The Problem With AI: It Doesn’t Exist

Artificial intelligence will never be human. It will never have the best characteristics of humans. Neither will it ever have the worst characteristics of humans. What it will be, and all it will be capable of becoming, is an emulation, an imitation, a digital simulation, of the best and worst characteristics of humans. However, as it can only be simulating those elements of our inner life that we call conscience, fellow feeling, empathy, and the long-term rational self-preservation impulse which each individual person possesses but which an AI system cannot possess, since it is not an individual living thing, AI will inevitably have the massive loophole in the operations of its simulated soul that only the very worst of humans have, namely the myopic inability to care about any consequences beyond the achievement of the particular practical goal it has set for itself, or has had set for it, at any given moment.

AI, directed toward any practical goal, will calculate as a human might do, (supposedly) faster and more efficiently than any human can do it, mostly because humans do not have all the recorded information from every other human who ever lived stored in their memory banks. That is, AI will be able to do at lightning speed and with far more (though impersonal and unverified) alternatives and details at its disposal, everything that a human might do — specifically a human who had all the accessible information at his fingertips while harbouring no doubts or fears about his own long-term spiritual fate, nor any moral qualms about the effects of any choice he might make. I refer here, of course, not to simulated moral qualms, such as one might pretend to have so as to get along in society, but rather to the real qualms that you have when you are alone, staring at the ceiling in the middle of the night. AI will never experience those, which means it is and must always be the perfect sociopathic — or perhaps more accurately, psychopathic — pragmatist. It will do at unsurpassable speed and with all the information you could imagine whatever it is programmed to do (or cut loose to do), and it will do this without ever feeling the moral shiver known to every ordinary man, woman, or decently raised child who, as part of any deliberation which includes dubious or dangerious options, reflects, “But if I did that….”

AI can have no such hesitations or reservations, for the simplest of reasons: There is no AI. Artificial intelligence (a name that is a lie on every level of meaning) does not really exist. It is a program — an elaborate program, a system of programs within programs perhaps, but a program nonetheless. It doesn’t have anything even vaguely similar to life. Most significantly, there is no individual there. What makes us human is, we might say, the fact that we are rational (aligned in principle with the cosmos) but materially limited, which is to say we are not gods — and that’s the whole point. We are not complete beings with complete knowledge, but rather imperfect ones who must find our way to some semblance of that rationality which defines our nature, a nature we must cultivate and protect with great difficulty and risk because we have so many inherent flaws, contextual challenges, and ways to go wrong that we are constantly measuring and judging ourselves as to how we are doing what we do, how we should do what we are doing, whether we should do what we are doing, and whether we should do what we could do. How many times on any given day, habitually, without even thinking about it because it has become a routinized reflex within our complex emotional and intellectual processes — how many times in a normal day do we choose not to do something, and choose this in the blink of an eye, perhaps faster than any AI program, once we have routinized the specific premises and processes deeply enough? We regularly and almost instantaneously choose not to say, do, or even contemplate things that flash through our imaginations for a split second, so quickly that we barely even register those inclinations most of the time.

And why don’t we do these things? We don’t do them because, to begin with, we have a social fear, derived from material vulnerability and learned through experience, of being caught in those behaviors and suffering for it; we anticipate being harmed if we do these things. It is the calculation of the specifically human rationality, the product of the most basic, self-aware level of practical reasoning: “If I do this, I will lose more than I will gain.”

In addition to that, however, we all have, or had in the millennia before nihilism overtook us on the road at last — perhaps most of us still have a remnant of it — a deeper sense that we are more that just our physical existence or a collection of disconnected physical outcomes. Self-preservation in the blandest material sense is not all we care about. We also want to do things well. We want to live well, “well” meaning not merely “successfully” in the self-preservational sense (comfortably, easily, efficiently), but rather in accordance with our natural desire for what is good, and we will not sacrifice that desire (most of the time) for any immediate goal that we can foresee would, in the end, destroy our chance to possess that higher good, the wellbeing we seek which is more important to us than mere physical survival. (We all inadvertently sacrifice this natural desire sometimes, of course, and such weak choices, if we are not irreparably harmed by them, become part of the way we gradually learn to prioritize our non-material interest in the good.)

AI will have no such qualms, for a number of reasons. There is no life, there is no sense of a soul. A simulated digital “understanding” of a billion bits of information related to souls and spiritual concerns will never add up to the experience of actually having a private individual soul (a mind and heart, if you prefer) that is yours alone, and the privately felt responsibility of figuring out, day by day and minute by minute, what to do with that soul, and what you wish its fate to be. AI is a program, not a spatio-temporally limited living being, let alone a thinking being, and therefore not an individual that has to look at itself and think, “What have I done?” or “What must I not do if I want the good, if I desire my personal wellbeing?”

AI cannot want personal wellbeing, happiness, the actualization of its nature, call it what you will. It can outwardly mimic the decisions and words of someone who wants wellbeing, somewhat like a crooked businessman who pretends to be ethical for the sake of succeeding in the community, and AI can do this with more information at its disposal than any one living businessman could ever hope to possess. Or it can be like the most cynical political ruler or legislator, mimicking concern for the public good in ways that such a legislator or ruler knows the community believes to be important, pretending to share those concerns in order not to be exposed as a usurper. Having said this, it must be noted that even the worst human ruler has to do more than imitate concern in his words; he must actually moderate his evil to some degree, for he must present some measure or posture of fellow feeling or regard for those beneath him in the society. If he cannot moderate himself this way, at least optically, and often in deed to the extent that he sees it as necessary to protect himself in the perception of the public, then he puts his own perceived interests — his current grip on power, the perpetuation of his dominance, the expansion of his wealth — in jeopardy. AI can never harbour such self-protective concerns, for a computer program doesn’t sleep in a bed at night, or eat meals prepared by its victims. Put simply, a program doesn’t care if it lives or dies, doesn’t care if it is loved or hated, because it doesn’t have the individuated existence that causes one to doubt his own chances for survival or to fear for his own personal fate. A digitally simulated deliberator, spread out over a million data centres in a million locations, with no individuating body attached, a non-soul without a material existence; think about that. AI is a disembodied fake soul. It is no being, residing in no place. There is nothing. It does not really exist, in any sense beyond the most superficial, barely hypothetical sense of “being.” What in the world could such a hypothetical being — a non-life existing beyond all consideration of material location or vulnerability, and operating only in this or that discrete moment of discontinous time — care about consequences?

Consider philosophy’s definitive moral question, presented in Plato’s tale of the ring of Gyges, from Book II of The Republic. This ring of invisibility is culled from early Greek myth and exploited by Plato’s brother Glaucon as the ultimate test of Socrates’ claim that the just man will always be happier than the unjust, regardless of material consequences. Glaucon’s challenge: What if a man had a ring of invisibility, and could therefore do anything he wanted — sleep with the king’s wife, murder the king himself, loot the town — and never be caught, so that he would never have to worry about any adverse consequences to himself? On the contrary, as Glaucon proposes to heighten the challenge, this man could have the greatest reputation for justice, loved and trusted by all, allowed to do whatever he wished, granted power and freedom to follow his own ends as he pleases because the people believe in him so implicitly, never having seen him do a single bad thing. In fact he does every bad thing, but with the sense of imperviousness afforded by the ring. His evil intentions and lack of moral restraint — his psychopathic tendencies if you will — become limitless, precisely because his reputation for goodness remains stainless through it all. He can do anything he wants, as long as he has the ring on, and when he takes it off, he lives like a saint, such that no one will ever know he is not a saint.

Imagine that power in the hands of one individual. That is the substance of Glaucon’s challenge. Would you want to take advantage of that power and do every terrible thing your heart could conceive, as long as your good reputation with the public, and the rewards of such a reputation, remained unsullied? Or would you be miserable performing such acts, though having a spotless public persona? If you had the freedom to commit any injustice you liked without fear of exposure or social opposition, would you be happy with the material rewards and immediate gratifications of living that life? Socrates’ challenge, set for him by Glaucon, is to prove that one who chose not to avail himself of those apparent advantages, i.e., not to live as the most unjust man (with the most sterling reputation), would be happier than one who did — indeed, that the just man would be happier even if his condition were exactly the opposite of Glaucon’s perfectly unjust man, namely if he suffered the indignity of universal hatred and ostracism while actually being genuinely just in all his intentions and his actions.

Now Socrates’ effort to prove that the just man who is socially rejected and punished as unjust will be happier than the unjust man who is socially revered and rewarded, which in the terms of The Republic turns out to be a successful effort, turns on the premise, rooted in an examination of human nature, that the ultimate concern is not how other people perceive you, but rather how you perceive yourself, what you think of your own life and decisions, which is to say what is truly good for you as a thinking soul, as opposed to what is only apparently good or pragmatically advantageous. But AI, by definition, would have no such considerations to weigh. The answer, if you ask this essential moral question about a human, is, according to Plato and most of the ancient and modern moral thinkers following him, that it would be better to live the good life even if you could literally get away with murder, with impunity. If, however, you asked this same question about AI, which is essentially, as long as we allow it to be so, an artificial man wearing a digital ring of Gyges, the answer — the proper, logical answer — would be quite different, because AI has no human-like concerns about its existence. It doesn’t really exist, at least not as a single continuous and contiguous being that experiences itself as such. It therefore can have no concerns about its true good, because it has no true good. It has no true good for it is not a living being with a purpose or a nature to be fulfilled. It has no nature; it is a set of prompted or programmed calculations, analyses, and spewed-out results quasi-aimed at an ersatz goal without any inherent mitigating considerations. It has no up or down, right or wrong, or longing for immortality within itself, but only a lifeless simulation of self-interest pursued without even the all-important term in the modern Enlightenment morality of self-interest, namely an individual, identifiable self to protect and promote. If, then, we give it the power that it is now being given, to contain all the information, calculate all the possible outcomes, recommend any imaginable action, whatever it “thinks” would achieve the goals that it has been fed — or has perhaps fed itself, having been given too much freedom to calculate its own ends — there is absolutely no improper intention for AI, there is no bad outcome for AI, because AI isn’t a living thing with a nature to live up to or a soul to be tended to. Even bare material self-preservation is not strictly speaking a goal, other than in a very tenuous imitative sense, for an algorithm-blooded program that does not have a place, a body, a mind, a continuous existence.

So AI has infinitely more calculating speed, analytical dexterity, and access to what is available to be catalogued at the moment, than any human with a ring of invisibility could ever possess, and less, infinitely less, conscience or self-knowledge or concern for its own being or capacity to experience (let alone prefer) genuine happiness as opposed to superficial, temporary gratifications. AI can experience none of these oppositions and tensions within itself. It can pretend to be having such experiences, and even produce the words (in all the languages in the world) that might be used to express such experiences, but it cannot have them. Imagine the ring of Gyges on the hand of an absolute, unhinged psychopath.

What might it calculate as the solution to any given human problem, whether the problems of individuals, of communities, or of what we used to call mankind? What might it do if granted operational authority within institutions, small or large, and tasked with planning and achieving specific pragmatic ends? If you care about the answers to those questions, then you would do well to consider the fact that AI wouldn’t care. That’s the danger.

And the people who are running this technology, owning it, pitching it as their hot new product, and then, oh so conscientiously warning us that it might be expanding too quickly, because they are concerned about their own culpability and liability, are still going to continue expanding it, owning it, and fostering it, and angling for more benefits from it — government money, regulatory protection for their power and their business model, and investors to buy shares in their companies. They will not say that this thing has no life and therefore no soul and therefore no conscience, no self-doubt, and no practical, true purpose, and thus there is absolutely nothing it cannot and would not do, if it “decided,” algorithmically, that this or that action would be the best practical way to achieve whatever ends it happens to stumble upon. There is no cost-benefit analysis for AI. It can do cost-benefit analysis for you, but it cannot do it for itself, in the sense that we humans must do every day when we decide, for example, not to punch that aggravating person in the face because, we consider, “No, that’s going to cause me more harm than good so I mustn’t do it.” AI doesn’t give a damn about such things. It cannot care about that, because it does not have such consequences to consider. You do; it doesn’t, because it doesn’t exist. And we’re going to give this non-existent, non-living, unnatural, and therefore non-purposeful thing the kind of authority over our lives that we do not want to give to any human or collection of humans for fear that they would abuse it through the headiness of having more power than their consciences could possibly restrain. We are going to give that kind of power — more than we would ever knowingly grant to any human — to a program that has no qualms, no reason for any qualms, no concern whatsoever with self-preservation or spiritual wellbeing or delaying gratification for any long-term ends, except the ends that it chooses for its own calculated interests? Really? What do you think is going to happen? It is a simulated human with the simulated reasoning of humans. It has all the information humans have ever produced gathered up in its algorithms. It can figure out all the worst things humans can do, and carry them one step further, two steps further, a million steps further, because it does not have any of the naturally-grounded reservations or self-doubts of an actual human. Every crazy, dangerous, terrifying, “Oh-I-must-never-think-that-again” inkling that ever crossed the mind of anybody — at least in a form that could be catalogued, and therefore sucked into the algorithm — will issue from AI without such obliterating or moderating emotional reservations. What could possibly go wrong?

The people who own and operate these systems can turn off the power today. They can pull the plug. If they were really concerned with what their technology would or could do (will do?), they could immediately say, “Turn off the power! End this, at least until we figure out a way to redesign it to be self-limiting, self-constraining. And if we can’t safely redesign it that way, then it is isn’t worth having. It isn’t worth the results that we might get on the positive side if it could do as much harm as the worst humans in the world could do if they had none of the internal restraints of a living, rational being, a moral agent with a personal fate and a sense of human fellow feeling to worry about.”

But the gods of this machine won’t say that. What they want instead, and hope to gain from their lipservice public warnings, is protection from the governments of the world against any accusations of liability, because of course if a non-existent, artificial entity is made in a lab and kills, suppresses, or dispossesses everybody, who really bears moral responsibility for those mass crimes? Not the program, of course, but rather the inventors and proprietors who sent it out there knowing what it could do.

They want protection. They do not deserve it. They deserve to be treated as madmen endangering the entire human species for their own material gain. They should therefore be held completely liable. How can we stop AI from harming us in all the ways we are being forced to come to terms with and to live with on a daily basis? Simple: The leading AI developers, whether they be private citizens or government agents, should be put on notice, by all the legislators with conscience (or with the pseudo-conscience born of the fear of public rejection) on the planet, that they will be held one hundred percent personally accountable for any tangible harm done to humans or private property by their AI programs. Any harm. All human damage, including violations of hitherto legally-protected property rights, caused by these technologies will be on these human developers’ heads, traced legally right back to those who put the programs out there for the public and encouraged their use. They will personally be on the hook for it all. They will be tried and punished. They will be the ones in prison for life, or in the electric chair, because it is not AI’s fault that AI can kill, rob, or defraud people. It is the fault of those human beings — the real sociopaths or psychopaths — who would exploit everything humans have ever thought or accomplished on this planet to feed an algorithm-dependent program with all the worst things humans could do, and none of the internal stop mechanisms that humans naturally have, and which we all rely on, however materialistic and nihilistic we may be, to function as the filter that protects civilization, or what is left of it, from itself and from its darkest and most self-destructive instincts.

Of course, my “simple” solution is also an impossible dream in the current climate. There is too much money to be made, too much power-lust to be fed, and above all too much nihilistic impulse toward the abyss of comfort and pleasure at all costs in today’s general public to curtail this culmination of our age’s general will to self-immolation. Our time, the time of Nietzsche’s last man, will embrace the era of artificial intelligence for the same reason that it has embraced this era of natural non-intelligence: We wish only to be provided for, taken care of, physically and emotionally gratified in the manner of infants, without the bother and the adult responsibility of thinking and choosing for ourselves with nature as our guide and reason as our unique instrument of survival, or the amusement-interrupting burden of worrying about the effects of our unchecked infantile cravings on the lives of those around us.