with efficient data-gathering techniques and breakthroughs in genetics, nanoscience, and cognitive science, AI will almost certainly entice us to effect a greater mastery of our planet. Perhaps AI will first pass through a stage of attempts, via surveillance, policing, and militarization, to also master other human beings.

Heretofore, ethicists have understood key ethical concepts, such as agency, responsibility, intention, autonomy, virtue, right, moral status, preference, and interest, along models drawn almost exclusively from examples of human cognitive ability and reasoned behavior.

Artificial intelligence will challenge all those concepts, and more, as ethicists begin to digest the problem of continued human coexistence with alternate (and perhaps superior) intelligences. That is to say, AI will challenge the very way in which we have tried to reason about ethics for millennia.

First, we (ethicists) generally learn of AI applications only after they appear, at which point we attempt to “catch up” and possibly alter or limit the applications.

In philosophy, an agent intends (upon reflection) its actions. It is aware of the selection of intentions, and it initiates actions based on them. In other words, artificial agents (for philosophy) do not have agency.

First, while we might indeed wish to acknowledge rights that belong to humans qua humans, we must note at least two immediate conceptual difficulties. That is, (a) is “human” a stable category? Many heinous instances of depravity and violence hinge precisely on the denial of the status to groups of entities targeted for violence or elimination.

we must ask (b) whether human is the relevant category for rights protection. Any rights regime involves some form of means test, to qualify for inclusion within the regime; but “human” is a biological category, at best a disputed one, and therefore seems an unstable basis for a program of rights protection. Why, after all, should biology determine whether an entity qualifies for the cover of law?

If universal is to mean anything, it can’t merely mean those we know and already recognize—this would very quickly toss us back to the problems of “human.”

But universal is itself a tricky property, not least because resisted by those who want to insist on particularity and distinction. Cultural difference and those who defend it seem to cut against universalism, which is often perceived as a top-down mechanism for eliminating distinct claims of identity.

Worse, universalism can sometimes appear, despite good intentions, to be allied with objectively harmful economic regimes of the so-called New World Order—now not so new—of globalization and deregulation in trade and capital.

Rights regimes are dominated by assumptions, again often Kantian, that the bearers of rights are not only individuals but deracinated, abstract individuals: the rational choosers of standard economic analysis, really, or the isolated contractarian actors that appear everywhere from Hobbes, Locke, and Rousseau to Rawls and Gauthier.

All of this is really to say that traditional rights regimes are conceptually unstable, but that this may provide us with an unparalleled opportunity to expand and revise their basic assumptions. This is happening already, as we know: many advocates argue for the rights of nonhuman animals, such that we human animals ought to be forced, morally or even legally, to alter our behavior with respect to treatment, eating habits, and duties of care.5 Other advocates consider nonspecific entities, such as the environment or the planet, or specific forests and regions, to possess rights that should be morally and legally claimable.6 There are also those who believe that groups or cultural identities themselves have rights, not just the individuals who fall under their umbrellas of meaning.7 In none of these cases is the target rights-bearer able to claim the rights explicitly. But this is not possibly a crippling objection. Even under straight-up presumptions of individualism and universalism, the sufferers of rights violations are not always able to speak for themselves. Sometimes they are incarcerated. Sometimes they have been silenced by threats and torture. Sometimes they are just no longer with us. In no case is it a valid objection that a violated entity cannot make its own claim. Thus there is no reason in principle to object to the inclusion in rights regimes of those who cannot speak for themselves, or those who are not even capable of speech. That is what advocates are for.

The most promising way forward in rights thinking, it seems to me, is to execute analysis on the basis of risk. Risk in turn is a function of vulnerability, and the distribution of risk, while dependent on many factors including birthright lotteries and structural limitations, is arguably the central concern of social justice. If the protection of human rights, and the punishment of their violation, are to mean anything, they must serve the ends of justice in this respect. Minimizing risk, or equalizing its distribution, are pragmatic goals of a valid rights regime. Now we must ask: is such a regime open to the possibility of nonhuman agents who are individual and conscious, but not biological? This is the main question before us: can AIs ever be persons?

Philosophers are divided, and have been for some time, concerning the prospect of generalized autonomous AIs (GAAIs), whether in human form or not, who could achieve consciousness. They are nevertheless united in thinking that such consciousness, supposing it possible, is at least a necessary condition of potential personhood. Further conditions conducing to sufficiency might then include decision-making ability, the awareness of choice and its consequences, and the ability to tell right from wrong, to suffer and be violated, and so on. Only such a being would seem a likely candidate for inclusion within any traditional rights regime, even one open to quite radical forms of otherness.

What if nonhuman agency develops, instead, as collective or system-wide property of interlinked complex algorithms? If we take seriously the idea that nonhuman animals, environments, and cultural groups are all worthy of rights protection under existing traditional regimes, it would seem perverse to deny such protection to a vulnerable and responsive system that, for various good reasons, does not (or not yet) exhibit the individual subjectivity of human agents.

the realization of a fully conscious individual nonbiological system—is that such an entity must entail embodiment, so that it can experience the world phenomenologically and therefore understand its emplacement within environments. This has seemed an insuperable barrier to many philosophers, since the deployment of a body appears to be beyond the technological capacities of AI systems. Once again, though, the assumptions here are revealed as tenuous, if not outright invalid, and based upon allegedly baseline behavior that is everywhere changing under existing technological conditions. Yes, most human agents have bodies that move through the world, exercising the basic physical sensorium to gather, collate, and filter external stimuli for internal purposes. A sense of emplacement remains an important dimension of human phenomenology, as do the sense-based experiences of seeing, hearing, tasting, and feeling. Without sunsets, symphonies, good meals, and relational intimacy, the world is a dry place for humans. But this norm admits of many exceptions: there are many human agents with less than optimal sensory or ambulatory ability, for example. I am myself deaf in one ear, which deprives me of some experiences without challenging my sense of self. I also can’t run or jump the way I used to—who can? Even more significantly, there are many extensions of personhood achieved precisely with nonphysical stimuli: online interactions, media immersion, disembodied tracing, and so on....

For current purposes, the essential philosophical question concerns the oscillations created between apparent or at least partial humanness and the elements of otherness indivisible from nonhuman conditions of existence. In many ways, this series of oscillations—They’re like us! They’re not like us!—exactly matches the same anxieties evident in social movements that expanded the range of legal status and rights regimes within the biological category of the human. Entities now clearly within that biolegal status were once excluded from it, shamefully, and even now there are mechanisms within legal and national regimes to exclude presumptive claimers of “inside” status.

It is a matter of record that the term “robot” was first used in recent recordable culture by Czech writer by Karel Capek, in his 1920 science-fiction play R.U.R. (“Rossum’s Universal Robots,” in English). Robot is a word that derives, in Czech etymology, from robota, or forced labor. Thus the robots in the play are, in effect, slaves—and not even wage slaves, because they are assumed, in their mechanical efficiency, to require no food, shelter, or clothing, let alone healthcare or pension plans. In short, they are the perfect solutions to the problem of labor. They work on command, do not tire or complain, and they don’t need to be paid.

The robots are us, and we are the robots, whenever there is a resented central government, state labor restrictions, and centralized authoritarian power.

The urge to bestow fire—maybe now in the form of consciousness, the fire of the mind—is essential. Prometheus is a symbol of human striving, especially in science and technology. But he is also a symbol of what happens when overweening ambition outstrips common sense or regard for whatever we mean by “the gods.”

Pundits and performers will fight rearguard actions on these matters as long of most of us are here on the planet, but they cannot win the day, because the category of “human” refuses to be pinned down.

Philosophers have typically responded to these quandaries by trying to shift to discourse from “human” to “person.” Human biology is sufficient for personhood, as long as there is a decent regime of law in place, but it is not strictly necessary. That is, there may be nonhuman persons. Indeed, corporations are persons in various legal jurisdictions, subject to both legal punishment for wrongdoing and, maybe less benignly, the legal right to express themselves politically and financially.

If you had what felt like reliable memories and experiences, how could you really know the difference between yourself and a created being? After all, we are created beings, just using flesh instead of silicon.

Thus, in the present case of GAAIs, we have a doubling effect. There is what looks very much like a racial or ethnic conflict (humans versus nonhumans) supplemented by, or crossed with, a technological conflict (those who see emancipation there versus those who see enslavement). Fear multiplies fear, and the prospects of any smooth integration of human and nonhuman entities are rendered more and more remote. The more advanced the technology becomes, while allowing the theoretical possibility of GAAIs, the more that very technology and its products are feared and resented. Otherness with human biology in play is hard enough to confront; otherness with no biological kinship would seem insurmountable.

First, then, the most likely scenario is that, like all present AIs including diagnostic programs, driverless cars, and military drones, the GAAIs will be considered wholly owned property. This would give them the status of, in effect, Aristotelian-style slaves, without personhood status though retaining abilities far beyond other animals and, of course, inanimate objects. It is not clear what advantages AI consciousness, assuming it were ever possible, would add to the ability of such property-based entities to function.

second scenario, then, where the GAAIs are granted secondary but significant status as welcome semi-individuals, like children, pets, or family retainers. There are many depictions of this scenario, but the most obvious is probably the extended Star Wars film franchise, where droids are considered by some to be second-class citizens, in an obvious suggestion of anti-digital racism, but are otherwise granted respect, affection, and responsibility. R2-D2, C-3PO, BB-8, and other examples of the scenario are played out in various scenes of the groaning catalogue of films. The attitude to droids changes somewhat over the now-lengthy duration of the franchise, so that more affection, humor, and autonomy are afforded the nonhuman entities.

This scenario is popular in part because it is the obvious counter-narrative to the dominant strains of fear in much of the cultural depiction of GAAIs. But it has its own internal difficulties. No matter how well liked, a servant remains at best a wage-slave and at worst a favored chattel-slave.

third possible scenario, the one which most people imagine when they think of androids or other forms of GAAIs, namely, full autonomy even under conditions of radical otherness. Here, once again, we are in the realm of pure speculation, but it is this scenario that receives by far the most attention.

Such nonhuman entities would seem, per assumptions, to be candidates for full personhood and hence recognition under rights regimes and other legal protections. But their existence would present a challenge to the extension of these regimes and protections that is without precedent. As noted, the umbrella of rights has been spread over previously excluded groups, including women, people of color, and LGBTQ people. These groups all had the advantage of being able to claim (eventual) recognition on the basis of species-resemblance—this despite efforts to deny such resemblance. Nonhuman entities have not fared so well, including animals and environments. For the most part, such entities have relied on trusteeship and stewardship to protect their well-being.

if these GAAIs recognize their inherent special status, whether recognized in law or not, new philosophical questions arise. Do they die, for example? If not, how does that affect legal status? Is there need for sustenance or maintenance? Is employment an option? Reproduction? Acquisition of wealth and its transfer? Are they vulnerable to pain, suffering, and emotion? Without these, could they make or understand art? And so on.

The fourth and final scenario I wish to entertain is at once more radical and, at least some people’s thinking, more plausible that the second or third. This is the possibility that, instead of fully separate GAAIs being created and then confronting us with claims to independence, a middle way will be pursued in the form of cyborg relations and posthuman hybrids. On this view, rather than complete otherness, we will confront GAAIs through a conjoining of algorithmic and technological elements with existing familiar biology. Since many human forms already contain aspects of the cyborg in the form of both internalized technology (pacemakers, metal joints) and external mediation (constant access to otherness via a smartphone, for example), it is not at all impossible that an evolutionary step is imminent in which these connections become more complex and permanent.

Introducing prostheses and other forms of body modification, or even enhanced cognitive ability, raise familiar issues of unequal distribution of goods. Posthuman transformation could become a justice issue, in short.

the more we press issues of sentient AIs and their possible autonomy, the more we are thrown back upon fundamental existential questions concerning human existence. The nonhuman autonomous entity is a necessary counterpoint to the dominant narrative of human identity and meaning.

We cannot allow the depredations of the technological attitude, as depicted by Ellul, Mumford, Bookchin, Heidegger, and others, to dominate the scene.14 What Heidegger calls Ge-stell, or enframing, is the rendering of all aspects of the world into “standing reserve”: a condition of availability and revealing that may retain poetic or creative elements but which, in the event, most often indicates a kind of instrumental use-value calculus of resources both natural and human.15 The forest is seen as lumber, the river as electric power, the human as … what? Perhaps a perpetual gig worker, victim of aspirational advertising, and unwitting supplicant to social-media addiction and upgrade anxiety? Yes, that sounds familiar.

Despite the rise of robotics and AI in wider society there are no “anthropology of AI and robotics” journals, research departments, or a clearly delineated body of research. There are billions of dollars worth of investment in robotic and AI research and business, the products of which are likely to produce far-reaching changes for humans, animals, and the environment.

The growing interest in the ethics of robots and AI has been largely developed by philosophers of technology, not anthropologists, and this has shaped the narratives, issues, and concerns of social scientists engaged in robotics and AI. Alternatively, I want to suggest that anthropological paradigms are capable of opening up new kinds of reflexivity, and as such there is a case to be made for increased anthropological engagements in these fields.

The origin stories of robots and AI are entangled with advanced industrialized economies and are simultaneously fictional, business, and research artifacts.

AI is also complemented by a mythical imaginary where the end point is a conscious machine, capable of reasoning, abstraction, speech, and language and even sentience. Just as robots are portrayed as a threshold technology where boundaries between human and machine and animate and inanimate are dissolved, so too is AI. These fantasies underscore the cultural imagination of robots and AI, and, arguably, contribute to an illusion fused with anxiety: will they or won’t they rise up?

Anthropologists explore categories of space, time, symbolism, ritual, and cosmologies capable of bringing to the fore extra layers of meaning that are invisible to the architects of robots and AI.

Arguably the rise of ethics of AI and robotics will create committees of stakeholders who will work together to reduce bias, stereotyping, and perpetual algorithmic inequality.