Why ethical pay is critical to the future of AI
From autonomous vehicles to chatbots, AI holds a pivotal role within modern tech. But what about the relationship between AI and data ethics? Many of the tech giants behind AI applications, like Meta and Google, are household names. But most people don’t know about the participants who train the AI models that power them.
The workings of AI remain unclear to most people. Some misconceive it as being “intelligent” like a human. In fact, it’s processing huge quantities of data very quickly. To do this, AI systems need extensive training.
This important task is often left to low-paid workers from developing countries. This, in turn, raises many ethical issues.
Why is ethical pay so important?
In the field of AI research, ethical pay is critical. There are two important reasons for this.
Participant wellbeing
Critical AI training tasks, like data-labeling of images and text, are often outsourced to workers in the Global South. Researchers and companies post requests for these jobs on online crowdsourcing platforms. Arguably one of the most well-known platforms of this type is Amazon’s Mechanical Turk.
Some firms have framed their outsourcing as “impact sourcing”. They argue that they've provided new opportunities for workers in poorer countries. However, many of these roles provide low pay and little opportunity for career progression.
What's more, participants frequently outnumber jobs on crowdsourcing platforms. This leads to high levels of competition and lower wages. Many platforms, including Amazon’s MTurk, don’t request a minimum pay rate for participants. In many cases, participants have earned far below local minimum wages. Speaking to TIME magazine, employees from AI outsourcing company Sama claimed that some participants took home only $1.50 per hour.
Data quality
When researchers underpay participants for high-turnaround tasks, they put data quality at risk for highly important work. Participants may try to complete a task quickly to move to higher-paying tasks. This could lead to them adding inaccurate labels to the data, for example.
This is more likely to happen when participants are rushing or not feeling invested in the work. In turn, this leads to poorer data quality for researchers. To encourage accurate data, participants need autonomy, respect, and real compensation.
AI and data ethics – key solutions
AI models don’t have to be trained using unethical practices. In fact, it’s vital to the future of AI that they aren’t. Several key factors can ensure ethical treatment for participants.
Fair pay
Research has found that remuneration is the main motivator for participants to provide high-quality data. Fair rates encourage participants to be attentive when giving answers or labeling data. For a higher-paying task, they will be more likely to spend time providing accurate responses. Meanwhile, low-paying tasks may lead participants to rush.
At Prolific we’ve implemented a minimum hourly rate for participants (£6 per hour). We also have a recommended hourly rate of £9 per hour. Need help calculating your reward? Use this online calculator to find the right remuneration for your participants.
Regulated research
Another key issue is the poorly regulated state of online crowdsourcing platforms. Many websites offer participants little to no recourse to challenge research clients. Most of these sites don't have a minimum rate, either. This means participants aren’t guaranteed a minimum level of pay.
At Prolific, we pause any underpaying studies so the researcher can adjust the pay to an appropriate amount. This ensures underpaying studies are not kept on the platform to exploit participants.
Freedom to choose
Participants should have a level of control over the tasks they choose to work on. In practice, this can be challenging to achieve. Many participants want to take part for financial reasons. On some platforms, participants often far outnumber available tasks. If competition is high, participants may feel obligated to choose any study available.
We want to keep competition to a minimum. So, we've restricted the number of participants on the platform to ensure there's enough work for everyone.
Provide support
Unfortunately, on some platforms there have been accounts of researchers engaging in underhand behavior to exploit participants.
This is an issue that has recently been raised with Amazon’s MTurk platform. In one example, a company placed over 70,000 tasks on the platform. They then rejected all completed work from participants. This means the company would continue to have access to the work but was under no obligation to pay the participants. MTurk participants usually have the option to negotiate with clients. In this case, they received little to no support from Amazon.
On Prolific, participants always have a very clear point of recourse if this kind of situation arises. They can easily report the behavior to our support team, who are on hand to help.
The future of AI
AI will continue to play an important role in society for years to come. To stay sustainable, those who train AI systems must be fairly compensated for their time.
So, how can you ensure that your research project meets the guidelines for ethical AI? First, you need a clear understanding of the relationship between AI and data ethics. Download our AI ethics eBook for everything you need to know about AI ethical principles and ethical data collection.
