A year ago I gave ChatGPT a question devised for our primary school maths circles pilot. It tried, tried again, errored. It kept trying to write a recursive Python script that blew its memory cap. Today there are multiple large language models that can complete Olympiad level maths problems. Anecdotes like these abound.
The bullish Silicon Valley take is for an AI explosion that automates the economy in <3 years. The bearish take is for industrial revolution type progress that takes one to two generations. Whilst there is disagreement on the timeframe (and I happen to lean towards the latter view), there is little meaningful disagreement that the lower bound of the coming change is still extremely large.
We don't know what the implications of this will be, and making predictions is a fool's errand, but that shouldn't stop us pondering on the important questions. Here are some that I think are important and have been wondering about.

What happens to people’s early careers?
In most scenarios short of the singularity there is still a role for humans with deep domain expertise. The problem is that the route for acquiring this expertise is demolished.
At the moment firms pay people to do junior jobs through their early careers. These people earn the marginal product of their labour, perhaps plus some bonus to account for the expected value of their future contribution if they stay at the firm in a senior role later on. It is through these junior jobs that they acquire deep domain expertise.
In an AI future the marginal product of early career labour for the knowledge worker is extremely low, as these roles can be replaced by AI agents. The economics of the early career no longer work, so what is the pathway to deep domain expertise?
Some possible options include:
We move towards an apprenticeship model for knowledge workers, who become ‘indentured’ for a significant enough period of time to pay back the costs of their apprenticeship
Early career knowledge workers take on debt in order to fund this stage of knowledge acquisition, perhaps through more sophisticated financial markets that allow you to sell a share of future income
The higher education sector expands to fill this role, bringing public subsidy into the mix and rationing the flow of early career knowledge workers
If we don't find some option like this then knowledge work will become limited to those who have the wealth to invest in funding their early career - think unpaid internships but much worse.
What happens to the relative value of different attributes?
Intelligence has been seen as the superior personal quality for some decades, and the returns to intelligence are significant (although Tyler Cowen already argues that it is overstated based on the historical data). As other sources of intelligence become relatively cheaper, the relative value of intelligence as a human quality will decline. Which other qualities will rise in relative value, and how much?
Two likely categories stand out to me.
The first is the cluster of characteristics around agency and persistence, that one might characterise as “getting things done”. In a world of lower cost intelligence it is going to be easier than ever to get things done. The less intelligent but higher agency person is going to have an edge over the more intelligent but lower agency person, as the latter’s competitive advantage is eaten away.
The second is the cluster of characteristics around empathy and influence, that one might characterise as “people skills”. In a world of scarce intelligence, people with bad people skills got by because their intelligence was a valuable and scarce resource. This is decreasingly true.
I should note that I think this is true on the margins, and that the margins here are probably very big, but it may well not be true at the extremes. I expect that some humans will still come up with original ideas in many fields as they do now (novel idea generation clearly isn’t just explained by raw intelligence, and it's far from clear that AI will come to dominate this). I also expect there to be an edge for deep domain expertise (as above). But the 60-90th percentiles of intelligence used to have a big advantage over the 30-60th, and I expect that to no longer be the case.
What happens to the service sector?
Imagine a world where we have a cadre of deep domain experts heading service businesses, accompanied by indentured apprentices. What has happened to all the people who used to be lawyers, management accountants, auditors, etc?
Obviously we cannot predict this accurately as we do not know what broader structural changes there will be in the economy that will shift labour demand.
What I do expect is an increase in demand for more ‘human’ jobs, both in services and manufacturing. Even in an AI takeoff scenario where the capital share of GDP increases significantly, the increase in total output will mean there remains plenty of consumption spending in the economy. Might we see more spending on art and the artisan, as costs in other areas of the economy are reduced by AI and are substituted to these more luxury goods? Might we see more spending in health and education, as people seek more bespoke schooling for their children and healthcare for themselves? These would all generate employment for people who had previously been the ‘middle rump’ of the knowledge service sector.
This disregards the likely initial scenario, which is that many of these sectors work with government to get themselves regulated into continued existence. One can imagine that lawyers do not go quietly, but manage to get the status quo regulated into continuing for at least a while.
What happens to the state’s ability to raise revenue?
It’s easily conceivable that, in the AI future, a large proportion of global income goes to a small handful of corporate actors. Whilst this may be less the case if AI ends up being more commoditised, the capital intensity of development makes it likely that returns will be relatively concentrated.
If revenue goes to a small group of internationally mobile firms, what does this mean for the state’s ability to raise revenue through taxation? This is the thing I am most concerned about, and have the least idea of what is likely to happen.