Artificial intelligence has been blamed for a surge in mass layoffs and a sluggish job market, upending once-stable career options and putting even highly skilled white-collar jobs at risk. But in a somewhat ironic twist of fate, the same technology is now offering a new source of income for professionals: teaching AI models how to do their jobs.
A growing number of companies are hiring part-time contractors to help train the models that underpin today’s most advanced AI systems — the same systems capable of automating even the most specialized work. Doctors, lawyers, scientists, engineers and other professionals can earn hundreds of dollars an hour correcting these models’ mistakes, adding nuance where necessary and contributing industry-specific knowledge that could only be possessed by someone with real-world experience.
What Is AI Training?
AI training is the process of teaching artificial intelligence systems by feeding them data and guiding their learning. This involves simple tasks like labeling images or transcribing audio, as well as more complex work where humans review AI outputs, correct mistakes and provide industry-specific knowledge. The goal is to help the AI understand patterns, make more accurate decisions and perform tasks that would normally require human involvement.
This demand for human expertise stems from what industry experts call a “data wall,” or a point where models have already consumed most of the high-quality data available online and have nothing else to learn from. To break through this wall, developers need fresh, domain-specific information that can’t just be scraped from the internet. For many professionals, that gap has created a potentially lucrative revenue stream in an otherwise uncertain economy.
You may not want to plan your retirement around these gigs just yet though, though. Once AI systems absorb enough human expertise, that knowledge — and perhaps even the professions themselves — will likely be replicated, and eventually automated entirely. In the end, the very people fueling AI’s rapid growth could be putting themselves out of work for good.
A Shift Away From ‘Sweatshop Data’
AI has always depended on human knowledge. Researchers develop the systems, feed them data, oversee their training and evaluate their performance. To help models make sense of that data, companies have long looked to human annotators for data labeling, which is tedious work that typically involves identifying objects in images, transcribing audio clips or categorizing snippets of text. To date, much of this work involved the exploitation of workers in developing countries, who sometimes earn less than a dollar per task.
But now that AI models have gotten smarter, they require a higher caliber of human knowledge. They no longer need help identifying pictures of dogs, they need experts who can spot chatbot inaccuracies, dial in subtle nuances of language and apply their industry-specific knowledge to particular situations.
AI training company Mechanize Work went so far as to claim that the era of “sweatshop data” is over. While that may be a bit of an overstatement (these low-wage data labeling jobs still exist), a shift toward more specialized data is clearly underway. If they want to build systems that can autonomously manage large-scale software projects or solve novel, real-world problems, developers are realizing that they’ll need the help of highly skilled professionals to do so.
Companies Hiring Expert AI Trainers
There are dozens of AI training companies offering a wide variety of jobs, from low-wage data labeling gigs to high-paying roles that require industry-specific expertise. Basic data labeling jobs can still be found on Amazon Mechanical Turk, Scale AI subsidiary Remotasks or Appen subsidiary CrowdGen. But increasingly, data labeling companies are seeking workers with white-collar expertise for their projects, offering more money in return:
- DataAnnotation, a Surge AI subsidiary, advertises hourly wages starting out at $20 per hour for generalists, but expert projects — like those requiring a background in law, medicine or finance — start at $40 per hour.
- Outlier, another Scale AI subsidiary, pays up to $15 per hour to generalists, but it also offers more to researchers, subject matter experts and experienced professionals. A physics expert could earn between $30 to $50 per hour, for example.
- Alignerr, a Labelbox subsidiary, pays $20 to $40 per hour for generalist roles, but contractors with specific credentials, like a doctoral degree in chemistry, can earn up to 90 to $200 per hour.
- Mindrift, which is owned by Toloka, boasts that 70 percent of its 10,000 so-called “AI tutors” hold a master’s or doctoral degree. While English language writers earn up to $25 per hour, more specialized occupations can earn more. Mechanical engineers, earth science experts and software developers can all make up to $55 per hour.
Those hourly wages pale in comparison to those offered by Mercor and Surge AI, which are offering the most lucrative, specialized gigs in the AI training industry.
Recently valued at $10 billion, Mercor pays $1.5 million per day to its bench of roughly 30,000 contractors, which includes doctors, lawyers and bankers. Mercor is offering primary care physicians $130 to $170 per hour to review datasets and evaluate AI-generated outputs for an AI-assisted primary care product, and it’s paying lawyers $110 to $130 per hour to craft and review legal questions, evaluate AI-generated legal responses and solve domain-specific problems. Sales managers and investment services professionals can make $80 to $120 per hour for lending their expertise on Mercor. Chemists, photographers, order clerks, medical secretaries and journalists can train AI models for roughly $60 to $80 per hour.
Most of Mercor’s freelance projects require anywhere from 10 to 30 hours of work per week and last roughly three to four weeks. The company is also paying salaries between $90,000 and $200,000 to full-time “tutors” who can teach AI models about everything from real estate to statistics to web design. For instance, the company is hiring a “personality and behavior tutor” that can evaluate a model’s tone, humor and flow.
Surge AI goes after an even more sophisticated class of professionals, saying it wants to “raise AI with the richness of human experience: curious, witty, imaginative and full of unexpected brilliance.” To that end, it has contracted with more than 20,000 professionals with doctoral degrees, including Wall Street economists, National Institute of Health medical advisors, Stanford University researchers and NASA engineers, to name a few. The company is offering an hourly rate of $250 to $450 to medical fellows, $300 to $500 to management consultants and investment bankers and $500 to $1,000 to venture capital partners and startup CEOs.
The Pros and Cons of Working as an AI Trainer
AI training or data labeling can be a convenient side hustle, as it can be performed remotely on one’s own schedule. And workers claim it’s typically easy to get started. One Redditor said an AI training company processed and accepted their application on the same day, and then they went through a brief onboarding and screening process. Once they were assigned a project, they had to do some unpaid preparation work and an additional screening process to prove they were qualified for a project.
Jobs range from low-paying data labeling jobs — like identifying objects in an image, transcribing audio or categorizing the sentiment behind a written statement — to higher-paying data annotation tasks — like evaluating the accuracy of chatbot responses and offering suggestions for improvement. Doctors, lawyers and other skilled professionals may receive significantly higher wages by sharing their industry-specific expertise with the models.
That being said, the work can often be monotonous, and opportunities can dry up from time to time. Some of these sites even use AI to manage workers, which has led to some reports of mistreatment. One Redditor said they have heard stories of people losing access to projects or being kicked off the platform without any explanation. And two of the biggest data labeling companies, Surge AI and Scale AI, have been sued for paying workers less than they were promised. Scale AI has also been sued for exposing workers to graphic, psychologically traumatizing content.
It also remains to be seen whether companies seeking industry-specific knowledge will be challenged in court. Brendan Foody, Mercor’s 22-year-old CEO, told TechCrunch that companies don’t want large AI labs to access and train models on their data. To get around that, Mercor hires former employees who “understand those workflows” to help train AI models. He said the company tries to prevent contractors from sharing trade secrets and proprietary information from their former employers, but he acknowledged that it’s possible “there are things that happen,” given the scale of the operation.
But perhaps the greatest drawback of working as an AI trainer is that the very expertise you contribute to these systems may ultimately render your role — or even your entire profession — obsolete. By teaching machines to perform tasks that once required years of professional experience and specialized human judgement, workers are simultaneously helping to build the systems that could replace them outright. And yet, when faced with the seemingly unstoppable rise of artificial intelligence, many are taking an “if you can’t beat them, join them” approach, betting that training these systems may be the only way to not get left behind.
Frequently Asked Questions
Can I earn money training AI?
Yes, several companies are hiring people to train AI models through both full-time and part-time work. These opportunities range from basic data annotation and content moderation to fact-checking and teaching more specialized skills. Compensation can vary from a few dollars an hour to six-figure salaries, depending on the job and expertise required.
How much money can I earn training AI?
It depends on the job. Simple tasks like basic data labeling usually only pay a few dollars an hour. But if you have a specific expertise — like medicine, law or engineering — you can earn much more than that, sometimes hundreds of dollars an hour or even a six-figure salary if it’s a full-time position.
What skills or experience do I need to work an AI trainer?
For simpler tasks, you just need to be able to pay attention to detail and follow basic instructions. For higher-paying roles, you need experience in your field and the ability to convey that knowledge so an AI model can learn from it. Any technical training needed to work with the AI model itself will likely be provided by the company hiring you.
Where can I find AI training or data labeling jobs?
You can find basic data annotation work on sites like Amazon Mechanical Turk, Remotasks and CrowdGen. More specialized roles are available at companies like Mindrift and Alignerr. Mercor and Scale AI offer some of the most lucrative AI training gigs in the industry.
