top of page

Questions to Ask Before Taking an AI Training Side Gig as a Doctor

  • Jun 29
  • 8 min read

Increasingly, physicians in our online communities for doctors are being recruited for and accepting side gigs with AI labs to train AI models. While there is lots of interest in this work, reviews of the experience are always mixed, as the industry is evolving constantly and the rapid growth in intermediaries results in varying levels of expertise and professionalism in working with physicians. Many labs are used to working with demographics very different than doctors, and don’t understand the unique nature of working with physicians. It’s best to ensure that both you and the hiring company have clear expectations of what you’re being asked to do, how you’ll be paid, and more. However, we’ve seen from repeated discussions on the group that the nature of the space doesn’t always lend itself to this, and you've got to be okay rolling with the unexpected as the projects change. Below, we’ll outline what questions to ask before taking a side gig working with AI labs to train AI models to minimize frustration with the experience, and more importantly, to set yourself up for success.


Disclosure/Disclaimer: This page contains information about our sponsors and/or affiliate links, which support us monetarily at no cost to you. These should be viewed as introductions rather than formal recommendations. Our content is for generalized educational purposes.  While we try to ensure it is accurate and updated, we cannot guarantee it. We are not formal financial, legal, or tax professionals and do not provide individualized advice specific to your situation. You should consult these as appropriate and/or do your own due diligence before making decisions based on this page. To learn more, visit our disclaimers and disclosures.


Questions to ask before accepting an AI model training side gig


What questions should I ask before accepting a side gig that involves training AI models?


It’s important to understand that this space is very heterogeneous, and the experience level of different vendors will often dictate your experience. Many physicians understandably get upset by what they perceive as bait and switches in the space, so the first thing to know is that you’re going to need to be patient as the industry figures out how to best do these projects and work with physicians.



What am I actually being asked to do when I’m told your company is hiring physicians to help train AI?


AI projects come in many different forms, and are evolving rapidly to increasingly complex tasks. While a lot of AI projects initially were very straightforward labeling tasks, they now include things like:

  • Looking an an AI generated answer to a medically related question and evaluating it for accuracy or for how it could have answered things better

  • Identifying flaws in the model that could result in unsafe recommendations, or areas where an LLM may be hallucinating 

  • Comparing different AI generated responses to the same question and ranking them in order of best to worst

  • Writing answers to complex questions based on standard of care

  • Creating actual prompts that would pose challenges to models and seeing how they do or where they fail

  • Reviewing the reasoning process used to generate responses to questions

  • Identifying appropriate sources for models to pull data from 

  • Reviewing auto-generated reports or interpretations on studies, such as in radiology or pathology 

  • Developing rubrics to assess models

  • Making sure the model avoids worst case scenarios by putting in stopgaps


Related PSG resource:


What liability insurance might I need to perform AI training work?


The good news is that most of these gigs do not use real life clinical cases or data, and you’re not typically asked to review actual charts, see patients, or give personalized medical advice. Usually the models are pulling from fake scenarios that have been synthesized based on use cases, or otherwise using aggregate de-identified data.


As such, you likely don’t need traditional malpractice since you're not practicing medicine or establishing physician-patient relationships, but rather evaluating AI outputs. That said, some gigs may require errors and omissions type coverage or have other insurance requirements. Make sure that you ask relevant questions and do your own due diligence. In some cases, your contract will specifically allude to this or indemnify you against claims. Ensure that your responses or work are not specifically tied back to you.


Related PSG resources:



What is the hourly pay for training AI models, and am I paid for my own training for the job?


A lot of these gigs are advertised as “up to x/hour.” Unfortunately, many of them change rates if they have a lot of physicians signing up, and we’ve had several physicians express frustration that they signed up when a gig was being advertised at one rate, only to find out that the contract offered to them is presented at a much lower rate. We don’t condone this at all, but it happens not infrequently.


You’ll want to ask:

  • Is my hourly rate fixed throughout the length of the relationship?

  • Am I paid for my time training on how to do the job? This is important as training for these positions can take many hours, and if you’re not good at the tasks, the company may decide not to move forward with hiring you for a project.

  • Are there different rates for training and projects, and what determines the rates?


Rates are often based on things like level of expertise, board certification, specialty, whether you’re actively licensed and practicing medicine, publications or academic appointments, other subject matter expertise, prior experience with AI training, and prior evaluation scores on previous projects you’ve worked on.



How will I get paid for my work on this AI training project?


In most cases, you will be an independent contractor hired by a third party that in most cases is not the lab you’re directly working with.


Each company pays differently, but most will either pay:

  • By the hour

  • Per accepted task that you take on

  • By a milestone you hit or achieve


Because this work is done asynchronously, many of these companies have technology in place to ensure that you’re really working during this time, and that you’re focused on the task at hand. This may involve browser extensions or activity monitoring. They may also ask for timesheets and that you hit minimum quantity scores.


Often times, they’ll have you set up a payment system when onboarding (at which point you’ll be asked to provide a W9). They may pay through ACH through your bank account or through a payment system like Stripe, PayPal, or other platforms. Related PSG resources:



Is work actually available?


It’s best to set expectations here, since a high hourly rate means nothing if there’s barely any work available.


While you may think they wouldn’t be recruiting unless they actually had a project on hand, some of these companies are now onboarding physicians so that they’re ready to go when a project drops. You’ll have to decide if this is right for you, because it means that you may have to give them the information above like social security numbers and bank account numbers to onboard with their system, before any work has actually been offered to you. Be very careful about providing this information unless you have a high degree of confidence in the company you’re giving information to. This can be particularly hard in this industry given that most of the companies in the space have only been around for a short period of time.


If there’s not work available at the moment, ask about the frequency of projects in your field, how many hours others in your specialty averaged in prior months, whether there’s a backlog of work, and how projects are doled out (first come first serve versus qualification system).



What are the time commitments for this particular AI training project?


Projects vary widely in the number of days, weeks, or months that they’re expected to take (and heads up, a project that they say will be going on for months may be abruptly cut short a few days in). Some labs ask you to be available for fixed hours, whereas others let you work as much or as little as you want. If they’re moving fast on something, they may ask you to guarantee availability for a certain amount of hours per week.


Most work tends to be asynchronous, although they may offer office hours or regular reviews to make sure the quality of the work and the output is what it needs to be.



Is there a qualification process for this AI training gig?


In many cases you’ll be given a training or a guide that you’ll be asked to read through outlining expectations for the project and how to do the job. They’ll likely have you try doing simple tasks first and then increasing the complexity of the tasks gradually to make sure that you understand the work and that you’re able to complete tasks in a timely and efficient manner, as well as accurately. This work is not for everyone. If you struggle with technology, it may not be a great fit. 



How is quality assessed in an AI training environment, and what are the consequences of low scores?


Since how much work you’re given often depends on how good they feel you are at the work, it’s important to understand the metrics by which you’ll be assessed, and how they will be assessing you. Many platforms are continuously scoring you as you do your work, so you want to know what those scores are.


You’ll want to know what opportunities they have to help you are if you are getting lower scores, what the implications are for pay on hours done if there are lower scores, whether there’s an appeals process if you get low scores, and under what conditions you get kicked out of a project.



What happens if an AI training project ends?


Projects come and go quickly in this space. While you may have been given one set of expectations, there are no guarantees. As such, you’ll want to know what next steps will be, so hopefully you didn’t onboard for very little upside. Ask things like:

  • Will I be automatically considered for other projects?

  • How long will my profile be active?

  • How can I find out about other projects to apply to?


Know that almost every AI project will require you to sign an NDA or other confidentiality agreement that will prevent you from discussing prompts or sharing screenshots. They may ask for return of related materials.



Conclusion


The decision to take on an AI training project as a side gig can be complicated. If you’ve decided to pursue one, we hope these questions will help you better vet opportunities that come to you. This is a young and evolving industry, and many physicians encounter frustrations with it, but it can also be very rewarding, both financially and professionally. Just do your due diligence, and understand that there may be some unexpected twists and turns with these projects!



Additional AI related side gig resources for physicians


Sign up for our weekly newsletter for alerts about upcoming or available AI training projects, as well as our other content, including new blog articles, physician educational webinars, and member resources.


Related PSG resources:

bottom of page