Aristotle and A.I.
Aristotle and A.I.
Ancient wisdom and future technology
In the year 350 BCE, Aristotle wrote:
Let us begin by discussing the relation of human and artificial intelligence, in the hope that we may be able to obtain something better than commonplace confusions. For some people think that it's good for artificial intelligence to do our work for us; but others think that it is inappropriate, because that confuses proper distinctions between the tasks best done by people, and the tasks best done by machines.
We do our work with the tools we have. Some of these tools are things, like a car or a hammer or an e-mail server. And some of these tools are people, like the auto mechanic or the carpenter or the programmer. But artificial intelligence is unlike all of these; it does its work merely by being ordered to do so, without a human mind behind it. And that suggests a new kind of life for human beings, one of mastery without any corresponding servility.
By now you may suspect that I made this passage up. But I did not; instead, I have paraphrased text from Aristotle's Politics (book 1, 1253b). (I must admit my paraphrase is very loose indeed.) In the original, Aristotle is considering the relationship between masters and slaves. As a thought-experiment, he imagines what would happen if the fabled “statues of Daedalus” were real: humanoid statues, without minds or feelings, but that nonetheless could move and work if ordered to do so. For him it was a matter of idle speculation, because in his time, such wonders only existed in the imagination.
But ours is a time of wonders, where machines do move and work without a human mind behind them. Indeed, some even seem to think.
OpenAI was founded in 2015, and it released the fourth iteration of its large language model AI, ChatGPT-4, in 2023. That version could already carry out natural conversations, solve complex math problems, and write, both prose and computer code, at least as well as some humans. In September 2024 OpenAI released ChatGPT-4o, the latest iteration of its model, capable (we are told) of rudimentary reasoning. Given the rapid speed of AI improvement, it seems possible at least that an AI capable of doing almost everything as well as an average human, an Artificial General Intelligence, may appear soon.
For thoughtful people, these developments inspire wonder, but also unease. Leave aside the obvious fear that AI might pose existential risk to humanity. More subtly, AI could simply dispense with humanity, or most of it. A sufficiently-advanced AI could carry out tasks, and manage projects, better than we could, such that we give up our freedom of action; or do all of our thinking for us, such that we give up our freedom of thought, becoming idlers or cattle.
It is not obvious that these outcomes are likely, or even possible. But stipulating that they are, should we be concerned? After all, there are already large swathes of human activity that we have automated away, and are happy to have done so. Courtesy of the Industrial Revolution, more people enjoy more prosperity and health and safety today than ever before. Nor does anyone propose that dishwashers or clothes dryers have fundamentally eroded human dignity by depriving us of noble toil. Perhaps the advent of powerful AI is only the latest expression of the Industrial Revolution, and will ease our burdens without diminishing our nature. Or perhaps not, and AI does pose a serious threat to our dignity. How are we to know?
Put another way, the question before us is what do human beings need to do to flourish? What are the activities that are fundamentally human?
If we knew the answer, then we would have a clear course of action before us, or rather, we would have a clear division of labour. We would have a comparative advantage for the twenty-first century: humans, and only humans, should do the things that are best and most properly human. And AI, to the extent possible, should do everything else.
This is why Aristotle’s views on the matter are relevant. Framing the question this way suggests that of all the classical thinkers, on this matter he will be the most helpful. In the Nicomachean Ethics, he addresses the subject directly: what is the purpose of life? and what is the art of living that will help or hinder us in achieving that purpose?
In Aristotle's view, the purpose of human life is to flourish, to exercise the capacities that are particularly human, and to do so well; a state he referred to as εὐδαιμονία, or eudaimonia. Aristotle has a great deal to say about the specifics of eudaimonia, much of which we must set aside: Athens in the 4th century BCE and Oxford in the 21st century CE are so fundamentally different that appropriate behaviour in one would be disastrous in the other. But if we read Aristotle generously, we will see that there are some qualities that make up eudaimonia that we can abstract to fit our times. In doing so, we can build a framework to help us understand what things we should and should not want AI to do for us.
What, according to Aristotle, are the elements of human flourishing, and what is the promise and pitfall of AI in each regard?
The first is emotional control. Part of being fully human is to develop our minds such that we feel things in an appropriate amount. We should not be like children, who feel things too much, and are often overcome by sorrow, anger, fear, and even joy. Nor should we be like Star Trek's Vulcans, who feel too little, and as such become remote from the world and from each other. Finding the appropriate balance, the "golden mean", between extremes is key to Aristotle's ethics. It is vitally important because, for Aristotle, our emotions are not merely internal states, they are one of the motors that drive external action. By developing our emotional control, we will neither be capsized by our emotions nor isolated from them, and so will act appropriately in the world.
On this view, we should approve of AI that can help us to understand and express our emotions in healthy ways. We might imagine systems that provide guidance on remaining mindful or calm, especially in the moments when we are at risk of being carried away by anger. Other AI might help us to spot and reframe negative or unbalanced thoughts. Especially when paired with wearable sensors and other smart technology, such AI could be helpful in our self-training. Similarly, this sort of AI could be a useful tool to therapists, social workers, and psychologists, to help them and their clients build up mental health.
But since it is in dealing with others that we learn and practice emotional self-regulation, we should disapprove of AI that replaces engagement with other people. Such AI already exists in the form of virtual therapists and companions (AI ‘boyfriends’ and ‘girlfriends’). Healthy therapy and companionship depends on the mutual engagement of human minds and desires. This has been known in psychotherapy for some time; research has shown that it is the therapeutic relationship or ‘alliance’ between therapist and patient that drives outcomes. The technique the therapist uses is less important than the person of the therapist. As with therapy, so with companionship: a relationship will help us to grow only if there is another person involved. A simulacrum of a person cannot be allied with, and can only give us what we think we want, not what we need. And so all of these instances of AI are, on Aristotle's account, suspect. Resorting to them risks stunting our emotional growth.
Emotional control is a prerequisite for another element that Aristotle thought fundamental to human flourishing, which is social engagement. He begins the Politics with an observation that humans are social (or "political") animals. We do not need him to point out that spending time with other people is important for our psychological health, but there is more to the case than that. We are members of families, of communities, and nations. We have, and are part of, business interests and other organizations; of religious bodies; of clubs, and fandoms, and political parties. In all of these bodies we have roles to play. For Aristotle, these roles define, at least partially, who we are. We need to know what parts we are to play, and to play them well.
On this view, we should approve of AI assistants that can help us to do these things. AI that can coordinate gatherings, facilitate conversations, help meetings to be productive, alert us to our responsibilities and help us to carry them out; all of these things should be welcomed. But we should disapprove of AI that removes the need for social connection. AI that draws us into private worlds is suspect. This is obviously true of virtual companions, but there are more subtle pitfalls. Many are enthusiastic about AI storytelling that produces bespoke experiences: a novel that is prepared just for you, or a film, or a video game. Experiences like these cannot be shared, and thus deprive us of connection for others. We should aim instead for AI to enrich our appreciation for the experiences that we can share.
We must continue to develop the capacity to reason well.
The final element of human flourishing is reasoning. For Aristotle reasoning was the fundamental human activity, the quality that distinguishes us from plants and animals. To reason well, we need to develop the intellectual virtues: imagination, practicality, intuition, and judgment. We need to learn how to assess an argument's soundness and validity. And we need to be courageous enough to follow the truth even when it is inconvenient: to loosely paraphrase Aristotle once more, "Plato is my friend but the truth is a better friend to keep" (Nicomachean Ethics 1096A).
From the perspective of the twenty-first century, the importance of reasoning well cannot be overstated. For Aristotle, reasoning was the exercise of the most human capacity, the ability to help us determine what the good life is and equip us to live it. But we need not share his views to agree on this point. Unlike him, we are heirs to the Enlightenment and the Industrial Revolution, and the incredible health, wealth, safety, and comfort that pluralism, democracy, and industrial production have given us. We are the beneficiaries of progress, which stems from human reasoning, as expressed in philosophy, science, and technology. To maintain what we have, and extend it into the future, we must continue to develop the capacity to reason well.
On this view, we should be very excited indeed about AI tutors that can help us master new material more quickly and easily, and AI research assistants that can help us brainstorm, generate hypotheses and experimental plans, and especially AI tools that can help us solve problems we could not otherwise address. AlphaFold is an example of the latter: protein folding is critical to biomedical research, and with AI we can do more in this realm than we thought possible before.
Conversely, we should be equally skeptical of AI that does our thinking for us. By this I don't mean AI that solves problems we set for it, or we'd have to be equally skeptical of calculators or spreadsheet programs. I mean AI that alters our relationship with reasoning itself. We have already seen AI-powered systems deployed to guide decision-making in fields as diverse as candidate hiring for large corporations, and target acquisition in warfare. It will take longer for AI to penetrate more directly-political realms like criminal justice or policymaking, but its advent is certain. By processing data and identifying patterns beyond the ability of any human, AI will generate immense temptation for users to bypass their own judgment in favour of the machine's. Indeed, this has already begun to happen. Precisely because these temptations are so strong, we should temper our enthusiasm for any AI system that might permit us to delegate our reasoning, and especially our decision-making, to it.
All of this is to say that we should embrace AI which enhances our autonomy and promotes human flourishing, and resist AI that detracts from these goals. Put another way, we should aspire to a humanist AI.
A humanist AI, aligned with Aristotelian principles, would support, but not replace, the necessary role of humans in the exercise of emotional regulation, social engagement, and reasoning. Humanist AI will help us to feel our emotions fully without being at their mercy; it will help us spend time with each other, so we can better carry out our responsibilities and enjoy sociability; and most importantly, it will help us sharpen our own ability to reason. It will improve our abilities to act and think, and not encourage their atrophy.
As we assess new AI tools, we should ask in which of these directions it will pull us. Use of AI that improves our human capacities should be promoted. Use of AI that encourages division or solitude should be deplored. And most importantly, use of AI that replaces human decision-making and judgment must be forbidden.
To do so will require us to strike a delicate balance, employing both the vast capabilities of AI but also our own creativity, empathy, and reasoning. If we do, we will then have tools with unprecedented ability to erase drudgery while remaining ourselves free to think and act. Put another way, we will have achieved mastery without servility, a state Aristotle could only imagine as a fable, a state he described as resembling the divine.
Off-Ramps
Thinking too much about elections erodes good governance:
Thanks for reading Changing Lanes! Please let us know how we’re doing by answering the poll below. And if you’d like to respond to this post, please leave a comment.