Life as a member of the communications team for frontier AI labs must be fascinating. Imagine you’ve got this big report ready to go out on the extremely elaborate model your economics team built that you want to position as a tool for policymakers and academics to understand how AI is going to impact the workforce, and the night before it publishes, your colleagues just casually (and publicly) declare that there’s a greater than 10% chance that your technology kills everyone within the next decade. Is that in the model? Do you address that?
Anthropic went with no to both of those questions, but it did, in fact, release its Economic Index growth model, which is designed to project what the economy could look like by the end of the decade, based on different levels of AI functionality and adoption. You’ll be shocked to learn that the big takeaway from the model is that everything is going to be a-okay in most cases—though the outcomes that aren’t on the chart are probably as noteworthy as the ones that are.
Per Anthropic, as long as we stay within the parameters of somewhere between “business as usual” and an economy where “AI increases growth to about twice the normal rate,” we can expect unemployment to stay within historically normal ranges and wages to “remain flat or rise depending on the industry.” However, if we reach growth that is “faster than anything in economic history,” then there will be “adverse impacts on wages and job prospects for knowledge workers.”
But don’t worry, because even when it’s bad, it’s good! “In those scenarios, society is far wealthier, so the challenge is making sure that the gains are broadly shared.” And since we’ve historically been so great at wealth distribution, there are certainly no red flags to raise there.
The model is interesting, at least. As the AI lab explained it, it sought to effectively break down jobs into bundles of tasks, then determine how the availability of AI might impact those tasks—what can be automated, what new work will be created because of AI, and what will likely remain human labor. Then it scales AI adoption levels to each job to determine how different industries and the economy as a whole will be affected by the technology.
It ultimately spits out a model that, at the extreme end of the spectrum, projects an economy that is 50% larger than it is today and has a nearly 30% unemployment rate. For what it’s worth, the furthest you can push the model the other way is zero AI adoption, which amounts to the economy functioning as usual. It doesn’t account for a potential AI bubble and how its popping may impact workers and the economy, which is currently wildly overleveraged on the idea that the AI boom is for real.
Even the model’s most extreme edges also don’t account for everyone dying, which is apparently within the internal parameters for possible outcomes. Evan Hubinger, Anthropic’s alignment science lead, posted on Twitter yesterday that, “We really do earnestly believe AI could kill all humans!” and put the odds at greater than 10% over the next decade while admitting, “we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
Feels like that’d probably result in some economic shrinkage, but maybe eliminating humans would just get rid of inefficiencies. Guess we’ll just plow forward unimpeded until we find out!
Read the full article here
