Three slow-moving trends
Productivity-pay decoupling
In the United States, labour productivity — output per worker hour — has risen steadily since the 1970s. Median hourly compensation has barely moved in real terms. The gap between the two lines opened in the mid-1970s and has not closed. The Economic Policy Institute has tracked this for decades, and the broad pattern appears in most developed economies, though the size of the gap varies.
The declining labour share of income
Karabarbounis and Neiman's much-cited 2014 Quarterly Journal of Economics paper documented a global decline in labour's share of national income over the last four decades — from roughly 64 percent to roughly 56 percent in the United States. The causes are debated (automation, market power, globalisation, measurement), but the trend holds.
Rising transfer dependence
Over the same period, the share of household income coming from government transfers (social security, healthcare, unemployment benefits, food assistance) has grown in most developed economies. In 1950 US households received about 8–12 percent of their income from transfers; by the 2020s the figure is closer to 18–22 percent, and higher in countries with larger welfare states.
Why these three matter together
None of these trends was caused by AI — they were all well under way before deep learning arrived. But together they tell a story: labour is paying a shrinking share of national output, and the state is paying a growing share of household income. If AI accelerates the direction of travel, policy will need to respond. If it does not, the trends will still matter.
Which of these three trends had you heard of before? Which, if true, would most change how you think about work?