Many of the most successful and widely used machine learning models are trained with the help of thousands of low-paid contract workers. All over the world, millions of crowdworkers earn their living on platforms like Amazon’s “Mechanical Turk” by performing countless small tasks that companies and researchers outsource to them. In the US, around 250. 000 people generate at least three quarters of their income this way . But even though they work for the richest artificial intelligence (AI) laboratories in the world, they receive far less than the minimum wage.

Saiph Savage, director of the human-computer interaction laboratory at West Virginia University, wants to change that. At the beginning of December, in a lecture at “NeurIPS”, one of the world’s largest AI conferences, she promoted an improvement in the working conditions of these gig workers.

Technology Review: Why did you have the AI Community told about the “invisible workers”? Doesn’t she know anything about it?

Saiph Savage: Because you need to advance your research with the work of these workers and realize that a self-driving car or something like that Things only exist because people don’t get a minimum wage. While we are thinking about the future of AI, we should be thinking about the future of work. These workers are people.

What types of work do companies and researchers seek help with?

Often it is about data labeling, especially of image data, so that monitored machine learning models understand the world better. Or transcribing recordings. When you speak to Alexa from Amazon, for example, crowdworkers often transcribe what has been said so that the speech recognition algorithm learns to understand language better. I also recently met with crowdworkers in rural West Virginia. They are hired by Amazon to read Alexa dialogues to help her understand how people in this region are speaking. Others flag websites that may be filled with hate speech or pedophilia. That’s why you won’t come across these things when searching for images in Google or Bing. These tasks are often assigned on platforms such as Amazon’s Mechanical Turk. Big tech companies like Facebook and Microsoft also have their own. In contrast to the Mechanical Turk, anyone from researchers to startups can use this.

What is the problem with that?

I don’t see crowdwork as a bad thing in itself. It gives companies access to external employees. However, for a study we followed hundreds of Mechanical Turk employees for several years and found that at two dollars an hour they earn far less than the minimum wage.

Nevertheless, these platforms are the main source of income for many. The next problem is that these platforms prevent career opportunities as full-time crowdworkers cannot develop their skills. Many also do not mention the platform work on their résumé. Most employers don’t even know what it is and that these employees are behind our AI.

Where does your great interest in the topic come from ?

I was doing a research project at Stanford University where I was basically a crowdworker and faced the same problems. I helped design a new platform that was similar to Amazon Mechanical Turk but was controlled by the workers. But I was also a technician at Microsoft, which also opened my eyes to work in a large technology company. You become faceless and that is very similar to the experience of crowdworkers.

How do we solve the problem?

In many ways. I help workers get an idea of ​​how long a task might take. In this way they can assess whether a task is worthwhile. Therefore, I developed an AI plug-in for these platforms that employees can use to exchange information and coach each other as to which tasks are worth their time and with which they can develop certain skills. The AI ​​learns what type of advice is most effective. It picks up the text comments that employees write to each other, learns which advice will lead to better results, and advertises them on the platform.

Let’s say employees want to raise their wages. The AI ​​determines what type of advice or strategy is best suited to help workers with this. For example, it might suggest that you do these types of jobs by these employers, but not them. Or you may be asked not to look for work for more than five minutes. The machine learning model is based on the subjective opinion of Amazon Mechanical Turk workers, but I’ve found that it could continue to raise workers’ wages and develop their skills.

How can this training be promoted?

For example, we could guide employees to do a number of different tasks with which they can develop their skills. Companies offering jobs on these platforms could offer online micro-internships for workers. We should also support entrepreneurs. I’ve developed tools that people can use to create their own gig marketplaces. This makes setting up a platform a bit similar to configuring a website.

(vsz)