Why we shouldn’t fight automation
Plus: the poverty decline isn’t just about China, the plummeting wages of British degree holders, and more
Welcome to The Update. In today’s issue:
British degree holders earn 15 percent less than 20 years ago
In brief: Anthropic could overtake OpenAI, dark traits yield dark beliefs, and more
Why we shouldn’t fight automation
When AI is finally good enough to do our jobs on a grand scale, how should we react? In recent years, many have argued that we should steer AI toward making workers more productive rather than replacing them. Economists such as Daron Acemoglu, David Autor, and Simon Johnson call for tax incentives, regulations, and federal research to encourage ‘augmentation’ and discourage ‘automation’.
But can we really steer AI in this way? Governments have a weak track record of shaping the economic impact of powerful technologies. I worry that if they tried, it would consume political bandwidth that’s better used elsewhere. AI advanced enough to cause widespread automation would transform society profoundly, and governments should prioritize issues where they can make a bigger difference.
While I certainly think that human agency will shape the future, rational use of that agency requires recognizing its limits. Even if it turns out to be more feasible to direct AI development toward augmentation than I think, I doubt it would last. Unless AI progress is completely halted, automation is coming, sooner or later. The arguments for steering away from automation are shortsighted.
And even if we could stop automation, should we? Acemoglu, Autor, and Johnson claim that ‘displacing workers will never be good for workers or for the labor market’. But in my view, that betrays the same shortsightedness. Over longer time spans, automation has been one of the great engines of human prosperity. While it can be painful for workers who lose their jobs, the compound effect has been to make societies enormously richer. We will need new policies to adapt to AI-driven automation – such as reforms of our welfare systems – but seeking to stop it is not the answer.
The decline in extreme poverty isn’t just about China
In 1990, over 940 million Chinese people lived in extreme poverty – less than three dollars per day at today’s prices – but since then, market reforms have led to the most spectacular fall in poverty in world history.
A common assumption is that China accounts for the entire global decline in extreme poverty. But as Max Roser and Pablo Arriagada show, this isn’t true. While China is an outlier, the rest of the world has also seen a considerable fall in extreme poverty, from 33 percent in 1990 to 12 percent today.
British degree holders earn 15 percent less than 20 years ago
Does it pay off more to go to university than it did in the past? In most rich countries, the answer is yes – but not in Britain. Since 1997, the British graduate earnings premium has almost halved.
Combined with weak productivity growth, this has left British degree holders earning about 15 percent less than they did two decades ago.
But there’s one part of Britain where the graduate premium has held up: London. While the rising minimum wage has flattened the earnings distribution elsewhere, this effect is weaker in London, where earnings are higher.
These charts show lower graduate premiums than the first one in this section (due to differences in methodology), but the trend is the same.
Today’s AI may pose lower pandemic risk than expected
By making biolab work easier, AI could increase the risk of a catastrophic disease outbreak. Do today’s AI systems already constitute such a threat? To test this, the biosecurity nonprofit Active Site recently ran a randomized controlled trial where people without molecular biology training attempted tasks required to reconstruct a virus. Before the trial, Active Site and the Forecasting Research Institute recruited biosecurity and virology experts as well as superforecasters to predict the results.
Interestingly, the experts and forecasters overestimated the usefulness of AI. On average, they predicted that AI access would more than double participants’ success rate on three core tasks. But it turned out there was no significant difference between participants with access to AI and those limited to using the internet. After learning the results, the experts and forecasters lowered their estimated risk of an individual or group causing a catastrophic disease outbreak in 2028.
Britain’s trade deficit is surprisingly large
Britain’s trade deficit doesn’t get much attention, but it is actually the second largest in the world in absolute terms:
As a fraction of GDP, it is larger than the American trade deficit:
In brief
People with dark personality traits find the world less beautiful and meaningful
Why are scientists more enthusiastic about using AI for writing than humanities professors?
American government is overly constrained, argues Francis Fukuyama
Citrini’s viral scenario of an AI crisis in 2028 is bad economics
The US now imports more from Taiwan (23 million people) than China (1.4 billion)
The Research Revival Fund supports projects that resurface forgotten research
AI and robotics could make automation less dependent on repetition
That’s all for today. If you like The Update, please subscribe – it’s free.
I agree that governments would struggle to effectively shape the economic impact of AI through policy, but I also think we shouldn't underestimate just how painful this type of transition could be for huge numbers of workers - I don't have particular faith in government safety nets responding adequately.
This links to a point that Alex Imas has been making, the more unpopular AI gets, which isn't helped by doomsday predictions from labs, the more likely it is that societies do end up fighting automation, as that will become a popular political strategy. Not really sure what the answer is here.