The robots are coming: Artificial Intelligence, Automation and Work
A colleague sent me a paper on AI and the coming of the robots a few months ago — ‘Artificial Intelligence, Automation and Work’ by Daron Acemoglu and Pascual Restrepo. It was published as a working paper in NBER at the beginning of the year, but has turned up in a few places, and you can find a pdf easily on Google. It was even referenced recently in a Bank of England speech by Andy Haldane.
I made some notes on it at the time, more from the public policy point of view as well as venting some sass, which I’m now posting more or less as I wrote it. As well as the paper, I also touch on the moral failure of AI research and alarmism.
The upshot of Acemoglu and Restrepo’s paper is a way of thinking about productivity and policy to address it, and more broadly about the labour market and industrial strategy. You know, the kind of thing the great minds working on economic policy should take a view on.
They use a task-based framework to look at the effect of automation and AI on labour, wages and employment. This is useful as it makes the displacement effects of automation clear — as the set of tasks the machines do gets bigger and replaces labour. Tl;dr: This is #bad for labour.
Ameliorating this to an extent are:
- Boosts to productivity
- Capital accumulation (increasing its rent)
- Deepening automation (better robots)
But they can never make it up, and so labour still loses its share and ends up totally hosed.
The off-set to automation, historically, has been the creation of new tasks. Thus expanding the set of tasks where labour has a competitive advantage.
They also explore what can go wrong, and lose us any productivity gains:
“Our framework also highlights the constraints and imperfections that slow down the adjustment of the economy and the labor market to automation and weaken the resulting productivity gains from this transformation: a mismatch between the skill requirements of new technologies, and the possibility that automation is being introduced at an excessive rate, possibly at the expense of other productivity-enhancing technologies.”
Now, I didn’t go through the maths in detail. But it does suggest some interesting ways of thinking about productivity and framing any policy responses. That is, our role in government is about keeping the scales balanced during growth. And recovering the missing productivity from a sub-optimal rollout of the robots — maybe solving some of the productivity paradox/problem.
This way of thinking is not a normal part of economists heuristics, as the production function is (usually) blind to this (blindingly obvious) displacement.
Some of the policy areas are motherhood and apple pie:
- Reduce labour market frictions.
- Smooth geographical labour market adjustments. (We’ve been very bad at these.)
- Re-skilling. Importantly this is not just about more funds, but about the right focus. The wrong skills are same as no skill / a shortage. Considering apprenticeships as an example, their great utility here would not be about the children/ new entrants to the labour market — that’s not the problem in Redcar. I’ll go deep on skills interventions in another post.
- The tax code around capital for equilibrium automation. The risk is you put your thumb on the scales, and push automation at the expense of labour for no productive benefit or even losses.
This doubles down on the risk that as the robots come, other parts of business are not ready for it. Then it’s a repeat British industry replacing steam with electric power, and not redesigning their factories.
The really important one is:
- Creating new tasks.
The jobs are never coming back. So we need new ones. And we can make choices about this.
The robots, and automation, are like a slow burning economic shock…
For various reasons the market can fail, and you can end up in a local minimum. You don’t even get the productivity benefits from automation.
Seamless segway to AI alarmism:
There is a moral failure in some of the AI focus in research. There is effectively a fixed pool of scientists / robot researchers, and they can be misallocated. I would compare it to the brain drain of finance, pulling in talent from other fields and wasting them. If you have the time (40mins), this AI alarmism video is good. On how irrational AI beliefs can really lead to a lot of wasted time and effort. (The image used in the header comes from this talk, a 70s sci-fi book cover by Shigeru Komatsuzaki. I found a collection of them here.)
For this reason you need:
- diverse research portfolio across your university sector
- diverse capital investments
To hedge against irrational or sub optimal market behaviour. As rational as I’m sure Elon Musk and Jeff Bezos are.
Any productive benefits from automation can also be lost from distributional effects. Inequality reduces any marginal consumption from the gains. Flat demand reduces profitable avenues, double down on automation as returns there are to capital. And we’re back in a Marxist crisis of profitability. Or swapping secular stagnation stories.
- Falling working population means we need all the robots we can get. (Labour’s share needs to fall?) We can both lean into the robots, and lean into creating new jobs.
- I think you can make a good argument that current copyright and IP law is reducing productivity.
- Data should be considered infrastructure.
Activities like automating the production of official statistics by government analysts using reproducible analytical pipelines is the deepening of automation (the intensive margin, in the paper). This kind of displacement should shift people to work on:
- the more creative use of the data and analysis,
- collecting the data to feed into the AI / robots.
To me, this data creation echoes the early shifts in agricultural employment, and in textiles, discussed in the paper:
“Another interesting example of this process is provided by the dynamics of labor demand in spinning and weaving during the British Industrial Revolution as re-counted by Mantoux (1928). Automation in weaving (most notably, John Kay’s fly shuttle) made this task cheaper and increased the price of yarn and the demand for the complementary task of spinning. Later automation in spinning reversed this trend and increased the demand for weavers. In the words of John Wyatt, one of the inventors of the spinning machine, installing spinning machines would cause clothiers to “then want more hands in every other branch of the trade, viz. weavers, shearmen, scourers, combers, etc…” (quoted in Mantoux, 1928).”
Which also casts the organisation that are poor at providing the properly curated data we need as the new luddites (I’m looking at you ONS, and executive agencies like the ESFA).