The Ethics Lab blog

Humanoid Robots and Low-Wage Work: The Jobs That Never Get Posted

A woman in Chennai earns $3 an hour filming her chores to teach robots. We count the humanoids actually working, look at what automation did to workers before, and lay out the market, labor and basic income answers in terms their own supporters would accept.

Written by The Ethics Lab Team

Nagireddy Sriramyachandra is 25, lives in Chennai, and earns about $3 an hour doing her chores on camera. She records household tasks for a data company called Objectways. The footage is called "egocentric data," meaning it is filmed from the worker's own point of view, and it is meant to teach robots to do those chores themselves.

Gulf News reported in June, with video from the French news agency AFP, that thousands of Indian workers now film themselves chopping vegetables, folding clothes and making coffee, with head-mounted cameras, phones and motion sensors. She isn't competing with a robot. She's teaching one.

If humanoid robots really arrive, in warehouses and factories and maybe our homes, what happens to the people who do that work now? In the United States that means stockers, packers, movers and caregivers, who earn about $17 to $20 an hour.

First, count the robots

How many humanoids are doing paid work today? A handful.

The clearest case comes from Figure, a robotics startup. Its Figure 02 robot spent 11 months at BMW's plant in Spartanburg, South Carolina, working 10-hour shifts, Monday to Friday. It loaded more than 90,000 sheet metal parts, against a required cycle of 84 seconds, 37 of them for loading. Figure says the robot "contributed to the production of" more than 30,000 BMW X3s, which is a careful way of saying it helped. The forearm was its "top hardware failure point." Then, as Figure released a newer model, the company announced it was "officially starting the retirement of Figure 02."

Agility Robotics says its robot Digit has moved more than 100,000 plastic bins, called totes, at a warehouse in Flowery Branch, Georgia, run by the logistics company GXO. The company's post describes one robot, not a fleet.

Then there's Tesla. Elon Musk predicted "roughly 10,000" Optimus robots in 2025, and Tesla missed that target. In January 2026, Electrek reported, Musk acknowledged that no Optimus robots were doing useful work in Tesla's factories. At Tesla's October 2024 robot event, Bloomberg and other outlets reported that the robots were remotely operated by humans. Bloomberg's sources said the robots could walk on their own, but people behind the scenes handled many of the chats with guests. One robot bartender told a guest it was "assisted by a human."

None of these numbers come from an independent audit. They come from the companies or from reporters. Today the real humanoid workforce is a few machines doing narrow, repetitive jobs, which is a reason for patience. It is also a reason to pay attention now, because the rules for a technology are easiest to set before it is everywhere.

The math the robots are built on

A humanoid spreads only if it costs less than a person. Agility's CEO, Peggy Johnson, said in 2024 that her company was "looking at an under 2 year ROI vs a human at a fully loaded $30 an hour." ROI means return on investment. "Fully loaded" means the whole cost of a worker to the employer: wages plus taxes, insurance and benefits. In other words, the robot should pay for itself in under two years.

Which workers is that math about? The Bureau of Labor Statistics (BLS), the federal agency that counts American jobs, lists about 2.9 million people who move freight and stock by hand, with a median pay of $40,240 a year. Another 2.8 million stock shelves and fill orders, at $37,330. About 555,000 pack things by hand. Pick up a box, carry it, put it down: that is the work Digit does.

Home health and personal care aides, a bigger group than any of those, number about 4.7 million. Their median pay is $17.21 an hour, and BLS expects the job to grow 18 percent over the next decade.

Japan has tried robot helpers in elder care. The researcher James Wright studied a care home that tested several. Staff stopped using a lifting robot called Hug after a few days, saying it was "cumbersome and time consuming to wheel from room to room." With Pepper, a small humanoid that led exercises, staff had to stand next to it, "copying its movements and echoing its instructions." In 2019, only about 10 percent of more than 9,000 Japanese eldercare institutions had any care robot at all. Wright's conclusion: "the machines failed to save labor."

So far, the machines do best where the job is a box and a conveyor. In care work, the low-wage job BLS expects to grow 18 percent, the care robots Wright watched made more work for the staff.

What history says happens to the workers

Humanoids are new. Machines that take over human tasks are not, and economists have studied what happened to workers the last few times. The answer is stranger than "they all got fired."

Start with the ATM. "One might expect such automation to decimate the ranks of bank tellers," the economist James Bessen wrote. Instead, "the number of bank teller jobs did not decrease as the ATMs were rolled out." Each city branch needed fewer tellers, 13 instead of 20 between 1988 and 2004, so branches got cheaper to run, and banks opened 43 percent more of them in urban areas. People who say automation creates as many jobs as it destroys love this example.

The story has a sequel. BLS now counts 339,200 tellers and expects that number to fall 13 percent by 2035, because of online banking, fewer branches, video kiosks and "enhanced" ATMs. The ATM lesson was real. It just had an expiration date.

What about factory robots, the closest thing we have to humanoids? Daron Acemoglu and Pascual Restrepo studied American local job markets from 1990 to 2007. Each time a market added one robot for every thousand workers, the share of adults with jobs fell by 0.18 to 0.34 percentage points, and wages fell by 0.25 to 0.5 percent. In plain numbers, about 2 to 3 of every 1,000 adults who would have been working weren't.

A team of economists including Wolfgang Dauth studied Germany and found something different. Robots cost manufacturing jobs, but those losses were "fully offset by new jobs in services," and by several measures the new jobs were of higher quality. The cost, they found, "mostly falls on young workers just entering the labor force."

In Germany, the people who lost out were mostly not the ones already on the factory floor. They were the ones who would have been hired into factories next, which, if you're 16, means people like you.

Amazon's plans, as reported, look like that pattern to us. In 2025 the New York Times described internal Amazon documents. According to those papers, Amazon's automation team expected the company could avoid hiring more than 160,000 U.S. workers it would otherwise need by 2027, and more than 600,000 by 2033, saving about 30 cents on packing and delivering each item. Amazon said the documents "appear to reflect the perspective of just one team" and don't represent its overall hiring strategy.

Taken together, the studies and the plan describe a kind of job loss that makes no noise. Fewer adults working, lower wages, entry-level jobs that never get posted: none of it requires a single layoff, and it is hard to go on strike for a job nobody knew existed.

Three serious answers

People who think hard about this fall into roughly three camps. Here is each one in terms its own supporters would accept, followed by the best objection from the others.

The market answer: let the machines in and help the people they hurt. Productivity is how societies get richer, and this camp reads the ATM story as the normal pattern. When Bill Gates proposed a tax on robots in 2017, the economist Larry Summers, a former Treasury Secretary, answered: "I cannot see any logic to singling out robots as job destroyers." He warned that "A sufficiently high tax on robots would prevent them from being produced," and that "in an open world, taxes on technology are likely to drive production offshore rather than create jobs at home." Instead, he backed retraining, infrastructure, "direct public employment programmes" and "targeted wage subsidies for groups with particularly severe employment problems." This camp can also point to the German finding that the new jobs were better, and to the 760,500 openings BLS expects each year for home health aides.

The objection: the American study found real job and wage losses where robots clustered. Retraining and subsidies have to reach people who were never hired, and nobody counts them.

The labor answer: bargain before the robots arrive. In October 2024, dockworkers on the East and Gulf Coasts went on strike. Their union president, Harold Daggett, demanded "absolute airtight language that there will be no automation or semi-automation." The union said, "Our position is clear: The preservation of jobs and historical work functions is nonnegotiable." The strike lasted three days. The six-year contract that followed gave a 62 percent raise and let the ports bring in semi-automated cranes, with "guarantees for new jobs linked specifically to each piece of equipment." In South Korea, Hyundai's autoworkers' union answered the company's plan for a factory able to make 30,000 robots a year by 2028: "without labour-management agreement, not a single robot using new technology will be allowed to enter the workplace." At the United Auto Workers convention in June 2026, the union's president, Shawn Fain, said: "The fruits of our labor have multiplied like never before, but workers aren't reaping the harvest." This camp can point to results: the dockworkers won money and written job guarantees, including for jobs that don't exist yet.

The objection: a contract covers the workplace that signs it. Summers' warning applies here too. A port or plant that blocks machines can lose work to one abroad that doesn't.

The basic income and ownership answer: share what the robots make, directly. If machines produce more with fewer workers, the gains should reach people as cash or as shares. One large recent test, run by the research group OpenResearch, gave 1,000 low-income people $1,000 a month for three years, while 2,000 others got $50 a month. The researchers say recipients "had increased agency to make employment decisions that worked best for them and their families," meaning more freedom to choose when and how to work. For an ownership version, Sam Altman, OpenAI's CEO, has proposed an "American Equity Fund." Large companies would pay 2.5 percent of their value each year in shares, plus a tax on land, and he estimated that "a decade from now" each adult would get "about $13,500 every year." Alaska already sends residents a yearly dividend from a state investment fund.

The objection: the same study found that recipients worked one to two fewer hours a week, were 4.2 percentage points less likely to be in the labor force, and earned about $1,900 a year less from other sources. Eric Boehm wrote in Reason, a libertarian magazine, that "Taxpayers should not be expected to fund an increase in individuals' leisure time." And the real dividend is small: Alaska's was $1,000 in 2025. That helps, but it doesn't replace a paycheck.

Who owns what the robot learns

All three answers lean on a question that gets less attention: who owns the robots?

In the biggest warehouse deal in this story, not even the company using them does. GXO signed a multi-year "Robots-as-a-Service" deal with Agility, which means it rents robot labor instead of buying robots. The home robot NEO, from the company 1X, can be bought for $20,000 or rented for $499 a month. At first, much of its work will be done by remote human operators, who see inside your house through its cameras when you approve it. 1X's CEO, Bernt Børnich, explained why: "If we don't have your data, we can't make the product better."

The robot learns from people: a remote operator steering it through someone's kitchen, a woman in Chennai filming her own chores. As far as these examples show, the people who supply that help are paid for their time, and the improvements belong to whoever owns the robot.

Meanwhile, workers' share of what the nonfarm business sector produces, counting pay and benefits, fell to 52.8 percent in the spring of 2026. BLS calls it "the lowest level ever recorded," down from 65.5 percent when the count began in 1947. BLS doesn't blame robots, and economists argue over the causes. But the ownership question is arriving at a moment when workers' slice is already the smallest on record.

Sriramyachandra earns about $3 for an hour of footage of her own hands. A home robot that learns chores from human help rents for $499 a month. She is paid for the hour. Whoever owns the robot is paid every month after that.

Sources

Free AI lessons for teachers