Political philosophy in the age of artificial intelligence

What can Socrates teach us about knowledge and politics in an era dominated by the science and technology of information?

Artificial intelligence seems to stand apart from every other field of study today, operating on the most exciting and dangerous frontiers of man's technological capability. Because of this unique position, A.I. research garners reactions ranging from unbridled optimism to catastrophic pessimism. Pessimists like Eliezer Yudkowski think efforts at building artificial general intelligence are so risky as to be practically suicidal. Optimists like Marc Andreessen, on the other hand, see in A.I. a utopian future of unimaginable wealth, productivity, and knowledge.

Setting optimism and pessimism aside, few people deny the exciting and fearsome character of the current generation of tools. Such is the excitement and novelty that it becomes tempting to look further and further ahead to the most extreme possible outcomes without ever thinking to look to the past. That is, it is all too tempting to believe that there is no room for old teachings to inform and moderate the A.I. discussion.

Yet a careful reader can find in ancient texts a startling degree of insight into the patterns emerging from the field of A.I. research and development. For the student of classical political philosophy, the headlines, whitepapers, and dramas unfolding across the A.I. landscape can yield to what might be characterized as a rediscovery of essential insights that Socrates, himself, brought to light in ancient Athens—insights which might give rise to a cautious and inquisitive moderation.

In order to investigate those insights, one must start with a spare and provisional definition of artificial intelligence as it exists today, in the form of machine learning.

What is machine learning?

It would be an error to speculate about what artificial general intelligence might entail, because such a thing does not exist today. The A.I. tools that people do encounter today, e.g., large language models and image generators, fall into a category of technologies known as deep learning. Deep learning is a subset of the broader field of machine learning. Machine learning most often refers to a category of statistical techniques that use various kinds of neural networks to manipulate information from vast data sets in such a way that general patterns can be detected and used to respond accurately to novel prompts.

When speaking of machine learning, it is all too tempting to anthropomorphize, in part because the design is loosely inspired by the human brain, but also because the surface-level interactions people have with machine learning technologies are meant to feel like interactions one would have with a person. For example, one can ask ChatGPT, in plain English, “What were the circumstances which led to the Thirty Years’ War?” and receive a plausible answer that might receive passing marks in a high school exam on the topic.

Behind the facade of human-like interaction, though, the fact remains that machine learning tools are fundamentally computers running software programs to manipulate information into a useful form. In the case of machine learning, the software involves transforming large amounts of data in service of two broad goals. The first goal is to “train” the neural network to generalize relevant patterns from the data it is given. For a relatively crude model that identifies letters and numbers in images, this might involve a human providing millions of images of written text, each of which is paired with the letter or number it depicts. At first the computer will do nothing but guess letters and numbers at random. But the software is designed to use the human-provided descriptions to correct its predictions, thereby “training” it to accurately detect the correct letters and numbers. The second goal is to use the now-trained model to respond to novel prompts. Following the previous example, a well-trained trained letter-detecting model can be given images it has not seen before, and respond with an accurate answer.

There is, of course, baked into this description a certain skepticism about the relationship between the machine learning neural network’s function and the function of the human brain. That is intentional, because it remains a fact that we simply do not know how the human brain works. Honest neuroscientists will readily admit that the field is “pre-paradigmatic,” meaning that it is still waiting on a fundamental breakthrough. Certainly, a great many neuroscientists and researchers would seek to draw stronger connections between human cognition and contemporary artificial intelligence work. But holding such a view commits one to the principle error of bad philosophy—to lack adequate awareness of what one does not know, and thus to commit one’s intellectual strivings to a delusion.

The most prominent philosopher to declare that he knew what he did not know was Socrates, to whom we now turn for a classical view of human understanding through philosophy.

Socrates on human knowledge

Leo Strauss dedicates the opening paragraphs of Chapter IV of Natural Right and History to a discussion of Socrates and the tradition of political philosophy he founded. According to Strauss, Socrates distinguished himself from the philosophers who came before him by his characteristic combination of wisdom and moderation, in contradistinction to their “madness.” Whereas they expounded theories about the workings of the cosmos, Socrates sought to situate human knowledge within the horizon of experiences that are natural for man in common life.

Socrates teaches that man does not directly experience the absolute truth of things, but rather experiences a variety of phenomena that comprise his world, including a direct common-sense intuition about the various things he encounters, as well as the opinions about those same things which he hears and shares among the people in his community. Strauss writes:

Socrates started not from what is first in itself or first by nature but from what is first for us, from what comes to sight first, from the phenomena. But the being of things, their What, comes first to sight, not in what we see of them, but in what is said about them or in opinions about them. (124)

Starting from one’s own common sense, and the common sense opinions of others, is crucial for Socrates because the alternative presents a tempting delusion, which is to become obsessed with universal and homogeneous abstractions. Thinkers of all ages have been drawn in by the allure of such abstract, systematic thinking—e.g. everything is water; or everything is information; or everything is atoms—because it allows the thinker to simplify and more easily draw rational conclusions. Abstractions are friendly, useful, and easy to manipulate. But they purchase their usefulness at the cost of drawing man into a kind of delusion about the world, that the world is homogeneous and easily understood, rather than a complex and often contradictory composition of heterogeneous parts. More often than not, systems of abstract thinking do disclose something, but they inevitably conceal more than they disclose. Their gleaming purity can come to dominate one of the most important human faculties, which is perception in the mind’s eye of what is real for man qua man. As Strauss puts it, referring to man’s common sense opinions:

Socrates implied that disregarding the opinions about the natures of things would amount to abandoning the most important access to reality which we have, or the most important vestiges of the truth which are within our reach… opinions are thus seen to be fragments of the truth, soiled fragments of the pure truth. (124)

But, of course, Socrates was not satisfied with mere opinion, as opinions are almost always wrong. Having an opinion does not imply understanding. In order to understand, opinions have to become knowledge. Socrates was famous for asking his interlocutors confounding questions, questions which shook their faith in their own opinions, because he saw in that method of conversation the opportunity to refine opinions about a topic into genuine knowledge pertaining to a fundamental question. This conversational method is known as “Socratic dialectic.” Strauss describes Socratic dialectic as “the art of conversation or of friendly dispute” through which contradictory opinions are distilled into knowledge. No one opinion about a thing will ever be purely true, but each genuinely-held opinion contains “soiled fragments of the pure truth.” “Philosophy consists, therefore, in the ascent from opinions to knowledge or to the truth, in an ascent that may be said to be guided by opinions.” (124)

To sum it up, then, Socrates believed that genuine knowledge—knowledge pertaining to the fundamental questions that man as man faces in his direct experience of living—is only attainable by having friendly conversations with people whose common sense opinions contradict each other. If the right kind of person, i.e., a philosopher, guides such a friendly dispute, then the interlocutors’ false opinions can be vanquished, while preserving and refining their fragmentary truths.

Adherents of modern physical and social science will, no doubt, bristle at certain aspects of this description of an ascent from common sense opinion to genuine knowledge. In more recent times, people are habituated to think of knowledge as something that institutional science creates. And by institutional science, people tend to mean large organizations, like government and university research labs, carrying out a process that includes formulating hypotheses, collecting vast amounts of data, and using statistical methods to try to falsify the hypotheses. Anyone who enjoys the fruits of 21st century medicine, engineering, and the like must admit that modern scientific techniques have tremendous power.

But likewise it must be acknowledged that the entire edifice of science is built atop the same foundation of philosophy previously articulated. That is, science does not reject but rather builds on the idea that there is a natural whole, which is composed of parts, and which the mind’s eye can perceive, and about which the faculty of reason can make informed judgments. Science cannot even begin without a basis in philosophy’s earliest insights, a basis which science will not, and cannot, ever reject without becoming completely detached from reality. Strauss, again, explains beautifully:

All knowledge, however limited or “scientific,” presupposes a horizon, a comprehensive view within which knowledge is possible. All understanding presupposes a fundamental awareness of the whole: prior to any perception of particular things, the human soul must have had a vision of the ideas, of the articulated whole. (125)

Thus we are faced with yet another piece of the puzzle of human understanding, prior even to the common sense opinions, which is the “fundamental awareness of the whole.” It would seem that without the soul’s capacity to intuit the whole of being, there would be no way to form opinions, or even to perceive the parts that make up the whole. Those who speak of “the hard problem of consciousness,” are possibly speaking about something along these lines. And this, again, is why honest neuroscientists consider the field to be pre-paradigmatic.

In light of the Socratic view of human intelligence, through the ascent from common sense opinion to knowledge, we can return to questions about artificial intelligence.

Artificial intelligence in light of Socrates

To attempt a Socratic analysis of a contemporary technology is a fraught task that requires the utmost caution. Socrates did not directly experience ChatGPT, so it would be hubristic to claim to know what he would have thought and said about it. But given the Socratic view of human intelligence, we can ask whether or not computers seem to be capable of anything remotely similar.

The first major issue that computers seem to have is a fundamental lack of awareness. Certainly, computers have all manner ingresses for information in the form of devices like cameras, microphones, thermometers, and gyroscopes. And more than a few people with smart phones have suspected their devices of discreetly listening to them. But measuring various kinds of stimuli, no matter how surreptitiously, does not meet the standard for “awareness,” and comes nowhere near the “fundamental awareness of the whole” that humans exhibit. Unless a computer can exhibit such a capacity, it would seem that the full scope of human intellect is unavailable to a piece of software. After all, without this fundamental awareness, the entire process that follows cannot begin. Parts cannot be identified, opinions cannot be formed, and friendly disagreement cannot commence.

But perhaps a lack of awareness, alone, is not reason enough to write off the possibility of machine intelligence. After all, the reason that machine learning algorithms involve a “training” step pertains to precisely this problem. Computers don’t just miraculously wake up and start learning, so people have to train them. And the fact that computer scientists train models indicates that machines are capable of learning, i.e., the “learning” part of the name “machine learning.” Perhaps it is the case that after humans have helped the machine off the ground by supplying the data—the opinions borne of natural human awareness and intellect— the rest of the process can continue successfully all the way to knowledge and intelligence.

For the sake of argument, then, we can say that computers can be bootstrapped by humans with opinions in the form of very large databases, containing data collected from a great many people over a long period of time—a formidable set of opinions, indeed! These opinion-data still need to be transformed into knowledge, as evidenced by A.I. developers going through the hassle of training their models. Put another way, it is not enough to mount a hard drive containing a very large database to my computer. The computer does not yet “understand” anything by virtue of plugging in a hard drive. Something else must happen.

Giving the benefit of the doubt, we can acknowledge that training a neural network has, however loosely, a quasi-dialectical form. The computer says, “This is a cat,” and the model trainer says, “No, that is not a cat,” or “Yes, that is a cat.” Such a dialectical conversation is not exactly friendly, nor is it of the same literary caliber as The Republic, one must admit, but perhaps such critiques simply provide a high standard towards which ambitious A.I. researchers can strive.

Unfortunately, there is yet another problem that must be faced—namely, the faultiness of the equation of “data” with “opinions.” It may seem like splitting hairs, but the distinction is a crucial one. Data are inert pieces of information. A datum is not able to provide further explanation, nor is it able to frame itself in a context. With opinions, on the other hand, further explanation and framing within a context are available, by virtue of the fact that a human holds the opinion. That human can be asked a question. A person reports, “I don’t like anchovies because they are stinky.” His interlocutor replies, “You say you don’t like anchovies, but here you are at dinner eating a Caesar salad!” The person responds, “Oh, well, this is the one exception because romaine lettuce is so plain, otherwise, and at any rate, the texture masks the…” The datum “anchovies are stinky” is an artifact of an opinion, but its lifeless inability to say any more betrays that it is not itself the opinion in question. And if an opinion is to contain a soiled fragment of the truth, it surely must exist in its living and questionable human form, and not in lifeless and inert data.

At this point, among certain readers there will be a strong temptation to reject this style of inquiry altogether. Perhaps intelligence is something simpler, something more tangible, which can be measured, e.g., in bits of information processed per second. Why not soften the standard for intelligence in a way that permits the inclusion of computers? After all, what they are doing so often seems to be intelligent. But this would be a mistake. After all, when artificial intelligence researchers make claims about intelligence, they are necessarily using human intelligence as a standard. What else could the standard be? A lesser intelligence, like a dog? Or a greater intelligence, like a god? If that is the case, then the entire project is wasted on the error of believing that man can understand and even become a god, or that man is not and has never been more than a dog, in his intellectual capacities. Both ways are in obvious error.

In the end, one is left rather unsatisfied with the faculties of machines and rather impressed with Socrates. If the Socratic view of human intelligence and understanding presupposes a fundamental awareness, which gives rise to opinions that can be refined into knowledge through philosophic inquiry and friendly conversation, then an analogous view of machines would seem to presuppose a fundamental unawareness, which requires the manual collection and upload of many opinion-like data that can be used to forcibly generate novel opinion-like outputs through authoritarian commands. The computer can never rise above the status of a mechanical slave, following simple commands from its master.

Then again, it is always the case that a brand new and completely unpredictable scientific breakthrough can occur, which invalidates old paradigms and ushers in a brave new world of conscious and hyper-intelligent computers. No philosophic reasoning can ever close, once-and-for-all, the door to the purely unknown. But arguments over what cannot ever be known always descend into dogma competing against dogma.

As ever, then, we must return to the human question. What happens to man in an age characterized by massive edifices of information, the appearance of intelligent computers, and elite teams of researchers working to build these tools?

Man and politics in the information age

Even if we cannot see in computers today the possibility of human-like intelligence, we likewise cannot underestimate the potential for tremendous upheaval as a result of introducing new information technologies. Many good and bad things that were once considered impossible have become possible through tools like the Internet, cryptography, and deep learning. Those who understand and wield the cutting edge tools of any age have the power to shape the economic and material reality of that age, for better or worse.

But new information technologies disrupt much more than just economic and material reality—they can radically modify politics and psychology, again for better or worse, and in both intended and unintended ways. How man understands himself, as man, and how he sets about organizing society will change as a function of the tools available to him. If these changes are substantial, they will prove to be more fundamental and thus more important than any economic or material change.

There are a great many political and psychological changes already underway that one can identify in the information age, but there are two underappreciated developments that relate directly to both A.I. and political philosophy. The first change pertains to man’s ability to coalesce common sense opinion into knowledge. The second pertains to a rediscovery of ancient political wisdom through great technical feats.

The degradation of common sense

As people develop more sophisticated tools for synthetically replacing real life encounters—e.g. A.I. companions and virtual realities—those who use these tools will become more deluded about the actual reality they inhabit. Strauss reminds his readers that man refines his opinions into knowledge through philosophical conversation. The knowledge he gains pertains to the subject of his opinions. Opinions about home remedies for illness refine into knowledge about medicine. Opinions about what is right and wrong refine into knowledge about the problem of justice. So what happens if a substantial portion of man’s opinions, which used to be about reality as he experienced it, are replaced with opinions about a virtual reality of another man’s design?

As man’s attention comes to be ever more dominated by feeds, videos, metaverses, and more, he increasingly trades away the attention he pays to the whole of being. His opinions about political affairs in his town are replaced by opinions about the information he consumes about his town, or about far off places of which he knows nothing. His opinions about the the love and friendship he experiences are replaced by opinions about dating app profiles, or feelings towards A.I. companions. In general, the man who focuses his time and attention on simulated worlds risks replacing his opinions about the nature of reality—of things that are not man-made—with opinions about things that are man-made, and are thus lesser, illusory things. Any refinement of these lesser opinions will lead not to knowledge of nature, but to knowledge of the veil that man has pulled over his own eyes. He loses contact with reality, which is a suitable definition of delusion.

Then again, the critique of man losing contact with reality must be moderated. As Plato’s cave analogy shows us, man has always lived in a world of shadows on cave walls. And a wise mentor reminds me that technology can never entirely chase the world out the back door. There will always be some contact for man with the ground of his being. His pain and his pleasure, for example, can never be escaped entirely, or at least not in this life. But man does not need to lose contact completely for a revolution to occur. As Strauss famously wrote, there are caves beneath the cave which draw man ever further from the light. Perhaps the illusory technologies of the information age present man with the temptation to descend, yet again, into a deeper and darker cave.

We already see pathologies emerging which have their origin in an alienation from common sense, thanks to the popularization of the Internet, smart phones, and “data-driven” epistemology. Jon Askonas has written powerfully on the challenges of data-driven knowledge in his essay series entitled Reality: A Post-Mortem. In the third installment, What Was The Fact?, he charts the course of human knowledge from the pre-fact world of common sense, through the fact-based world of the scientific revolution, and all the way to today’s “post-fact” world of superabundant information. He writes:

If the temptation of the age of facts was to believe that the only things one could know were those that procedural reason or science validated, the temptation of the age of data is to believe that any coherent narrative path that can be charted through the data has a claim to truth, that alternative facts permit alternate realities.

As facts are replaced by data, and data come to refer circularly back to the domain of information, one cannot help but anticipate a destabilization of the ground of knowledge, itself. And once a majority of man’s own common sense opinions are directed away from his natural experience and towards the virtual world, his care for reality is subordinated to a fixation with unreality, and he risks becoming a lesser being, a slavish creature, essentially lacking both genuine knowledge and self-sufficiency.

The resurgence of the closed society

The second major upheaval that can be seen in artificial intelligence research pertains not to a loss of knowledge, but to a rediscovery of ancient wisdom about powerful and effective political structures. As elite cohorts of A.I. researchers iterate on their designs and techniques, and even their own organizational structures, in the service of achieving new breakthroughs, there has been a notable shift from an egalitarian premise, that the common good is ensured by open sourcing training data and models, to an aristocratic orientation that prefers to keep training data and models closed off from the outside world. The field of A.I. research is rediscovering the virtues of closed society.

The most prominent example of A.I. research shifting from the open to the closed society occurred in March of 2023 when OpenAI, the company behind ChatGPT, decided not to open source its latest model, GPT-4. In the GPT-4 technical report, the reason OpenAI provided for this radical shift related to concerns about both “the competitive landscape” and “the safety implications of large-scale models like GPT-4.” In an interview in April 2023, OpenAI’s co-founder and chief scientist, Ilya Sutskever, gave some remarks along similar lines, which I’ve lightly edited here for clarity and brevity:

A.I. comes with many different challenges, many different dangers, which come into conflict with each other. And I think the open-source versus closed-source is a great example of that. What are some reasons for which it is desirable to open source A.I.? The answer would be: to prevent the concentration of power in the hands of those who are building A.I. But if one believes that eventually A.I. is going to be unbelievably powerful, e.g., capable of autonomously opening a biological research lab, should this be open sourced also? When the capability is on the lower end, I think open sourcing is a great thing. But at some point the capability will become so vast that it will be obviously irresponsible to open source models.

One sees, in these remarks, a bit more than just a concern about safety. Sutskever is acknowledging that A.I. presents a complex web of conflicting concerns, of which some must exceed the dangers of concentrated power. He suggests that open egalitarian governance cannot solve the political problem of balancing these concerns. Rather, he thinks it would be better if the expert class of A.I. researchers acted like a nocturnal council for A.I. models, operating in behind closed doors to adjust the rules that govern them.

The decision to close off the code behind A.I. models is only part of the story. It is perhaps even more interesting to consider what happens to A.I. models, themselves, when they are built in the image of the open society, i.e., permitted to openly ingest data from the Internet. In an infamous episode, Microsoft unveiled an A.I. chat bot called Tay, which was premised on the idea that it would learn to engage with people by interacting with users on Twitter. Within hours, Tay had learned the art of Twitter discourse in the form of hurling obscenities and racial epithets. The technical lesson to learn from the implosion of Tay mirrors the political lesson to learn from the closing of OpenAI; namely, excellence often requires a closed society.

Millennia before Tay imploded and OpenAI closed its doors, the tradition of classical political philosophy taught the virtues of the closed society. Ancient thinkers wrestled with the problem of cities because cities exacerbate problems of justice and common good, but without the city man seems to be less than he could otherwise be. As Strauss put it, “Man cannot reach his perfection except in society or, more precisely, in civil society” (130). But civil society, “or the city as the classics conceived of it, is a closed society and is, in addition, what today would be called a ‘small society.’” (130). Why not a large, or even global, society?

For the ancients, the problem with political structures of a certain size was rather obvious. Beyond a small closed society, excellence became a practical impossibility. The chaos of increasing factions, dissipating care, and hyper-complex governance inevitably leads to the destruction of the highest ways of being. Strauss, again, writes:

For all precious things are exceedingly rare. An open or all-comprehensive society would consist of many societies which are on vastly different levels of political maturity, and the chances are overwhelming that the lower societies would drag down the higher ones. An open or all-comprehensive society will exist on a lower level of humanity than a closed society, which, through generations, has made a supreme effort toward human perfection. (131-2)

In Strauss’s words, we hear an ancient echo of precisely the same wisdom that artificial intelligence research is rediscovering today in the political element of the world of information. If they are too open, both A.I. companies and the models they make, will dissolve into vulgar and dangerous things. The harmony of these modern discoveries with ancient wisdom may strike us as anachronistic, but serves to strongly validate both the honesty and prowess of these modern efforts, and the ageless value of the ancient teachings.

In a strange and subtle way, the second upheaval mirrors the first. While the public, at large, stands to stupefy itself through delusion and degradation of self-sufficiency, the expert technological class stands to rediscover old and powerful political lessons. In this way, the organization and balance of political power could radically alter. It is not hard to see how “the many” becoming more dependent and less knowledgeable, coupled with “the few” coming to control the shadows on the walls of the cave ever more adeptly while simultaneously coming to grasp the virtue and power of the closed society, suggests the possible reemergence of political arrangements that moderns have worked very hard to suppress—either the rebirth of a virtuous aristocracy, or the resurgence of a vicious tyranny.

Subscribe to Niko Kovacevic

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.