Learning and Superlearning

I’ve been reading a lot more science fiction over the last few years. I don’t have one explicit motivation for it. Maybe I keep it up because I am trying to process the 2020s artificial intelligence revolution. 1. For me, at least, I enjoy seeing how well some authors could see the dilemmas of the 2020s from the 1980s and before. One work that I think about a lot lately is Dan Simmons’ Hyperion (1989), in which a computer superintelligence bestows unfathomably advanced technology on humanity before retreating into a separate, hidden domain known as the “TechnoCore.” The book is all about the hazards of the ongoing relationship between humans and computers, but it does not imagine some sort of final competition between them where one emerges and supersedes the other. In Hyperion, no matter how “smart” the computers get, it does not lead to the definitive obsolescence of the human. Instead, the differences between these two intelligences lead them to exist further apart from one another.

I think of this today because I worry a lot about how human beings learn. I want us to continue to push the boundaries of learning while living alongside machines whose ceiling for capability has no defined limit.

Learning is an obscure process. To my knowledge, there has never been a great answer for why two people with the same prior background, training and measured ability can sit in the same room, look at the same material, and come out with very different levels of understanding. As for the term you want to apply to whatever it is computers are doing today, calling it “learning” and “intelligence” only takes us to a higher plateau of obscurity about what is actually happening when anyone (anything) learns. At least human beings learn in places where they can be observed; the same cannot be said for machines, trained as they are by private enterprises operating under trade secrecy.

Regardless of how much machines do for some of us in the years ahead, I hope that more people–and institutions–create distinct enclaves dedicated to human learning. Rather than trying to meld with machines, an alternative pathway for civilization might see it withdraw, with great effort, from living so closely alongside computers.2 That is, a future society might see itself developing “in parallel” with an advanced computer intelligence, proceeding along separate tracks in which the two forces (human and artificial intelligence) depend on each other in innumerable ways, but still understand themselves to have separate interests and conditions of flourishing.

I am thinking about this Substack post on software deskilling, which has been creating a lot of discussion over the last few days. The post draws a parallel between the detachment from hands-on work that so many people in knowledge-related industries are confronting right now, and the experience of airline pilots, who have been required to accept increasing levels of automation over the last few decades. In both cases, there is a worry that the use of routine, blanket automation will degrade the ability of people to exercise good judgment during non-routine situations. To state it differently, is there any way to perform at a high level on novel problems without building one’s skills on routine work? If so, how will so many institutions, which are at least nominally motivated by gains in efficiency, continue to allow for routine (i.e., inefficient) work that builds competence for more complex tasks? Or, if routine won’t supply people with the tools to hone their abilities, what will?

One answer, which I can’t accept, is that the problem will simply go away, and that the machines will be largely in control soon. This is neither likely nor desirable. Even if computers were to achieve superintelligent capacities in specific domains, general intelligence is not the same thing as general judgment. Human judgment, into which I would fold dimensions like wisdom and humility, is to my knowledge barely a goal of current AI research–because it’s not easily quantifiable or reducible to an economic output. And there are a host of legal and institutional reasons why humans will need to be a major presence in even very automated situations.

So, assuming that there will be a place for humans working alongside machines for the long term, how will we create the conditions so that humans can continue to flourish, learning in their own way, outside of mandates of effiency? It’s reasonable to want a world in which both computers and people are as smart as possible. But we don’t get there by accepting a monoculture in which the most efficient way is always the right way.


  1. If I’m reading science fiction to deal with the present, I might be moving contrary to trend. Simon McNeil argues in a 2024 article that sales of science fiction are down because the genre raises uncomfortable questions about the present. Maybe the best 20th-century science fiction has undermined itself by being too prescient, by hitting too close to home: if you succeed in anticipating the future, you no longer provide the thrill of escape or mere entertainment. ↩︎

  2. The backlash over the last decade against internet bullying, social media and screens in general might be viewed with hindsight, someday, as the first great moves of disentanglement made against computing culture and computer dependency. ↩︎

Permalink