The Unstoppable Force Driving Inequality in Every Nation on Earth — and Why AI Is Making It Worse

This year, an odd scene is taking place in offices from Manchester to Manila. Thanks to a chatbot she pays $20 a month to use, a mid-level analyst who used to spend three days creating a quarterly report now completes the same task before lunch. Her supervisor observes. Higher-ups take notice. The number of analysts the company truly needs is calculated somewhere in a boardroom. All of this is not disclosed. Seldom is it. However, the pattern is now unmistakable and is occurring in every nation on the planet.

Inequality is not a recent development. For a century, economists have studied it and debated its causes, including globalization, taxation, weakened unions, capital flight, and the collapse of manufacturing. There is some truth in each explanation. However, beneath all of them lies something more structural, which Thomas Piketty covered in 700 pages in 2013 and which Sam Altman summed up in one sentence in his essay from 2021: the returns on capital continue to exceed the returns on labor. The owners of the machines benefit more when they generate more value. The wages received by everyone else are inadequate.

This old tale is made quicker, sharper, and more difficult to refute by artificial intelligence. Approximately 40% of jobs worldwide are exposed to AI in some significant way, according to IMF staff research published earlier this year. In advanced economies, this number rises to nearly 60%. This wave is unique in that it affects jobs that were thought to be safe until recently. associates in law. young programmers. writers of copy. teams that provide customer service. The 1980s saw a terrible decline in blue-collar jobs. The professional middle class is the target audience for this quieter version.

Gains in productivity might eventually improve everyone. There is precedent for that optimistic scenario, including electrification, the internet, and shipping containers. However, it took decades for each of those to spread, and they were all accompanied by difficult changes that destroyed lives long before the advantages became apparent. One unsettling observation made by Allan Dafoe, who oversees frontier safety at Google DeepMind, is that once a potent technology becomes available, societies that use it typically outcompete those that do not. In 1853, Japan’s Meiji Restoration started as a desperate reaction to Commodore Perry’s warships. Since resistance was not an option, the nation modernized. Every economy is feeling the same pressure related to AI at the same time.

The infrastructure itself is concentrated, which is what the AI moment adds and exacerbates the inequality narrative. Only a small number of businesses can afford the approximately 10^25 floating-point operations required to train a frontier model. The output can be used by a hospital in Karachi. It is unable to construct the model in a realistic manner. Countries that don’t have access to affordable electricity, cutting-edge chips, or research universities are being asked to rent the future from those that do. The similarities to previous colonial arrangements, albeit in more amiable language, are difficult to ignore.

However, the situation isn’t always dire. Prices do tend to decrease as a result of diffusion. With a smartphone, a student in Nairobi can access tutoring that cost families thousands of dollars ten years ago. A model that has been trained on millions of medical records can be consulted by a rural physician. When used properly, AI has the potential to close some gaps while expanding others. There are still unanswered questions about whether governments can tax capital quickly enough to finance that shift, whether labor can reorganize quickly enough to remain relevant, and whether frontier businesses can be relied upon to share what they create.

How this ends is still up in the air. According to history, those in control of the machines typically prevail in the initial round, and society as a whole only catches up when it organizes. As this develops, there’s a sense that the 2020s will be remembered not for the chatbots per se, but rather for what they revealed about the people who already held power in every nation.