In Arthur C. Clarke’s famous short story "The Nine Billion Names of God", a sect of Tibetan monks believes that human existence has a goal set by divine forces: to write down all the countless names of God. They believe that once this list is complete, God will put an end to the world’s existence. After centuries of writing the names of God by hand, the monks finally decide to use modern technology. Two sceptical engineers arrive in the Himalayas, bringing a super-powerful computer. With its help, the 15,000 years it would have taken to list all possible variations of the name of God are no longer needed; the work is finished in three months. The engineers mount their ponies and begin their journey down the mountain. Clarke’s story ends with one of the most stark final sentences in the history of literature: "Overhead, without any fuss, the stars were going out."

This scene portrays the computer as a shortcut to objectivity or ultimate meaning, and it is precisely this view that is at least partially responsible for our current universal fascination with artificial intelligence. Although the technologies underlying AI have existed for some time, it is only since the end of 2022, with the introduction of OpenAI's product ChatGPT, that the impression has emerged that we are very close to accessing an artificial intellect. In a 2023 Microsoft Canada report, President Chris Barry declares that "the AI era has arrived, bringing a transformative wave with the potential to impact every facet of our lives" and that "this is not just technological progress; it is a social change that is moving us towards a future where innovation will take centre stage." This was one of the more level-headed reactions. Artists and writers are in a panic that they will become redundant, governments are scrambling to keep up with the process and regulate it, and scientists are engaged in passionate debates.

Companies are eagerly seizing the opportunity to jump on the hype train. Some of the world’s largest corporations, including Microsoft, Meta, and Alphabet, are placing massive bets on AI. In addition to the billions spent by big tech companies, funding for AI startups reached nearly 50 billion US dollars in 2023. At an event at Stanford University in April, OpenAI CEO Sam Altman said he was not worried about his company spending 50 billion dollars a year on AI projects. He also hopes to create a kind of super-assistant, a "super-competent colleague who knows absolutely everything about my life, every email, every conversation I've had, but who doesn't feel like an extension of myself."

Yet alongside this, there is a deep-seated conviction that AI is a threat. One of the most well-known critics who claims that AI poses an existential risk is philosopher Nick Bostrom. As he explains in his 2014 book "Superintelligence: Paths, Dangers, Strategies": "If we build machine brains that surpass human brains in general intelligence, the fate of our species will become dependent on the actions of a machine superintelligence."

The classic example on this topic is the story of an AI system whose sole and seemingly harmless goal is to produce paperclips. Bostrom believes the system would quickly conclude that humans are an obstacle to achieving this goal because they might turn it off at some point. They might also use up the resources needed to produce a larger quantity of paperclips. This is an example of what AI catastrophe forecasters call the "control problem"—the fear that we will lose control over AI because any security protocols we have built into it will be eliminated by an intelligence that is a million steps ahead of us.

Before we concede even more to our technological masters, it is worth looking back at the mid-90s and the arrival of the World Wide Web. It, too, was accompanied by various predictions of a new utopia, a fully connected world where borders, differences, and scarcity would come to an end. Today, it would be highly questionable to claim that the internet is an entirely unproblematic good. The fantasies did indeed come true: we can carry the world’s knowledge in our pockets. But this has had a rather strange effect—it has made people a little crazy, increased discontent and polarisation, contributed to a new surge in the activity of far-right forces, and destabilised democracy and truth. It would not be right to say that we should simply resist technology, as it also acts in the opposite direction, increasing freedom. Rather, it should be said that when big technology comes to us with a gift, we should first cautiously look at what is inside.

What we call AI currently focuses primarily on LLMs, or large language models. These models are fed huge datasets—ChatGPT has essentially vacuumed up the entire public internet—and they are trained to recognise patterns in this data. Units of meaning, such as words, word parts, and letters, become tokens and receive specific numerical values. The models memorise how each token relates to others, and over time they learn something akin to context: where a word might appear, in what sequence, and so on.

By itself, that does not sound particularly impressive. However, I recently asked ChatGPT to write a story about a cloud that has feelings and is sad because the sun has come out; the result was surprisingly human. The chatbot had not only included various elements typical of a children’s fable but had also developed a plotline where the cloud, named Nimbus, finds a corner of the sky for itself and makes peace with the fact that it is a sunny day outside. Perhaps one wouldn't call it a particularly good story, but it would certainly amuse my five-year-old nephew.

Robin Zebrowski, a professor of cognitive science and chair of the department at Beloit College in Wisconsin, explained the human-like quality I felt this way: "The only truly linguistic things we have ever encountered are those that have minds. So when we encounter something that seems to handle language like we do, all our past experience kicks in and we say: 'This is undoubtedly something thinking!'"

Therefore, for decades, the standard test used to check whether technology was approaching intelligence was the Turing test, named after its creator, the British mathematician and Second World War codebreaker Alan Turing. The test involves a person who, through text messages, questions two invisible subjects—a computer and a human—to determine which of them is the machine. The roles of questioner and respondent are performed in turns by several people, and if the machine succeeds in fooling a sufficient number of questioners, it is acknowledged as exhibiting signs of intelligence. ChatGPT is already capable of fooling at least some people in some situations.

Such tests show how closely our notions of intelligence are tied to language. We tend to think that beings capable of "handling language" are thinking: we marvel at dogs that seem to understand complex commands, or gorillas that can communicate in sign language, precisely because it is closer to the mechanism by which we ourselves make the world meaningful.

But perhaps the ability to use language without the simultaneous ability to think, feel, want, or even exist is the reason why the texts written by AI chatbots are so lifeless and vague. Because LLMs essentially work by looking at massive sets of data relationships and analysing how they connect to one another, they often spew out statements that sound perfectly reasonable but are actually inaccurate, absurd, or simply strange. This reduction of language to a dataset is also why, for example, when I asked ChatGPT to write my biography, it claimed I was born in India, studied at Carleton University, and received a degree in journalism—and was wrong on all three counts (the correct facts: Great Britain, the University of York, and English philology). In ChatGPT's understanding, the form of the answer and a convincing formulation are more important than the content: correct relationships matter more than the correct answer.

However, the idea of the LLM—a repository that stores meanings which are then variously combined—aligns with some 20th-century philosophical claims about how humans think, perceive the world, and create art. The French philosopher Jacques Derrida, further developing the work of linguist Ferdinand de Saussure, suggested that meaning is differential: the meaning of every word depends on the meanings of other words. Imagine a dictionary: the meanings of words can only be explained by other words. What is missing is always some "objective" meaning that would exist outside this endless chain of signification and finally bring it to a halt. But we are forever stuck, continuing to circle in this loop of difference. Some thinkers, such as Russian literary scholar Vladimir Propp, argued that folklore narratives could be broken down into their constituent structural elements, as set out in Propp’s groundbreaking work "Morphology of the Folktale". Of course, this does not apply to all narratives, but it becomes understandable how one could combine plot components—initial action, crisis, resolution, and so on—to then create a story about a thinking and feeling cloud.

The view that computers are capable of thought is quickly beginning to seem increasingly viable. Rafael Millière is an assistant professor at Macquarie University in Sydney and has devoted his entire career so far to consciousness and, more recently, artificial intelligence. He has described his involvement in a large collaborative project called the BIG-bench test. In it, AI models are deliberately set tasks that go beyond their "normal" abilities to see how fast they "learn". So far, the level of AI has been tested, for instance, by having it predict the next move in a chess game, act in a role-played trial, and combine various concepts.

Today, AI can take previously unconnected, even randomly selected things—say, the skyline of Toronto and the style of Impressionist painting—and combine them, creating something that did not exist before. But this leads to a conclusion that makes one feel a little uneasy or unsettled. Is this not, in some way, how we ourselves think? Millière says that, say, we know what a pet is (a creature we keep in our homes) and we also know what a fish is (a living being that swims in large bodies of water); we then combine these two things, keeping some features and discarding others, and arrive at a new concept: a house fish. The latest AI models are endowed with this capability—combining to form a seemingly new concept—and that is exactly why they are called generative models.

Even relatively complex arguments can be interpreted this way. The problem of theodicy has been an inexhaustible topic of theological discussion for centuries. The question it poses is this: if an absolutely good God is omniscient, omnipotent, and omnipresent, how is it possible that there is evil in the world, even though God knows it will happen and is able to stop it? This is a radically oversimplified approach to this theological problem, but theodicy is also, in a way, a logic problem, a set of ideas that can be recombined in a certain way. I do not mean to say that AI can solve our fundamental epistemological or philosophical questions, but it does suggest that the line we can draw between thinking beings and pattern-recognition machines is by no means as clear and bright as we might have hoped.

The feeling that an AI chatbot is based on a thinking being is also fuelled by the now generally accepted view that we do not know exactly how AI systems work. The so-called black box problem is often presented as mysticism: robots have surpassed us so far, or are so alien to us, that they are doing something we do not understand. While true, it is not quite in the sense one might think. New York University professor Leif Weatherby says that the models process so many permutations of data that it is impossible for a single human to grasp them. AI mysticism is not some unseen or unfathomable mind remaining hidden from us; it is related to scale and power.

However, even if we see the difference, recognising that AI is capable of using language only with the help of computational power, an interesting question remains unanswered: what does it mean to think at all? University of York professor Kristin Andrews, who studies animal cognition, points out that there are many cognitive tasks—remembering how to get food, recognising objects or other beings—that animals manage even if they are not self-aware. In that sense, one could quite well attribute intelligence to AI, as it is capable of engaging in what we usually call cognition. But, as Andrews notes, there is no evidence that AI possesses identity, will, or desires.

So much of what creates will and desire resides in the body—not just in the obvious sense, like erotic desire, but also as a more complex connection between inner subjectivity, our unconscious, and how we as a body move through the world, processing information and reacting to it. According to Zebrowski, it could be argued that "the body is important to how we can think, why we think, and what we think about". She adds that "you cannot simply take a computer program, stick it in a robot's head, and get an embodied being". It is entirely possible that computers are already close to what we call thinking, but they do not dream, want, or desire, and that is much more important than AI proponents would like to admit, not only in terms of why we think but also in terms of what thoughts we arrive at. When we use our intellect to seek solutions to an economic crisis or combat racism, we are guided by our ideas of morality or a sense of duty towards those around us and our descendants—the sense cultivated within us that it is our duty to improve things in some morally significant way.

Therefore, it is possible that the computer in Clarke’s story, a kind of shortcut to transcendence or absolute knowledge, is the wrong example. Douglas Adams’s supercomputer in "The Hitchhiker’s Guide to the Galaxy" might be closer to the truth. When asked the ultimate question about "life, the universe, and everything", the supercomputer, of course, spits out its famous enigmatic answer: "42."

This absurd humour is completely self-sufficient, but the funny answer also reminds us of a truth that is easy to forget. Life and its meaning cannot be reduced to a simple statement or a list of words, just as human thinking and feelings cannot be reduced to a formula expressed through ones and zeros. If we reach the point of asking AI a question about the meaning of life, the answer is not what is flawed. The question itself is flawed. And at this particular point in history, it seems it would be worth thinking about what exactly is making us seek answers from a benevolent, omniscient digital god, who is likely neither.

In March of this year, I spent two days at the Microsoft headquarters near Seattle. Microsoft is one of those leading tech players that has bet everything on AI. To prove how much, they had invited journalists from all over the world for a "tour" of their "innovation campus"; the event offered a dizzying programme of lectures and demonstrations, meals at the seemingly endless range of campus restaurants, and a couple of nights in a level of hotel that journalists cannot usually afford.

We were taken through a data centre, having been issued earplugs beforehand to filter out the hum of a small field of fans. We listened to several expert presentations on how their teams are integrating AI and how specialists working on "responsible AI" are tasked with keeping the technology in check before it goes off the rails. There was plenty of chatter about how this work will affect everything that happens in the future. In one session, the likeable Seth Juarez, principal programme manager for AI platforms, compared the introduction of AI to the transition from a shovel to a tractor: in his words, it "will allow humanity to move to a new level".

Some of what was seen there seemed truly inspiring, including a presentation led by Saqib Shaikh, who is blind and has been working for years on the Seeing AI project. It is an app that is increasingly capable of defining objects in the field of view in real time. Point it at a desk with a tin can, and it will announce: "A red drink can on a green desk." There was also optimism about the idea that AI could be used to save endangered languages, more accurately search for tumours, or more effectively predict where to place resources for disaster relief—mostly by processing large volumes of data and then recognising and analysing patterns within them.

However, despite all the pretentious talk about what AI might one day do, the impression was that for now, it handles trivial things best—reconciling numbers in financial reports, transcribing and summarising meeting content, sorting emails more effectively. Such an emphasis on everyday duties suggests that AI is unlikely to create some miraculous new world, but rather, depending on your perspective, will make existing processes slightly more efficient or perhaps intensify and solidify today's structures. Yes, some of your duties might become easier to complete, but it is even more likely that these automated tasks will simply be part of a new, even larger volume of work.

It is true that AI’s ability to simultaneously process millions of factors can more than compensate for the human skill of analysing certain types of problems, especially those where the factors involved can be reduced to data. After the Microsoft expert lectures on AI research, each of us was offered a book titled "AI for Good", which explores potential altruistic AI applications in more detail. Among the projects discussed was the use of machine learning to analyse wildlife observations collected via sound or satellites, as well as the potential to predict where it is best to place solar panels in India.

This is encouraging and, especially now, in the 2020s, allows for a momentary glimmer of relief or hope that something might improve after all. But the problems hindering, say, the use of solar energy in India are not just caused by a lack of knowledge. They are tied to resources, good will, deep-seated vested interests, and, to put it simply, money. This is an aspect that utopian future models so often leave ignored: if and when changes happen, we will have to address questions of whether and how a certain technology will be distributed, deployed, and adopted. Crucial will be how governments decide to allocate resources, how the interests of the involved parties will be balanced, how the idea will be presented and propagated, and so on. In short, everything will be decided by political will, resources, and rivalry between competing ideologies and interests. The problems facing Canada or the world at large—not just climate change, but also the housing crisis, the toxic drug crisis, or the growing anti-immigrant sentiment—have not been caused by a lack of intelligence or computational power. In some cases, the solutions to problems, viewed superficially, are very simple. For instance, the homelessness crisis would diminish if more cheap housing were available. But the solutions are difficult to realise because of social and political forces, not because of a lack of understanding, thinking, or innovation. In other words, progress on these issues will be hindered by the very same thing that hinders everything else—ourselves.

The notion of exponentially growing intelligence, so dear to Big Tech, is a strange fantasy that abstracts intellect, turning it into a kind of superpower that can only multiply; problem-solving is viewed as marks on a scale along which one can move ever further. This understanding is what is known as technological solutionism; the term was coined 10 years ago by Evgeny Morozov, a progressive Belarusian author who has made it his mission to provide a scathing critique of Big Tech. He was one of the first to point out the Silicon Valley tendency to view technology as the solution to all problems.

Some Silicon Valley entrepreneurs have taken technological solutionism to the extreme. The ideas of these AI accelerationist preachers are the most terrifying of all. Marc Andreessen was directly involved in the development of the first web browsers and is now a billionaire, a venture capitalist who believes his calling is to fight the "woke mind virus" and promote general capitalism and libertarianism. In his manifesto "The Techno-Optimist Manifesto" published last year, Andreessen laid out his conviction that "there is no material problem, whether created by nature or technology, that cannot be solved with even more technology".

When historian and journalist Rick Perlstein attended a dinner at Andreessen's 34-million-dollar house in California, he encountered a group of people there who passionately oppose any rules or other kinds of restrictions that would control the development of technology (at the end of 2023, in a tweet, Andreessen called the regulation of AI the "new foundations of totalitarianism"). When Perlstein recounted this to a colleague, he "noted a certain similarity to one of his students, who categorically claims that all the centuries-old problems that historians worry about will, self-evidently, be solved quite soon by better computers, and therefore considers all this humanist fussing to be a bit ridiculous".

Andreessen’s manifesto also includes an absolutely normal and not in the least threatening section in which he lists all possible enemies. Among them are all the usual right-wing bogeymen: regulation, know-it-all scientists, restrictions on "innovation", and the progressive activists themselves. In the venture capitalist’s perception, these are self-evident evils. Since 2018, Andreessen has served on the Facebook/Meta board, and this company has allowed democratic institutions to be attacked through deception and disinformation. Yet he claims—seemingly without the slightest irony—that it is the experts who "play God by interfering in other people’s lives, all while remaining completely insulated from the consequences".

There is a generally accepted view that technology is a tool. You have a task that needs doing, and technology helps you get it done. But there are significant technologies—a roof over your head, the printing press, the atomic bomb or a rocket, the internet—that have almost recreated the world, thereby changing something in how we perceive ourselves and the reality around us. This is not simple evolution. When the book appeared—and with it the ability to document complex knowledge and spread information beyond its previous limits of accessibility—the foundations of reality itself changed.

AI occupies a strange position: it likely belongs to these technological transformations, but at the same time, its role is significantly overstated. The idea that AI will lead us to some great utopia is deeply questionable. Technology can indeed turn a field like a plough, exposing new ground, but what was in the soil before will not just disappear.

After speaking with experts, I have the impression that AI will be a great assistant that will process data existing in volumes that humans simply cannot work with. Pattern-recognition machines, applied in biology or physics, will likely yield exciting and useful results. Other potential AI applications seem more mundane, at least for now: it can increase work efficiency, streamline certain aspects of content creation, and facilitate access to simple things like travel itineraries or text summaries.

But this does not mean that AI will bring only benefit in every respect. An AI model can be trained on billions of data units, but it cannot tell whether any of it is good or in any way valuable to us, and there is no reason to think that will change. We arrive at moral judgements not by solving logic problems, but by taking into account everything in us that is irreducible—subjectivity, self-respect, inner life, desires—everything AI lacks.

If anyone says that AI will ever be able to independently create art, they misunderstand the reasons why we turn to the world of aesthetics at all. We yearn for things created by humans because we care what a human says and feels about their experience, gained from living as a person and as a body in the world.

There is also the question of quantity. By tearing down the barriers to content creation, AI will simultaneously flood the world with complete rubbish. Google is already becoming increasingly unusable—precisely because the internet is clogged with AI-generated content designed with one goal: to collect clicks. Here we encounter a problem within a problem: digital technology has created a world that is so full of data and so complex that in some cases we now need technology to sift through it all. It is very likely that whether you consider this magic circle a curse or a blessing depends on whether you belong to those who derive some benefit from it or to those who have to dig through this mess.

But there is also cause for concern regarding AI’s integration into already existing systems. As Damien Williams, a professor at the University of North Carolina at Charlotte, pointed out to me in conversation, models assimilate massive masses of data during training that are based on what currently exists and what existed before. Therefore, it is difficult for them to avoid the biases and partiality that exist in the past and present. Williams points out: if AI were asked to depict, for example, a doctor shouting at a nurse, in its execution the doctor would certainly be a man and the nurse a woman. Last year, when Google hastily launched Gemini, which is meant to compete with other companies’ AI chatbots, it introduced "racial diversity" into its images of Nazis and American founders. The cause of this strange error was a clumsy attempt to preemptively address the problem of bias in training data. AI relies on what has been, and it seems that attempts to take into account all those countless ways in which we encounter past biases and react to them are simply beyond its power.

The structural problem related to bias has existed for quite some time. Algorithms have already been used in fields such as credit scoring, and the application of AI in, say, recruitment repeatedly demonstrates a biased leaning. In both cases, previously existing racial biases have clearly manifested in digital systems. So, often the real problem is precisely such issues, rather than the old cliché—an AI system gone out of control that arbitrarily launches nuclear missiles. However, that does not mean that AI will not kill us. Quite recently, it became known that Israel, in attacking targets in Palestine, used an AI version called "Lavender". The system’s task is to label members of Hamas and Palestinian Islamic Jihad and then indicate their locations as potential air strike targets—including these people's homes. As reported by +972 Magazine, civilians have died in many such attacks.

AI as such does not threaten us as a machine or system that will suddenly slaughter humanity. But the assumption that AI is truly intelligent encourages us to hand over a wide range of social and political functions to computer software; not only the technology itself but also its specific logic and ethos, its libertarian-capitalist ideology, are seamlessly embedded into our daily lives. So the question is: for what purposes is AI being used, in what context, and within what boundaries? "Can AI be used to let cars drive themselves?" That is an interesting question. But other questions are more important: should we allow self-driving cars on the roads, under what conditions, within what systems should they be embedded; or should we perhaps strip the car of its priority status altogether? And no AI system can answer these questions for us.

From the tech-industry type who links his hopes for human progress to superhuman intellect, to the militarist who relies on an AI system when choosing targets—all demonstrate the same longing for some kind of objective authority to which they could appeal. When we turn to artificial intelligence, hoping it will help us understand the world, when we ask it questions about reality and history, or expect it to depict the world as it is—have we not already become entangled in AI’s logic? We are drowning in digital rubbish, in today’s cacophony, and our response to it is to seek some superhuman assistant that would fish grains of truth out of a mire of falsehoods and errors. But we are so often led astray when AI makes mistakes.

Perhaps the determining factor that creates this urge to hear an objective voice where there is none is the fact that our meaning-seeking apparatuses have been undermined beforehand in ways that are not unique to AI alone. The internet has also been a destabilising force, and AI threatens to make everything even more complicated. Conversing with University of Vermont professor Todd McGowan, who works in the fields of film theory, philosophy, and psychoanalysis, I suddenly realised that our relationship with AI is essentially explained by the desire to overcome this destabilisation.

We live in a time when truth is unstable, changeable, and constantly disputed. Consider just the turn towards conspiracy theories, the rise of the anti-vaccination movement, or the emergence of racist pseudoscience to the fore. Each era has its losses: for modernism, it was a balanced and coherent self; for postmodernism, the stability of master narratives; and now, in the 21st century, there is increasing pressure on the notion of a shared view of reality. At the same time, public figures, from politicians to celebrities and publicly known intellectuals, seem more than ever subject to the lure of fame, suffer from an ideologically narrow view of the world, and preach obviously false ideas.

McGowan says the missing element is what psychoanalyst and thinker Jacques Lacan called "the subject presumed to know". In society, there should be plenty of persons presumed to know—teachers, clergy, political leaders, experts—and they should all function as authorities that provide stability to structures of meaning and ways of thinking. But when the systems that give things concrete shape begin to vanish or are challenged, as has happened with religion, liberalism, democracy, and many others, humans are left to seek a new god. Something especially poignant can be sensed in our desire to ask ChatGPT to tell us something about this world, where sometimes the feeling creeps in that nothing is true. For people filled with subjectivity, AI embodies the transcendent—the impossibly logical mind that can tell us the truth. In Clarke’s short story about the Tibetan monks, a similar notion of technology as something that allows us, simple mortals, to overcome our limitations can be sensed somewhere in the distance.

But the outcome of this overcoming is the end of everything. By enlisting technology to make their deeply spiritual, manually, and carefully performed work more efficient, Clarke’s characters erase the very act of faith that gave strength to their path towards transcendence. Here, in the real world, our goal is probably not to meet God. The goal is the torment and ecstasy we experience in these strivings. Artificial intelligence will continue to increase in scale, power, and capability, but the assumptions underlying our faith that it could, so to speak, allow us to get closer to God can only take us even further away from him.

In 10 or 20 years, AI will undoubtedly have developed further than it has at present. Most likely, its problems with inaccuracies, hallucinations, or bias will not have been solved, but perhaps it will finally write a sensible essay. And yet, if I am lucky and am still around, I will go out of the house armed with an AI assistant that will whisper in my ear. There will still be cracks in the pavement. The city where I live will still be under construction. Traffic will likely still be chaotic—even if cars drive themselves. Perhaps I will look around or lift my eyes to the sky and my AI assistant will tell me what I see. But everything will still be proceeding just a little differently than it is now. And the stars? Despite all the changes, which for the moment may seem so huge, the sky will still be as though strewn with them.

© The Walrus, 29 May 2024
Article published in the July 2024 issue of Rīgas Laiks magazine