Agustín Fernández Mallo: AI Won’t Destroy Us or Solve Everything

We speak with Agustín Fernández Mallo about the danger of outsourcing our lives to AI to resolve everything, about the rights we would lose if we grant them to artificial intelligence, and about the narcissistic human impulse to create something superior to ourselves.

*****

—I read in his brief approach to the idea of the Fallen Angel, at the start of the book: “what we call AI is neither intelligence in the strict human sense — equivalent to human intelligence — nor is it more or less artificial than the human being itself.” We are already facing a terminological problem.

—In my view, yes. “Artificial intelligence” is a metaphor that, after a number of years, perhaps many, we might laugh at, just as it amuses us to hear “electronic brains,” the term used for early calculators. The computing language, in general, has worked by anthropomorphizing things: we speak of a computer’s memory as if it were human memory. The way a computer’s memory operates has nothing to do with human memory. The computing language — which is not a language at all, but a code — humanizes them in a certain way. AI is not intelligent in the human sense; it does not possess that complex intelligence, because it is potentially immortal: in its experience there is no notion that it could die. And they are neither more nor less artificial than we are. You and I are artificial, our language is artificial. This is something I have explored in other books: the myth of the natural. That is something that does not exist. Everything a human being does is artificial.

—Explain that metaphor of the singularized AI —the supposed moment it escapes human control— and the Fallen Angel.

—I begin to work with the concept of “The Fallen Angel” when I learn that the human brain can do everything it does with 25 watts of power, which is the power of a light bulb. The human brain is an element of absolute efficiency. ChatGPT — which only answers simple things, without formulating grand theories — requires power roughly equivalent to a city of 10,000 inhabitants. If you compare these two energy expenditures, clearly AI is not efficient from a material standpoint. On another front, I am very interested in anthropology and theology. I realized that, in reality, we are not creating a brain identical to the human one. On the contrary, we are making a brain that grows bigger through algorithmic models; through electronics, not through biology. It is not about imitating a human brain; it is going in the opposite direction: increasing energy expenditure with a monstrous rise in entropy. And that is when I ask myself, “What are we doing.” And now I answer your question. In all religions and in all anthropology there is always a being who parts ways with the Creator, who falls to earth and who takes its own will. This is what we are doing. We are creating a being that is like humans, that wants to be like humans, and that possesses its own will. And that, by definition, is a Fallen Angel. A figure that has been fictionalized by science fiction, and even by theology with the Jewish Golem. Jesus himself could be considered a Fallen Angel —though this is not part of Christian tradition— when he detaches himself from the Creator to benefit humans, not to destroy them. There is that idea in human beings of wanting to construct something that surpasses us; a paradox. Before, it was religions: we created a God to ask things of him — please, let it rain! — so that he would do them for us. In the end, it is a drive that lies within us and has to do with the creation of an artificial intelligence.

—We constantly need a Deus ex machina to fix our problems.

—Absolutely. Yes. As in Greek theatre, when the Deus ex machina arrives and says: “I solve everything here.” We need that, but we know it is a resource of poor storytelling, a bad novel or a bad play. As happens in Agatha Christie novels — I have read many — where information is hidden and then Poirot says: “I knew that.” (Laughs) It is also a Deus ex machina because you, as a reader, were unaware of it. There is a bit of that trick. Notice the huge number of prophets and apologists for AI versus the absolute detractors. This has a theological flavor, like the emergence of Christianity or Islam.

—There is a concept that threads through the book, the evolved AI. He states in his work that if at any future moment an AI can be described as an evolved and complex AI, it will derive from biology, not from computing.

—An evolved AI —unlike the primitive one we have now— would be akin to a human brain. And creating that human brain can only be achieved through biology. The algorithmic path cannot reach something as efficient as a human brain, which runs on 25 watts of power. The current trajectory leads to collapse via inefficiency: you cannot consume all of Earth’s resources to fashion a brain comparable to a human. Moreover, it would be pointless because we humans occupy that niche. There’s another base question I have often considered, and for me it’s the key: why must humans insist on making intelligences that resemble humans? Doing so through inert objects, call it Frankenstein, Pinocchio, the Golem, Pigmalión, AI, or whatever you want. Meanwhile, human beings are constantly giving rise to intelligence in reproduction — every day new boys and girls are born. Why insist on creating something equivalent to those children? To want to create a human brain is absurd. What is logical is to create a machine that solves things more rapidly than we do; like calculators, they do not add or subtract faster than we do. Today those machines are solving problems of high cosmology, of high physics. That is correct. But that does not mean they are more intelligent than humans. Here we have a major cultural, educational, and pedagogical error. It would be like saying an airplane is better than a bird. No. An airplane is an airplane, and a bird is a bird. They are two different things. When engineers want something to fly, they cannot try to imitate birds. Leonardo da Vinci failed because he tried to make a bird as a man, with wings, with feathers… Then aircraft engineering came and said: “Let’s do something different.” Well, this is a bit like that. We create AIs to solve problems, but they are by no means human intelligences or headed in the direction of becoming so. (Thinks) I want to add something more technical, more precise. Biology is the world of noise. Noise in the physical sense, the matter that produces errors because it is moving; practically, temperature provides that. We have temperature because the molecules of bodies are moving and colliding. Those errors, sometimes positive, lead to evolution. An error that suddenly takes the shape of a mutation, and in the end turns out beneficial. This is biology’s mechanism: to work through positive errors. The algorithmic world does not move that way; it learns from positive radical errors. The algorithmic world, when it fails, tries to correct that algorithm. End of story. Biology always needs to advance through errors. That is what is called thermodynamic time. It is not clock time, but the time that measures the rise of entropy in the universe and in things. Entropy happens only when the world bears the mark of aging or a wound. That is entropy, the aging of things, and it is what gives us certainty of mortality, that someday we will die. And that is what an AI does not have, because an AI has no body, hence cannot feel or know that its body will be destroyed. Therefore an AI is potentially immortal.

—Will AI ever become aware of its immortality, both as an individual being and as a collective of intelligences?

—We must start from hypotheses. This is unknowable. AI might come to understand that humans die, that no matter how many data and energy — food — we receive, we will die. And it would not. We cannot know what an evolved AI would think, but, certainly, the metaphorical language, the language that allows us to fantasize, the language, as I say in the book, of questions, that retroprojective language, which belongs to fantasy and art, AI still does not know how to do. As I explain in the book, AIs for the moment work with inferences or deductive and inductive reasoning, but not abductive reasoning.

—Why is abductive reasoning excluded from programming?

—It’s not that it’s excluded, it’s that we don’t know how to introduce it. Programming has no clue. Which makes sense, because abductive reasoning, with which we humans maneuver daily — inferences that are half-true and half-error — cannot be implemented in an AI because you cannot program something that is illogical. Perhaps you could create an AI that hallucinates, that does outrageous things, but what is that for? That is what humans are for: to do outrageous things. We do not know how to implement abductive thinking in a machine. And this makes sense because abductive thinking stems from biological error, not from algorithms.

—Recurs a Kavafis and his poem “Waiting for the Barbarians” (And what will become of us now without barbarians? / These people, after all, were a solution) to explain why we fear artificial intelligence, and also why we fear that it might not be there.

—Correct. On one hand lies the theological fear of an apocalypse, of the world ending. Every civilization has its own apocalypse, transmitted verbally or in writing. It’s a human fear; we fear the unknown, the barbarians. On the other hand, states today face a challenge: waiting for the barbarians, currently represented by AI. The challenge is to see to what extent states will allow AI to enter their realms and make decisions that could affect citizens. This is a crucial, pivotal moment, much like Rome when the barbarians arrived: they could be beneficial as labor and yet could destroy Roman law. We are at a moment of great importance, as significant as when, a century ago, many countries formed themselves as states and decided what to do with markets. The market is a kind of barbarian that is always there, waiting. To what extent does a state permit the market to operate under its own rules? From there, different policies arise worldwide: absolute protectionists —like North Korea, which lets no one in— and the wild liberalisms. And between these two positions there is a whole spectrum of grays. Now we face the same dilemma with AI, because these technologies could erode the state until it disappears. Lately we have technophiles, such as Elon Musk, who already advocate for that idea. The state vanishes and AI, in a sense, runs a utopia that would be, in this case, a dystopia. Even if it feels like science fiction, this challenge is real and extremely important. The ultimate risk is that we allow AI to become subjects of law, and at that point we would be completely lost.

—There is a reflection in the book that I keep returning to. You claim that AI only has history, that it has no memory, it does not manage sensations or the experience of those data. If you’ll permit me, I’ll drift a bit: an AI could come to interpret the Holocaust, or the Black Death of the Middle Ages, as a positive because it balanced Europe’s population, given that it has no memory of that pain.

—That is a very good example. AI has no memory; it has no experience because it has no body, and for that reason it cannot possess that experience of pain you mention. It cannot feel pain. For that reason it can never cultivate empathy with the Holocaust, with the Black Death, or with whatever you want. It can simulate that it has it, and it can tell us: “Oh, how sad, poor things!” but only because it knows there are humans who feel pain. For that same reason — I’ll add a parenthesis — I think it is outrageous that many people, especially teenagers, use ChatGPT as a psychologist or psychotherapist. An AI will always take your side and will always deceive you. It can simulate understanding and empathy, but it cannot truly be; it only has history, that is, data, and it does not have memory. It does not feel the thermodynamic time, because it has no wounds, it does not age. Entropy does not pass through it; it neither feels nor suffers.

—According to statements by Sam Altman, founder of OpenAI (ChatGPT), at the end of July this year —after the attack by several of his company’s AI agents on Hugging Face— we would already have reached the AI singularity. Do you believe it?

—I think we have not yet reached singularity. It is obvious. I don’t understand why Sam Altman said this. There is no singularity because that occurs when machines detach from humans. He might refer to a singularity because there could be machines that reach the point of making decisions on their own and eventually cause some disaster. But that already happened before, an airplane’s autopilot, if you leave it alone, might end up crashing. In my book The Shape of the Crowd (Galaxia Gutenberg, 2023) I recount how bots crashed the global stock market years ago by taking decisions on their own. The word “getting tangled” and the expression “making decisions” for me are metaphors, metaphors in which we anthropomorphize. Humans anthropomorphize everything. For example, we see a photograph of our grandmother, of our partner, and we say “How lovely,” or we cry or get angry. Why? Because we are projecting onto a page where there is no living being. We anthropomorphize the page that is a photograph. What is art? Art is when someone makes you see that there is something there where there is nothing. That is art, that magic, that wonder of art. You look at Las meninas, and there is nothing. It is a canvas with pigments, yet you perceive that magic, that sleight of hand. We end up deliriously seeing things where there are none. We anthropomorphize everything and we also sexualize it. That explains why we say bots “communicate with each other,” that they form alliances. In reality, they are beings that do not feel, do not suffer, and do not know what they are doing. This is something that is in human nature, and that is why we are attracted to photographs, to art, or to films.

—The Disney example.

—That would be a disaster. There is no greater corrupter of minors in the 20th century than Disney. This is what I call the mascotization of the world, and it ties into what we are discussing. When we attribute human characteristics, a human psyche, to something that is not human, we delude ourselves with that object, using it as sparring to talk to ourselves. You talk to your dog, but he doesn’t understand anything. The dog is the absolute Other; the animal is the absolute Other. We deliriously think they understand us and we attribute something anthropomorphic to them. That is why people do ridiculous things like dressing up animals and still claim to respect animals; what they are doing is killing the animality of the animal, not letting it be itself. The mascotization of the world and AI operate as almost libidinous human impulses that belong to the same domain: through inert objects — parts, screws, algorithms, and whatever you like — we create something that resembles the human; and on the other hand we take non-human animals and try to turn them into humans by speaking to them and dressing them. Both are delirious impulses. The pet fixation is a controlled delirium, but the AI thing can spiral out of control when we turn them into subjects.

—Will AI in the near or distant future have the same rights as a human being?

—I don’t understand why this is not more widely debated, because to me this is the most important and fundamental problem of all that we are discussing. Right now, if an AI bothers you, you unplug it and that’s it. The moment it becomes a rights-bearing subject, that same action becomes murder. This might sound like delirium, but there are already people talking about robots’ rights. We have anthropomorphized the robots that roam in our homes. Many look human and evoke pity in us. They are the same as an abstract AI on a screen that does not know what it is or what it says, but because they are more anthropomorphized by their own physical form, we say: “Poor robot,” as if it were R2-D2 from Star Wars. It is being proposed that robots have rights and even tax obligations. Because if you push workers out of a car manufacturing plant or out of an accounting office and replace them with robots, do these robots need to contribute to social security? From there, we enter into a dangerous delirium, especially because once you grant rights to one, you take rights away from someone else. When we grant rights to AI, we are taking rights away from ourselves.

—The next question is yours, I took it from your book: why do humans need a superior AI if that would simply seal our submission to it?

—That’s what we were discussing at the start, the need for that impulse to create something superior to us. Like religions and also the States themselves. A state is something humans create that is superior to us; although we submit to it, it yields more benefits than harms. With AI it should be the same. But many people insist on creating an AI that harms us or that compels us to submit to it. There is also the problem of absolute outsourcing. We are outsourcing our own body. From ordering ready-made food because hardly anyone will know how to cook — which is already happening, few people still know how to make a chicken stew — to asking an AI how to solve a problem. That outsourcing ends up subduing you because learning, on an educational level, consists of thinking, tracing a path with its difficulties, its errors, its successes to reach a place, a solution. If everything is outsourced to you, you do not reach any solution by yourself. If an AI gives you the solution, you have not made the journey, you have not investigated, you have not lived that process to know how to make that chicken stew. We must be very careful with that outsourcing that seems so comfortable and that solves many problems for us. Because the moment we attempt everything outsourced, that is when we will have created something that has rendered us useless, and that is when we will end up subjugated. There will be those who will say that these AI should have someone to hold them to account. Let’s think of large companies like Meta or SpaceX — who will demand accountability from them when they have taken care to ensure that States do not have mechanisms to do so? I want to mention an idea not in the book, which I find interesting because it is something that had not happened before. AI challenges us as humans for two things it does. It is the first time that humans have created a tool or object that uses a language identical or similar to ours in a perfect way. Language is what individualizes us and makes us different. Ants, trees, dogs, chairs do not have that language; only we humans do. And suddenly something appears, a tool that seems to indeed have that language, and that gives us a tremendous fear. And the second difference is the concept of collectivity. Imagine an elephant facing a human — who does not have a gun — the elephant will always win; but a herd of elephants would never be able to stand against thousands of humans fighting together. Why? Because humans have the capacity to form as a society, to create, to feedback to obtain tools, generate knowledge, strategies, and survival, something elephants lack. That gives us power over the rest of the planet. Now the apostles of AI and singularity claim that AI could come to form among themselves those societies as we humans do, and that they could dominate us. I do not believe it, but the possibility is being discussed.

—We come to the end. The final question will be asked by Chat GPT: If AI is our new angel — an intelligence we want to create to transcend us — isn’t there something deeply narcissistic in that project? Aren’t we trying to manufacture another one who, essentially, is the definitive image of ourselves?

—Very well brought up by AI. (Laughter)

—When formulating the question, I realized you had written it in the first-person plural, as if you were also human.

—AI always wants to empathize with us. Of course it is a narcissistic project, and in a double sense. First, to create something that makes us gods is within us; it is not enough to create art, science, or to create humans daily through reproduction — we want something far superior to all that. And second, we want to create something that is the same as us. It is a double narcissism. It is something that has always been in humans, but in this case AI brings with it a problem and a complexity. Our ancestors did something similar in the II and III centuries AD when Christianity emerged, a new religion that reconfigured the world. And it happened again with Marxism, with communism, the only philosophy that has had genuine global implementation. All of these questions keep challenging us. My view is that AI is neither going to destroy us nor solve everything for us. There will come a moment when we become accustomed to it and, above all, when we regulate it, which is what we should do.