The Algorithm is our AI weekly newsletter. Get stories like these delivered to your inbox. Sign up for our newsletter .
Silicon Valley is awash with talk of an AI-driven jobs apocalypse. It’s so gloomy that Anthropic researcher of societal impact, in response to Wednesday’s call for more information optimistic In her visions for AI’s near future, she said that there could be a short-term recession and “a breakdown of the early career ladder”. Her colleague Dario Amedei, CEO of the company, was less measured, calling AI “a general substitute for human labor” which would do all the jobs within five years.
These ideas do not come from Anthropic alone.
Unsurprisingly, these conversations left workers feeling panicky (and probably contributed to the support of efforts to halt construction of data centres which some gained momentum last week). This panic hasn’t been helped by legislators, who haven’t articulated any coherent plans for the future.
Even those economists that have warned that AI may not have yet resulted in job losses and may not lead to a cliff are now coming around on the notion that AI could have an unprecedented and unique impact on our work.
Alex Imas is an economist at the University of Chicago. When we talked on Friday, he shared with me two things: an honest assessment of our ability to predict what AI will look like and a call to action for economists.
Consider that every job is composed of smaller tasks. For example, a part of the job of an agent in real estate is to find out what kind of property clients want to purchase. In a catalogue that was first published in 1998, and has been updated periodically since then, the US government documented thousands of tasks. Researchers at OpenAI analyzed this data in December, judging how “exposed a job” is to AI. (They found that a real-estate agent was 28% exposed to AI). Anthropic then used the data to analyze millions of Claude conversation to determine which tasks were actually being completed by people using AI.
Imas says that knowing how exposed a task is to AI can give a false impression of the risk of retrenchment of a particular job.
He told me that “exposure is an ineffective tool to predict displacement”.
It is only illustrative for the most gloomy of cases — for a position in which literally Every task AI could do it without human guidance. Imas believes that if an AI model can perform all these tasks for less than you are paid, which isn’t a guarantee, as reasoning models and AI agents can be expensive, then the job will likely go away. It’s the case that elevator operators were referred to in the past. Today, a similar situation could be if a customer-service agent was solely responsible for triaging phone calls.
For the majority of occupations, it isn’t so easy. It’s important to know the details: some jobs may be in for a bleak future, but being aware of them is crucial.
How to get started You can also find out more about the following: When you are looking for a way to improve your finances, then look no further than It is difficult to predict the outcome of a situation based on exposure alone.
Consider writing code as an example. Let’s say someone who creates premium dating apps uses AI tools. They can do in a day what would have taken three days before. This means that the worker will be more productive. Employers can get more work for the same money. Will the employer then want to hire more or less employees?
Imas believes that this question should be a source of concern for any politician, as the answer depends on the sector.
We are working in darkness.
These efficiencies allow dating apps, in this case of the coder, to reduce their prices. A skeptic may expect that companies would simply keep the profits, but they could be undercut in a highly competitive market if they did. The lower prices are always going to drive You can find out more about this by clicking here. Demand for apps is increasing.
How much more demand is there? The company could grow and eventually hire if millions of people wanted it. More information about the product Engineers are needed to satisfy this demand. If demand is barely increasing – perhaps people won’t even want premium dating apps at lower prices – fewer coders will be needed and there may be layoffs.
This hypothetical can be applied to every task that AI is capable of, and the pressing question for our age will become clear: how price changes affect demand. This is what Imas stressed last week. We do not have data on the entire economy. We don’t have this data across the economy.
You can also find out more about the following:.
Imas says that the University of Chicago has partnered with supermarkets in order to obtain data using their price scanners. We don’t know the numbers for dietitians, web developers, or tutors (all of whom were found to be “exposed” to AI). It’s not a widely accessible or compiled data for researchers. Sometimes it is scattered among private consultancies or companies.
Imas says, “We’d need a Manhattan Project-style project to gather this data.” We don’t just need to collect statistics for the jobs which are clearly affected now by AI:
Imas argues that this would cost time and money but it is worth it.
It would allow economists to see how AI could affect our future and policymakers the chance to plan accordingly.