AI and Individuality
A Millian View
Very short summary: This essay addresses, from a Millian perspective, the potential benefits and costs of AI technologies. While these technologies can broaden the base of our knowledge, they can also affect our ability to turn information into genuine understanding. A serious possibility is the emergence of a new cognitive aristocracy where a minority of individuals cognitively take advantage of the new opportunities offered by AI. From a Millian point of view, this may in turn undermine people’s individuality. That calls for a skeptical stance toward the experiment in living that is the development and adoption of AI technologies.
The rise of AI technologies has triggered two general and related types of concerns. The first – the one that is by far the most discussed in the media – is the economic impact of these technologies, especially on the job market. The second is what could be called the sociocultural effects that AI will have not only on the organization of human societies, but more generally on the way we live, communicate, and think. These concerns are clearly connected. A society where human labor becomes mostly unnecessary is not only facing the economic issue of guaranteeing people’s subsistence, it also has to rethink how to give meaning to their lives. It also works in the other direction; even if jobs are not all destroyed, their very nature may change significantly in a world where, for instance, people now find most of the information they are looking for or make decisions using AI technologies.
As with all technological revolutions, we are prone to focus on the immediate social costs generated by AI technologies while underestimating, mischaracterizing, or simply ignoring the short and long-term benefits. This is not only because many intellectuals and researchers (philosophers, economists) and, more surprisingly, the business actors developing and marketing those technologies keep on brandishing doomsday scenarios while, in reality, their judgments are based on no evidence and shallow theoretical analysis. As philosopher Carlo Cordasco argues in a recent Aeon essay, costs are more salient than benefits because we lack the linguistic and conceptual apparatus to identify and express the latter, while this apparatus is fine tuned to locate the former.
The identification and estimation of costs and benefits of a new technology (or of a policy or any structural change in society) always take place against a background of assumptions about how society works, what mechanisms trigger what kinds of effects, what is valuable, and so on. For many reasons, some of them maybe related to our cognition, we have no choice but to think ceteris paribus, that is, by keeping everything constant except for the variables that are directly affected by the technological change. This way of thinking is fine for incremental changes. Let’s say a government wants to know how demand will react to an increase in taxes on sweet food products. Here, keeping everything else equal, including people’s preferences (i.e., their “marginal rates of substitution”) is fine, at least in the short and mid-term. There is no reason to think that the policy will immediately affect people’s preferences or any structural variable that would make the underlying economic model used to predict the effect of the policy obsolete.
We will immediately see clear costs (e.g., the jobs that are destroyed) and maybe some benefits (e.g., the time gained by automating simple tasks like updating one’s agenda) but we’ll miss the deeper transformation that the technology will induce. What are those transformations? Nobody knows. Besides more or less plausible conjectures, we just lack the conceptual apparatus (and the imagination) to formulate with strong confidence predictions of the structural change that will take place.
Carlo’s Aeon essay nicely illustrates this point with how AI has changed his own activity as a researcher and a writer:
“What changed was that the cost of preliminary exploration collapsed. I could sketch an argument, identify the first serious objections, test whether they were fatal, and reach a provisional verdict in an afternoon rather than a fortnight. This sounds like a simple acceleration, and the more profound effect was on what I was willing to abandon. Dropping a question after an afternoon’s work feels nothing like dropping one after three weeks. When the exploration costs are low, the sunk cost attachment disappears, and you find yourself dropping bad questions earlier and more often, which means the questions you keep are better. I explored far more ideas, and my working portfolio became both larger and better curated. I arrived at this outcome not through any deliberate plan but simply through sustained engagement with a tool that changed what exploration cost.”
An unexpected benefit of using AI to do research and write, one that emerged through the practice rather than an ex ante prediction, is that it increases the “ability to find problems that are both tractable and important.” Interestingly, this is mostly a tacit form of knowledge (as Carlo notes, nobody teaches this kind of skills) that is only accessible to practitioners. No one else can really understand what that means beyond a purely abstract and superficial level, and even for practitioners that may not be obvious until they engage seriously with the technology in their work. There is no reason to think that this is the only example of this kind.
Now, arguably, AI technologies may also have hidden or non-obvious costs that we will only discover as we progressively incorporate them in our social practices. As Carlo observes,
“My ability to hold together a complex position verbally, under pressure, in a seminar or a conversation, has probably not improved and may have declined somewhat. When preliminary exploration is cheap, you spend less time grinding through arguments from first principles, a grinding that builds fluency that shows up in live exchange. Friends have pressed me on this, and they are right to worry. The shape of the disagreement is itself instructive, because the cost is immediately describable as a subtraction from a capacity I have been exercising for years, while the benefit was prospectively invisible and retrospectively obvious.”
Two observations are in order. First, insofar as AI technologies have potential adverse effects on capacities or skills that we know we have and are able to articulate within our current linguistic and conceptual apparatus, the related costs are salient in the same sense as above. But, second, consider how AI technologies may have deeper, more structural effects on how we come to turn information into knowledge by “appropriating” ideas that become “ours.” In the second chapter of On Liberty, John Stuart Mill develops at length the argument that freedom of thought and speech is warranted partly because it puts the human mind under the pressure to affirms its beliefs against competing views.[1] In the process, the individual is trained not only to affirm what it takes to be the Truth, but they are also “learning the grounds of one’s own opinions.”[2] As Mill puts it eloquently, one
“must feel the whole force of the difficulty which the true view of the subject has to encounter and dispose of; else he will never really possess himself of the portion of truth which meets and removes that difficulty. Ninety-nine in a hundred of what are called educated men are in this condition; even of those who can argue fluently for their opinions. Their conclusion may be true, but it might be false for anything they know; they have never thrown themselves into the mental position of those who think differently from them, and considered what such persons have to say; and consequently they do not, in any proper sense of the word, know the doctrine which they themselves profess.”
Let’s grant this Millian claim that there is something in the fact of knowing that goes beyond the mere behavioral capacity to state true statements. This points toward a hard-to-measure but nonetheless potentially significant social cost — a decline in people’s ability to understand arguments and ideas and use them in a deliberative (especially political) context. The concern is primarily not directed at academics like Carlo and me who, by using AI, may indeed tend to acquire a slightly more superficial understanding of complex issues. In this case, it’s plausible that there is a tradeoff between “genuine” understanding and knowledge broadness, as well as other considerations, and all things considered, AI technologies may be beneficial. The concern is more acute for individuals who have neither the incentives nor the habit of thinking hard about complex issues, starting with children at school.
“The Land of Cockaigne,” Pieter Bruegel the Elder (1567)
In a recent The Atlantic essay, David Brooks addresses this concern by drawing a typology of individuals depending on how they will cognitively deal with AI. What Brooks calls the “productive passengers” will use AI to reduce their cognitive burden. Because they have a low need for cognition, their productivity may increase, but using AI will undermine their mental capacities. The “reluctant optimizers” are more aware of and concerned by the cognitive risks of AI, but mostly will have no choice but to engage in an instrumental use of AI to reach goals that are imposed on them, especially in a professional context. Over the long run, they will become increasingly deferent to machines and less prone to effort. “Mental marathoners” have a high-need-for-cognition. They will continue to put on cognitive effort, like marathoners continue to run long distances though they could do it more efficiently with a car. Partially, they will do it for the sake of doing it, just because they take pleasure. But that’s also because, in a Millian fashion, they understand that they mustMental marathoners will use AI technologies, but they will do their best to maintain their independence of thought and their cognitive agency.
If Brooks’s typology sounds elitist, putting high-need-for-cognition people as those who are the most likely to master AI technologies and, therefore, as those who will dominate in AI-ruled societies, this is a feature, not a bug. Last year, I already alluded to the likely emergence of a new cognitive aristocracy. As I put it then,
“The problem with doomsayers who claim AI announces the end of thinking is that they generally only consider the aggregate. Literacy will recede and, consequently, people will become less intelligent. That may be true on average, but it doesn’t mean it’s true for everyone, nor that it’s necessarily bad for society. In the most plausible scenarios, (generative) AI will not eliminate all intellectual jobs. Many intellectual roles will still be fulfilled by humans, if only because of comparative advantage. Moreover, not everyone will stop reading and writing. Quite the contrary: in a world where literacy becomes a rarer skill, its value may increase insofar as it’s still needed to correct thinking machines, to feed them with new inputs, and because some humans will still grant significance to human intellectual output.”
AI-ruled societies may be way more unequal than our current ones, not only because AI companies are likely to concentrate a large fraction of profit-making and many people will no longer be able to earn an income by working. The most significant inequalities will not be economic but cognitive and will affect the Millian core of a free society: the capacity to argue and deliberate about the Truth. This capacity is not only valuable as a precondition to be an active political member of society. It is also needed to be a genuine individual in Mill’s sense, that is, someone who has the ability to settle on life projects and to pursue them with an unshakeable conviction, to exercise judgment even, and especially, when it goes against the dominant view, and to understand, endorse, and turn ideas and values into one’s own. The great potential social costs of AI technologies may well be that it will undermine individuality in this Millian sense.
I willingly acknowledge that over the last few paragraphs, I have been doing exactly what Carlo warns us against in his essay: focusing on costs that we can articulate in our linguistic and conceptual framework, at the risk of underestimating potential benefits. So, let me be clear: I don’t deny that there are potential benefits, more or less understandable, tied to the use of AI technologies, including in terms of cognition. Because these benefits are more difficult to anticipate, the only sensible approach in my view is a skeptical one that puts a low confidence value on our beliefs about the effects of AI, whatever they are. Only experience will tell us what the costs and benefits really are. But, and this is another Millian theme, openness to change through “experiments in living,” even when weighted with significant skepticism, is part of what it is to be an individual in a liberal society.
[1] John Stuart Mill, On Liberty, Utilitarianism and Other Essays, Second edition, ed. Mark Philp and Frederick Rosen (Oxford University Press, 2015).
[2] Ibid., p. 36.