How one company hopes to alleviate poverty in India with “ethical data”

It can be easy to overlook the people behind all the technology we use. But a startup based in India called Karya is putting them front and center, both in its method and its marketing.

Karya’s stated goal is to alleviate poverty for Indians living in low-income communities by paying them approximately $5 per hour, a wage that’s higher than the market rate, to create data.

It works by sending text prompts to Karya workers through a smartphone app. Those workers then record themselves speaking their native language and send the recordings back to Karya. After the recordings have been quality checked, they become the foundation of massive datasets, which Karya sells to tech companies, universities and others, who use them to build AI models.

Marketplace’s Lily Jamali spoke with Vivek Seshadri, Karya’s chief technology officer and co-founder, about how his company fits into the lucrative data collection business.

The following is an edited transcript of their conversation.

Vivek Seshadri: Karya is a worker first company, and all our co-founders strongly believe that AI, while it is empowering all the world with the new technologies and its capabilities, also has the potential to empower people on the ground by giving them not just supplemental income, but also access to the technology, which they may never get to use. Today, we pay the highest wages in the data market, $5 an hour, for just recording sentences on their phone and transmitting the speech data to us.

Lily Jamali: Karya doesn’t employ these workers long term, it’s supplemental, right? Sort of like gig work?

Seshadri: It is gig work, and we want to keep it that way. The mission for Karya is to create pathways out of poverty. It’s not to employ people in data work. So, we see this as a first step in that process and as we speak, we are creating two other programs, one called Karya Learn, and another called Karya Grow which essentially cater to that goal. So those two programs are in progress right now and our goal is to, through the data work, get the workers to a milestone and then graduate people into these two programs.

Jamali: A big part of your pitch is the idea that you’re collecting this data ethically. Can you talk about what that means? What is ethical data?

Seshadri: Ethical data to Karya means two things: one is do not exploit, and the second is to empower.  Gig economies typically tend to become exploitative, because often the amount of skill that is needed to complete the gig work is very low and as a result, it can very easily fall prey to a race to the bottom kind of situation. So, to prevent ourselves from falling into that trap, we have essentially set our organization up as a nonprofit ecosystem wherein our goal is not to make more money. Our goal is very simply to enable wealth transfer to people in low-income communities. We have to ensure that workers are paid ethical wages for the work that they’re doing and ensure that any work they’re doing is not harming them in any way mentally or physically.

We feel there is an extended pathway for empowerment. Today there are lots of laws that are being passed around private data. Unfortunately, such legal frameworks do not exist for data generated for training AI models. We want to essentially move towards an ecosystem, wherein the value that is generated by AI services today is proportionately shared with the people who generated the data that enabled the service in the first place. We are essentially working towards incremental steps towards reaching that goal.

Jamali: And would you say that the ethical data model is scalable? Or are we going to find out with you?

Seshadri: We will definitely find out with us, but our strong belief today is that it is scalable. In some sense, that is the only pathway forward. As a society, we have to ask the question of what sort of ecosystem we want to build for all of society, and not just the top one percent of the population.

Jamali: You’ve laid out a goal for your company to reach 100 million Indians by the year 2030, which is very ambitious. As you grow, I’m wondering how a company like yours can make sure that you don’t stray from this ethical mission to something more exploitative? This is an issue that a lot of tech companies before yours have faced and continue to struggle with.

Seshadri: Let me begin by saying that if Karya ever had to stray from these principles to sustain itself, that’s the day I would call the Karya experiment a failure. In some sense, we would rather fail rather than turn into an exploitative company. We strongly believe that what we are setting as a goal for ourselves is achievable because we see a huge gap in the amount of value a piece of data is generating and the wages that workers are paid on the ground. All our mission needs is for the ecosystem to realize that and become ethical as a whole. We all benefit from the power of AI. We all benefit from the efficiency that AI adds to our lives, and I think it is time we ask the question, where is the data that is fueling this AI coming from? And what is the standard of living of those people? How can I ensure that they are part of the ecosystem and benefit from the ecosystem and get empowered by the ecosystem, rather than become exploited?

More on this

In our conversation, Vivek Seshadri mentioned he wants to ensure that the tasks Karya sends to workers don’t cause mental or physical harm. Some big tech companies like Meta and OpenAI, the company responsible for ChatGPT, have been criticized for outsourcing some content moderation work to developing nations like India, Kenya, and The Philippines. In those content moderation roles, workers are sometimes paid less than $2 per hour to view and label graphic and violent content.

On a lighter note, if you’re interested in learning more about Karya’s origins, check out the profile of the company and it’s other co-founder, Manu Chopra, in Time Magazine last month.