New gig alert: Training our robot replacements
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Aïda Amer/Axios
As layoffs mount, training AI is becoming a lucrative new side hustle — even if it means helping build the model that could one day replace you.
The big picture: The AI giants desperately need money, energy and data. They also need people. At least for now.
- When CEOs promise humans will stay "in the loop", it's unclear if that means to make big consequential decisions, simply review AI's work, or take the wheel in case of emergency.
- Right now humans are still a key part of that loop, acting as the mechanical turks of generative AI's magic.
- Like the 18th-century chess automaton secretly steered by a hidden master, today's "AI trainers" perform unseen labor to make machines appear smarter than they are.
How it works: Humans select, clean and label data to fine-tune AI models, teaching them how to answer questions or understand images.
- Uber recently announced an initiative to allow drivers to perform simple AI tasks to make money during the times they're not driving. Some of those tasks will help self-driving tech companies develop the tech that could help train robots to drive.
- Amazon announced augmented reality glasses this month designed to help delivery drivers do their jobs more safely. An Amazon spokesperson did not answer Axios' question about whether the data from the glasses would be used to train autonomous driving systems or delivery robots, but it's conceivable this could be a future step, given the company's goals and recent announcements.
- San Francisco startup Mercor pays doctors, lawyers and others to train AI so machines can perform like human professionals. "This is part of a mad rush to fine-tune AI with true human expertise so it can do for free what junior employees do now — and, later, what senior ones get paid good salaries to do," Axios' Jim VandeHei and Mike Allen report.
- OpenAI is reportedly working with Juilliard music students to teach a model how to compose like humans, according to The Information, and with former investment bankers to train models to do Wall Street's entry-level work, per Bloomberg.
The other side: Many workers are embracing an "if you can't beat 'em, join 'em" mindset, betting that training AI may be the only way to stay relevant as automation accelerates.
- Historically, the impact of technology on humanity is that it has required us to improve ourselves, NYU Stern professor Vasant Dhar tells Axios. Dhar has over 30 years of experience in machine learning research and has been studying the future of work in the age of AI for nearly as long.
- Dhar doesn't pretend to know exactly what world looks like more than a decade from now, but he does see a bifurcation between those willing to work with AI and those who are not.
- "What I'm seeing is the AI just gets better," Dhar told Axios. "We get challenged to up our game. Some of us up our game. Many of us don't."
The bottom line: Humans are fueling AI's growth.
- And possibly training themselves out of future work.
Go deeper: Bots are elbowing out humans in skill at office work
