Can we calm down?

AI doomerism isn’t helpful. Neither is AI boosterism.

Imagine a world where everything is better. Your job is more fulfilling and pays more. Your healthcare is better AND cheaper. Even war is improved, with less unnecessary bloodshed. That’s the vision advanced by Marc Andreesen in a blog post titled ‘Why AI Will Save the World’. Article titles are often more dramatic than their content. The online marketplace is competitive and writers need to attract attention. Andreesen’s post is unique in that its content is more extreme than the title suggests. It’s quite long and filled with multiple extreme statements so I won’t cover all of them but I’ll focus on two that struck me.

The first is a broad statement: ‘In short, anything that people do with their natural intelligence today can be done much better with AI’. There are numerous examples to challenge this. There’s the number of radiologists in the US increasing despite Geoffrey Hinton’s claim AI would make them obsolete. Or the ruinous results of the UK government’s attempts to automatically calculate grades. Or Watson Health being sold for an amount described as ‘salvage value’.

The second is a claim about AI’s impact on jobs. I think this is particularly important as it’s an area of substantial discussion since the release of ChatGPT. While estimates vary widely on how work will be affected by AI most agree that it will be affected. Andreesen employs the common argument that the net effect of AI – and technology in general – is to create more jobs than it destroys and to raise wages. Andreesen adopts a chastising tone, as though it’s so obvious that AI will be a net positive for jobs that people should be embarrassed to believe otherwise. My objection isn’t to the claim itself which is borne out by every analysis I’ve seen. But this hides the fact that the benefit isn’t shared equally. Writing in Brookings, Harry J. Holzer sums this up succinctly:

?

‘More broadly, workers who can complement the new automation, and perform tasks beyond the abilities of machines, often enjoy rising compensation. However, workers performing similar tasks, for whom the machines can substitute, are left worse off. In general, automation also shifts compensation from workers to business owners, who enjoy higher profits with less need for labor.’

This raises multiple questions. In what kind of jobs can workers complement AI instead of being replaced? How can people change from replaceable jobs to non-replaceable jobs? How long does this change take? What’s the cost of this change to the individual? If we don’t have answers for these questions we should stop wagging our fingers at people who worry about job impacts.

Rather than raging about how AI is going to destroy society or proclaiming that AI is going to bring us to the land of milk and honey could we take a breath and just calm down?

?