Ethical AI
Moral Intuition and the Ethics of Consent
As the world grows in complexity, the need for a formal ethical philosophy has become more pronounced. Though there is ongoing disagreement among philosophers as to what the term actually means, we may regard ethics in general as ‘how one ought to act’. But more than just a guidepost for personal morality and decision-making, ethics has a social (political) aspect built into it; it concerns itself not just with how one ought to act, but by extension, how everyone else ought to act as well. The enshrining into law of this ‘moral fiction’ — from the Hammurabi Code to the 10 Commandments — laid the foundation for the earliest of civilizations. By standardizing a moral code, this allowed for the cooperation and interaction among large groups of people.
Then came the separation of church and state which created a slight distinction between morality (principles that guide how one ought to act) and ethics (rules that govern how everyone ought to act). While religion continued the cause of morality, politics — now free of the oversight of religion — still remained at its core a philosophy of ethics. Law itself is the codification of ethics. There is of course a distinction between what is illegal and what is morally unethical, but the notion that the government can prescribe certain behaviors that its citizens should or should not do is an ethical claim. Furthermore, it is generally seen by the government (if not by many citizens as well) as unethical, even morally unethical, to break the law. While in philosophy ethics carries the same sense as morality, in political usage, ethics has been stripped of a higher notion of morality (one is not moral for instance just because they avoid breaking the law). But in other fields such as education, medicine and business, ethics still retains its connection with morality — when businesses engage in unethical practices, we mean to say that they are acting immorally, even if it’s not always illegal.
From religion to politics to business, the new frontier now for ethics is that of technology. From biotech to self-driving cars to AI — where the stakes can be far higher than business hiring practices and taxation — there is a need for a more formalized ethical philosophy, or rather a return to ethical fundamentals. Just as science cannot do without metaphysics (as Einstein is reported to have quipped) ethics cannot do without metaethics.
We might have been able to avoid a formalized philosophy of ethics, or to take the pragmatist approach as we do now in adhering to different moral theories depending on the context, but particularly when it comes to the possibility of AGI (artificial general intelligence) and other technological systems that may one day be capable of autonomous reasoning, it becomes of paramount importance that we imbued these systems with a cohesive and coherent ethical foundation.
While we may have been able to get along with applied ethics without ever really confronting the metaethical principles that underlie our reasoning — given that for the most part we seem to understand metaethics primarily through intuition rather than reasoning — machines and systems do not have such intuition, at least not unless we program that into them, and therefore, in the programing of ethics, it is necessary that we formalize our underlying principles.
Discourse Ethics and the Principle of Consent
Ethics can be broken down into applied ethics, normative ethics, and metaethics. In order to do applied ethics — in order to determine what the ethical position to take on particular issue is — we appeal to normative ethics, or the framework with which we judge ethical and unethical actions. In normative ethics the two major camps (leaving aside others such as virtue ethics) include deontologicalism — where actions themselves are either ethical or unethical — and consequentialism — where actions are either ethical or unethical based on their outcomes. In general, we tend to be a bit of both: we are a deontologist when we say it is always wrong to kill our child no matter what good might result from it, but a consequentialist when we lie to our child to make them not feel bad.
Major thinkers in the deontological camp include Kant and Rawls. Mill heads up the consequentialist camp with his utilitarianism, supported by Hobbes and Locke. But another thinker who does not often share the moral theory spotlight but whose ideas are nonetheless compelling, even if idealistic, is Habermas. Habermas’ discourse ethics is, I argue, an improvement on Kant and Rawls’ moral theories, and provides, at least implicitly, a more nuanced alternative to both deontologicalism and consequentialism.
In discourse ethics, for a norm to be valid (or to judge an action ethical), it must meet with the approval of all affected. If we extrapolate a normative principle from this axiom, we may say that what determines whether something is ethical or unethical is not the action itself, nor the consequence of the action, but the situation in which the action occurs, or rather, the relationship between the acting agent and whatever is acted upon. If the relationship is consensual then whatever actions are taken are ethical. What is ethical then is what is consensual.
Whereas Kant and Rawls begin from the standpoint of the individual — both in a sense rewording the Golden Rule — Habermas extends that consideration to all those involved. ‘Do unto others as you would have them do unto you’ becomes ‘do unto others as you all agree you should do unto them’. This solves a problem inherent in the categorical imperative. If I were a Muslim that supported Sharia Law, I may be acting in accordance with Kant’s axiom in beheading blasphemers, though my neighbor might not be happy about that. Rawls reworks the categorical imperative and builds a theory of justice based on the liberty principle. But again we have a problem: If I am a proponent of polygamy, even from behind a veil of ignorance and abiding by the liberty principle, I would be acting in accordance with Rawls in legalizing (and practicing) polygamy, though my neighbor might not be happy about that either. Habermas solves the unhappy neighbor problem by requiring the consent of all affected parties. But discourse ethics introduces some problems of its own, including:
A. The bottleneck of unanimous consent
B. How to determine ‘affected parties’
C. Limitations on consent and the self-defense paradox
In the political arena (in democratic countries) we have adopted a more butchered version of Habermas’ ethics. Marijuana is now legal in a number of US states because the majority has decided so. It is now ethical, legally speaking at least, to buy and smoke weed. Politics has solved the bottleneck in discourse ethics by introducing majority rule. While quite effective in practice, majority rule doesn’t actually solve the problem in principle: what moral significance lies in the notion of majority? We might say then that it is the responsibility of the minority to acquiesce and consent to the will of the majority — which is in fact the norm in practice (except when the minority decides otherwise) — and thus we can attempt to show that we maintain a coherent ethical system. But on what principle am I to acquiesce? Being required to bend my will to the majority, just because everyone else is against me, is certainly not abiding by the principle of ‘what is ethical is that which is consensual’, but some other principle such as ‘what is ethical is that which is the general will’.
In the social world though (given our pragmatic adoption of differing principles as it suits us), we at times consider consent to be foundational. What makes a sexual encounter ethical is the consent of all parties — nothing more and nothing less. Of course, consent is not usually our overriding principle, not even in sex. If one of the parties is a minor for instance, consent is insufficient. Or even if they are both adults but a monetary transaction takes place for sexual services, consent (in most of the US at least) is again insufficient. But by what principle is consent insufficient? We may be able to retain the coherency of the principle of consent by saying that it is not consent that is insufficient, but rather that the scope of the affected parties becomes enlarged to include the whole community. Given that the care of minors and the effects of prostitution are matters that are said to concern the community, and given that the community (by majority vote) does not give consent, the act then is unethical. But who gets to determine the scope of the affected parties? If the condition for what is ethical is that it meets with the approval of all affected parties, how do we determine who is affected?
Additionally, who gets to be counted as potential affected parties in the first place? (A dog seems to society to count as an affected party with a claim to consent in a way that a hunted deer does not.) In addition to pets, what about fetuses, are they considered equal in their role as affected party, or like pets, is their consent valued less? The consent of the mother, in the case of a pregnancy that if left to term would prove fatal to both, triumphs the consent of the fetus even in such jurisdictions that reject abortion on the grounds that the unborn child is a full individual. We might still avoid this coherency dilemma (given that consent by definition is egalitarian) by determining that it is not consent that is diminished but that certain parties are limited in their scope of affectedness. We might say that in the act of the mother saving her own life, the fetus is an only partially affected party — not because they are not wholly affected by the mother’s decision, but because their capacity to be affected by death in this instance is less than the capacity of the mother.
In the example of a sexual encounter among two adults, the scope of affected parties would be limited to the two of them. Yet if one had a spouse, we would consider the spouse as one of the affected parties, just not equal in affectedness and not demanding the same level of consent — which is why infidelity is not a crime, but rape is. But who gets to determine who is counted as an affected party and to what degree their consent counts? (And can we even make these determinations without consent losing its coherency?) The answer society provides us with is of course, majority rule.
Yet despite the adoption of majority rule, in our everyday interactions in general we tend to uphold the principle of unanimous and equal consent, with a notable exception. If someone, such as a burglar, were to act in opposition to this principle and act towards us in violation of our consent, we would also abandon our commitment to consent. We might in a Kantian manner say that the burglar (like the inquiring murderer) has exempted themselves from ethical consideration, or has removed themselves from the equation by acting first in contradiction of consent. But this attempt doesn’t actually solve the problem, since by what principle are we able to eliminate the burglar as one of the affected parties? We might then say that there is a limit to what we can consent to or not consent to, and the burglar cannot do other but consent to me resisting him — so I am free to proceed, given his tacit consent, and resist the burglary. But can consent ever be compelled?
In a libertarian manner we can say that my consenting to swing my arms ends where the consent of your nose begins. Therefore consent is limited by consent. This of course is another way of expressing discourse ethics: only those parts of consent that overlap can be considered ethical. In this way, consent may delimit consent, but my consent cannot ever compel your consent; consent has to always be freely given. So we remain stuck. Though the burglar has transgressed against my consent by attempting to rob me, by what principle is it ethical for me to transgress against his consent by punching him in the face? Society offers its own solution. Since the earliest civilizations, the principle of judgment served as the counterweight to consent. For all society’s talk of consent, it remains that judgment, rather than consent, is still regarded as the highest ethical principle. (But do we really want to embed AI at its core with a judgmental ethics?)
How do we keep the principle of consent from being contaminated by judgment? If we want to say that what is ethical is that which is consensual, then how can we ever act in self-defense without contradicting ourselves? This is the self-defense paradox: it would be unethical for me to allow a burglary to continue since it does not meet the requirement of an ethical action given that I am an affected party and I did not give my consent; and yet in acting in self-defense I am acting contrary to the principle of consent given that the burglar, let’s assume, does not approve of my preventing him from robbing me. In the example before of abortion, we might say that the affectedness of the burglar does not match my own affectedness — given that my possessions being stolen outweighs a punch to the face — but this principle of determining affectedness gets even less convincing here.
An alternative (albeit highly idealistic) attempt at solving his paradox comes to us from the Biblical injunction to turn the other cheek and to not repay evil with evil — a position which Leo Tolstoy took up in his works on the principle of non-violence. Not only is this principle of non-violence consistent, but it is as if in consenting, in turning the other cheek (and not because we are required to do it, but because we ‘love our enemy’), we are liberating the aggressor from his unethical or sinful condition and thus maintain a coherent ethics. This transcendent ethics was not taken up by politics for obvious reasons, but might yet be the ethics we would want to program into our machines and systems: an incapacity for violence, even though it would mean an incapacity for self-defense.
It seems to be the case that we hold to two (conflicting) fundamental principles: justice and consent. But regardless of our natural propensity for justice, we seem drawn towards consent as the highest ideal. But what do we mean when we say consent? In the case of a drunken friend who is attempting to drive themselves home, we would feel ourselves acting with their total consent in preventing them from getting into their car, despite the fact that they themselves in their drunken state may accuse us of acting against their consent. We know better. We know that our future non-drunk friend would thank us, and it is that future (or ‘higher’) self that we have in mind when we say we are acting with their consent.
But again we have a dilemma: who is the one to make that claim? Many dystopian fictions (and not just fictions) revolve around elites as the gatekeepers of morality, acting on behalf of the masses ‘for their own good’. But like the case of the intoxicated friend, our claim to be acting in their best interest can be (and ought to be) justified by the future, when their future self has consented — a justification that dystopian elites always find a way of forestalling. But who determines this in the interim? Do we designate a medium who can speak for the future?
If we want to abide by the principle that what is ethical is that which is consensual, then what we mean is ideal consent. This harkens back to Plato’s notion of the Good. What is most ethical is that which is most consensual (as Peterson would say) across all levels of analysis, taking into account ourselves as existing on a continuum from past to the future, and as having lower aims and higher aims. Love is the word we have to describe this kind of consensual relationship — that treats others the way they would (as their future and higher self would) want to be treated; or put another way, to act in their best interests even if they themselves cannot see it (like sending a child to school, or hiding your drunk friend’s car keys). This kind of consent can only be justified retrospectively and from a position beyond oneself; it therefore takes a kind of intuition.
Relational Metaethics and the Principle of Intuition
In what I have been describing as an ethics of consent, what is ethical is that which is consensual. Peeling back this normative principle further, what is ethical refers neither to the act nor to the result, nor even to the situation, but to the relationship. This provides us with a metaethical position that has moral facts residing not in the Self, or in the Other, but in the space between. What is ethical is not simply an imposition of the Other as Levinas might say, nor simply rising from the Self as Levinas argued against (ethics wouldn’t mean anything if only a singular Self existed), but rather ethics arises in the relationship between the Self and the Other. This relational ethics as a kind of transcendent morality, requires one to step outside of oneself into something dynamically experienced rather than statically deduced. It is therefore not something that can simply be reasoned but must be intuitively felt. This intuition, I claim, is the wellspring of ethics.
Will we ever be able to iron out a coherent governing principle which to encode in our machines — one that can avoid the unanimous bottleneck, that can solve the paradox of self-defense, that can account for our notions of consent and affected parties, and that can avoid having to maintain this system through a threat of punishment? Or is it simply too idealistic and the best we can do is rely on a pragmatic compromise as we have done for ourselves in the past?
But while pragmatism may not pose a dilemma for smart technology like self-driving cars that are simply algorithmically-based, if we ever expect to create reasoning technology — devices and systems with autonomous reasoning capability — we cannot just hope that such technology will somehow be imbued by way of osmosis with this feature of intuition which seems to serve as the touchstone of our reason.
Like non-violence, like consent, intuition is idealistic, but it highlights what should not be overlooked in the pursuit of programming ethical machines. In spite of it being idealistic, or rather for the very reason it is idealistic, an ethical intuition such as the one described herein ought to be what we program into our machines. It is therefore on us if we are to pursue reasoning machines, to make machines not just capable of reason, but capable also of intuition.
Bibliography
Habermas, Jürgen, and William Rehg. Between Facts and Norms: Contributions to a Discourse Theory of Law and Democracy. MIT Press, 1996.
Habermas, Jürgen. Discourse ethics: Notes on a program of philosophical justification. In Moral Consciousness and Communicative Action. MIT Press. 1990
Kant, Immanuel, et al. Groundwork for the Metaphysics of Morals. Oxford University Press, 2019.
Levinas, Emmanuel. Humanism of the Other, Trans. Nidra Poller, Introduction by Richard A. Cohen. Illinois University Press, 2003.
Plato. The Republic. Books, Inc., 1943.
Boddington, Paula. Towards a Code of Ethics for Artificial Intelligence. 2017.
Rawls, John. A Theory of Justice. Cambridge, Massachusetts : The Belknap Press of Harvard University Press, 1971.
Tolstoy, Leo, & Leo Wiener. The Kingdom of God Is Within You; Or, Christianity Not As a Mystical Teaching but As a New Concept of Life. New York: Noonday Press, 1961.
