Practicing my Handwriting in 1439
Tyler Cowen’s latest Bloomberg column on the dangers of AI predicts, among other things, that AI will disrupt the current high status and earnings of “wordcels” and “ideas people.”
The larger theme is becoming evident: AI will radically disrupt power relations in society.
AI may severely limit, for instance, the status and earnings of the so-called “wordcel” class. It will displace many jobs that deal with words and symbols, or make them less lucrative, or just make those who hold them less influential. Knowing how to write well won’t be as valuable a skill five years from now, because AI can improve the quality of just about any text. Being bilingual (or tri- or quadrilingual, for that matter) will also be less useful, and that too has been a marker of highly educated status. Even if AIs can’t write better books than human authors, readers may prefer to spend their time talking to AIs rather than reading.
It is worth pausing to note how profound and unprecedented this development would be. For centuries, the Western world has awarded higher status to what I will call ideas people — those who are good at developing, expressing and putting into practice new ways of thinking. The Scientific and Industrial revolutions greatly increased the reach and influence of ideas people.
AI may put that trend into reverse. It is not hard to imagine a world, less than 20 years from now, in which a skilled carpenter is seen to have better prospects — professional and otherwise — than an articulate lawyer.
As a wordcel myself, I find this prediction concerning. The “moving history” of the scientific and industrial revolutions is worrying enough, but as Tyler said, at least they increased the returns to ideas people in the long run. Collapsing institutions, international arms races, and a permanent decrease in the rewards for my skills just sounds terrible. I am not a good carpenter!
But will this prediction really come true?
On one hand, almost all of AI’s realized impact so far and the closest technological analogies to AI point to it being a complement, rather than a substitute to “those who are good at developing, expressing and putting into practice new ways of thinking.”
We are still in the very early days of AI development. Still, the current uses of nascent AI tech is evidence for what the future will bring. So far, AI has been complementary to knowledge work. It makes software developers twice as fast on some tasks and improves the quality and speed of consultants. In my personal experience its a very useful complement to my research and coding, allowing me to work and produce more.
As AI matures, this may change. Scribes and illuminators were complements to printing in the early days of the press when illustrations and complicated formatting were more difficult, but it didn’t last.
We do have some good reasons to expect the complementarity between AI and ideas people to continue, though. Google search is perhaps the most precise analogy to more mature AI tech we have and has been a massive boon to idea producers. Google uses AI to interpret user prompts and turn them into high quality text and images. This technology replaced some previously important research skills, but ultimately it is a positive multiplier on the productivity and influence of ideas people. Tyler says the printing press is one of the better analogies for AI and that too made idea producers better off in the long run.
On the other hand, even if we grant that AI is a productivity enhancing tool for ideas people rather than a competing substitute it could still be very difficult to be an idea producer in the future. Think about farming and the industrial revolution. Mechanization brought massive productivity enhancements to farmers but the number of farmers dwindled. Today, farmers have good per capita incomes and influence but are a small proportion of the economy as whole.
Comparative advantage makes me confident that there will be some room left for humans in the world even if AIs become super-human agents. But we might not like our comparative advantage. It seems plausible to me that human comparative advantage is in moving around in physical 3D space. That is what most of our evolutionary engineering is built for. The frontal lobe is an afterthought compared to our cardiovascular system. It’s comparatively easy to imitate human-quality art or writing than human-quality basketball or carpentry.
So I am left wondering: am I practicing my handwriting in 1439? I can see some plausible futures where my investments in research, writing, and thinking are moot, and some where they are invaluable but I’m not sure which is more likely.
It occurs to me that (relative to my consumption ability) there is already approximately infinite content on the Internet.
My consumption of this sea of content is not evenly distributed. I do not randomly sample, and I don't think anyone else does either.
What do we actually do? Our content consumption is socially biased.
1. We consume content that is popular in our social group so that we can talk about something with our friends. (For everyone who is progress-adjacent, Maximum Progress is a must-read!)
2. We become attached to particular online characters, and want to read their work. (I like Max's takes on metascience, so I also want to see what he has to say about land reclamation.)
3. We do searches for particular information where it is in fact more transactional. (This might go away because I can just ask GPT-N≥4.)
To Robin Hanson at you, online writing isn't about information.
Epistemic status: highly speculative.
Your analogy with farming for me thinking. Nowadays bad and ok quality food has been commoditised, but rich people pay a premium for high quality, more artisan, more unusual foods. Here in the UK there are exceptions of relatively cheap but high quality content, such as the BBC and the Guardian. On the quality/cost graph, these would be outlier plots, with that status attributable to alternate revenue streams ie. the public license fee (BBC) and donations (the Guardian), coupled with a public service mission. But the general point I’m making stands. I could see the work of ideas people following a similar trajectory. Largely a commoditised and bad-to-ok quality, generated by AI, for a free or low price to the masses, with a premium being paid for the really good stuff (human created, above the level of what AI can do). So pretty similar to how things are now, with paid subscriber-only content. Only that the quantity and quality of the free and cheap stuff will increase. Pushing the average quality from human idea generators higher. The big assumption here is that some humans CAN still generate better ideas than AIs. We’ve seen AI beat the best humans at Go and StarCraft. And do protein folding and other crazy-scale things. But will similar achievements become dominant for ideas? None of us know right now. The tech is improving rapidly but we don’t know when the limits will be reached and what they will be. And over what time frame we are talking - tens of years, hundreds of years? Also: ideas are easy to have, but putting them out with high production values is much harder (but let’s not go down that rabbit hole now!). I suspect we will be waiting a very very long time before AI is routinely able to out-idea a brilliant human with a few AI copilots 🙂