AI starts with data. Clean, curated, labeled, ready-to-plug-into-your-models data. But it’s not always so easy to get your hands on a dataset that you can use to quickly and efficiently push your models into production. That’s why the individuals who focus on data labeling are so important, and why Sama is making huge efforts to cultivate the next generation of AI specialists.
Rather than sourcing from the world’s top universities or boot camps, Sama has spent several years delivering a course entitled Artificial Intelligence 101 to individuals in Kibera, the largest slum in Africa. This course teaches students the basics of AI work, simultaneously training AI talent and taking aim at systemic poverty.
To discuss the results of these efforts, Sama’s CEO Wendy Gonzalez joined our podcast, How AI Happens. Wendy explains:
🖥️ Why providing work is a more impactful approach to ending poverty than providing aid;
🎓 How Sama’s program has measurably improved women’s income and employment rate by 60%;
✏️ How data labeling skills provide talent with access to employment in AI and beyond.
Stream the full episode below, and if you’d like to see the nuts and bolts of how Sama’s mission has affected the community, be sure to check out MIT’s recent RCT study measuring the impact of AI education on student employability and income.
And don’t forget to subscribe to How AI Happens on your favorite podcast streaming platform!
0:00:00.0 Wendy Gonzalez: Talent is distributed equally, but opportunity is not. And the best way to solve poverty, which is basically the root cause for every major social ill in the world, is by giving work, not giving aid.
0:00:16.8 Rob Stevenson: Welcome to How AI Happens, a podcast where experts explain their work at the cutting edge of artificial intelligence. You’ll hear from AI researchers, data scientists, and machine learning engineers as they get technical about the most exciting developments in their field and the challenges they’re facing along the way. I’m your host, Rob Stevenson, and we are about to learn how AI happens.
0:00:48.1 Rob: Today on How AI Happens, we’re going to talk about the next generation of AI specialists. While this might bring to mind the image of a starry-eyed Stanford PhD or a youthful self-taught prodigy, I’m thinking of a different source of talent, because over the last several years, Sama has been delivering a training program entitled Artificial Intelligence 101 to eager individuals in Kibera, the largest slum in Africa, just outside Nairobi, Kenya. To learn more about how an AI company can train technical talent while simultaneously taking aim at systemic poverty, I sat down with Wendy Gonzalez, CEO of Sama.
0:01:29.5 Wendy: I have, I’m a little embarrassed to say, over 25 years of experience [chuckle] in enterprise technology and SaaS and AI. I started my career in management consulting, really helping large companies figure out how to leverage technology. I switched over to the enterprise side, working to really implement disruptive technologies. And then I co-founded an Internet of Things startup about a decade back to build a SaaS platform, because again, IoT was a way to really disrupt the industry, and then I switched over to AI. So I joined Sama in 2015, and Sama is the leading trainee data as a service platform. We work with the Global 2000 to really help enable their mission-critical AI applications.
0:02:13.1 Wendy: I was really compelled to join Sama because I believe deeply in Sama’s mission, and really, our philosophy that Sama was founded on is that talent is distributed equally, but opportunity is not. And the best way to solve poverty, which is basically the root cause for every major social ill in the world is by giving work, not giving aid. Over a trillion dollars’ worth of aid has been donated to sub-Saharan Africa since the 1960s, yet, GDP hasn’t changed. While it seems like the right thing to do, the best thing that we can do to really solve the problem of poverty is by providing financial independence, and that’s through giving work. And so, Sama’s mission of purposely hiring people in underserved communities to give work was something that really spoke to me, because as a child of immigrants and marrying my husband who was the first person in his entire family through all generations to go to school, the power of work is transformative in terms of lifting up your community and lifting up the people in your family.
0:03:11.4 Wendy: So that mission spoke very, very deeply to me. Sama had a really unique and audacious view on this, which is, it’s not just about kind of paying living wages and providing employment and benefits; it’s about purposely hiring people who’ve got the greatest barriers to employment. And so, Sama’s model is to hire, in underserved communities, 50% women, 50% youth who have household incomes of less than $2 a day, which is the World Bank standard for poverty. And so I was fascinated with this idea of taking purposeful action and change to hire people and not just sort of provide wages, but really provide a transformative career path to hopefully break the poverty cycle permanently.
0:03:52.1 Rob: This notion of giving work rather than aid has taken the form of Sama’s Digital Basics Program, that’s the AI 101 course I mentioned earlier. As Wendy explains, this training is the first step in removing the barriers to access between eager, underserved workers and well, work.
0:04:12.8 Wendy: One of the challenges is not just about, “Hey, I’m in this situation, I didn’t graduate from a fancy school or college,” it’s also by just having access to the network to get jobs. So I say barriers to employment, it’s not just kind of education and where I live, but it’s like, “Do I know the right people, how do we even get connected to a job?” So the Sama Digital Basics Program really started by working with community partners in underserved communities, I’ll give you an example in Nairobi, to where we work with partners like the Human Needs project that has a facility that is embedded in the Kibera slum, which is the largest slum sub-Saharan Africa. So, it was an idea to bring digital skills training into communities, so that, again, to provide and reduce that barrier to access.
0:04:57.7 Wendy: And so what the Digital Basics Program does is it provides everything from basic skills like mounting and keyboards, but also the basics of AI. So what is artificial intelligence, what role does training data play in empowering artificial intelligence applications, and it’s really the initial training that is necessary for somebody to come and join the Sama program.
0:05:20.2 Rob: It should be pointed out here that the goal of the Digital Basics Program isn’t merely to create a farm team of future Sami employees, though many of them do wind up working there; rather, the skills that go along with data labeling end up providing individuals with a much wider breadth of opportunity.
0:05:38.4 Wendy: It’s not just about building training for the purpose of, “Okay, now you can do data labeling,” what we found, if they’re doing this program ’cause we were actually launched in 2008, is that the skills necessary to do labeling and tagging are critical thinking skills that actually apply to many different jobs on a go-forward basis. Our intention was always not just hiring people into Sama and you’ll be with Sama for the rest of your lives; it was really about building the skills that allow our workers to go on to higher-paying jobs, return to university. That’s the entire idea of when I say permanently breaking the poverty cycle, is to build the foundation so that people can move on to higher-paying jobs.
0:06:16.1 Wendy: So while we have a pathway, of course, to move up within Sama, the other objective of this is to build a core set of technical skills. So the training typically occurs within community or at our offices, and then of the people who are trained, some go on and move on to other jobs, many come and apply at Sama. After doing this level of training, we would hire people in as employees, so we’re different, we are not freelancing, we are not crowdsourcing. A part of our mission is to provide living wages, benefits, and professional development that’ll allow people to further their careers. And so they get hired into Sama, and typically, they will do data labeling as an entry point. And so data labeling, just to take us back a little bit in terms of artificial intelligence, is that…
0:07:03.7 Wendy: You think about it this way, AI is as intelligent as the trainee data it’s built on, because machine learning is all about recognizing patterns, then using deep learning techniques for the application to make decisions. Put another way, before a car can drive itself, like a self-driving car, it needs to be able to detect roads, pedestrians, vehicles, and traffic signs, and training data is basically structuring the data so that a computer vision application can identify, what is a car, what’s a drivable space, what’s a road sign, et cetera. And while that sounds like it’s relatively simple, it’s actually incredibly, incredibly complex. Some of these data labeling activities include data labeling in 3D. I don’t know if you’ve ever seen radar or light or images, it’s very, very complex. And then beyond that, it’s not just about detecting and after being that information correctly; it’s about the precision of it. And so, tagging a car is not just tagging a car, sometimes you need to include the side-view mirrors, maybe you need to include the shadow under the car, maybe you do need to include what’s behind the car, maybe you don’t.
0:08:05.3 Wendy: So there’s actually quite a bit of complexity, and so, in terms of the types of work and the skills that are being built, you have to tie it back to a taxonomy, there could be business rules. So, a part of what’s being developed as critical thinking skills as well.
0:08:22.9 Rob: As I mentioned before, this program has been underway for years, so, this podcast episode isn’t meant to be an audio press release. The reason I wanted to bring Wendy on to discuss the program is because of a recent study measuring the effects of these efforts, a six-year randomized control trial conducted by MIT and Innovations for Poverty Action, a research and policy nonprofit promoting effective solutions to global poverty. Wendy explained some findings from the report, and whether those findings were in line with initial goals.
0:08:55.4 Wendy: The thing that we are really trying to understand or MIT was trying to understand is this purposeful hiring model that this part of Sama’s mission, that’s core to the way that we operate, is that intervention of purposely hiring somebody from an underserved community, does that actually improve their employability and their income rates in the long term? Does it actually break the poverty cycle that I was talking about before, or would these people just would have succeeded anyways?
0:09:18.5 Wendy: So when we talk about really, how do you measure that, it’s called a counterfactual study, which, when friends ask, “Well, what do you mean? What is an RCT? Why did you do this study”, it’s kind of like, the FDA does it for drug approvals, right? Somebody takes a placebo, somebody takes the medicine, and then you find out at the end, well, did the medicine really work? And so that’s really what we’re trying to do here. Did this purposeful intervention of hiring, did it make a difference in somebody’s employability and growth and income?
0:09:46.0 Wendy: The study wasn’t just a matter of, “Hey, let’s take some surveys.” It’s actually even something in the works for six years. So we plan for a year, a year and a half, to identify folks and create a very detailed study, three years of actually surveying people on a regular basis. So, fun fact, over 2000-plus hours of surveying time [chuckle] and interviews and calls. But at the end of the day, the idea was to say, “Did the intervention work? Is there a meaningful and material difference, and do we have all over the data over time to prove it?” I believe all of our internal tracking that yes, we were, so it was amazing to get that ratified, that, yes, indeed, over the long term, this purposeful hiring model made a huge difference in terms of income and employment levels. But in particular for women, that was the thing that was really exciting to learn, and that was a little bit surprising, is that Sama’s model has been to create a purposeful higher model of 50% women and 50% youth.
0:10:48.0 Wendy: And we focused there from a mission perspective, because what I think has been well-researched and understood is that when women are lifted up in the economy and when women are lifted up in income, they contribute back to their communities, more kids go to school, there’s a real network effect of investing in women. And so that was part of our purposeful hiring model, to where we have a criteria in hiring at least 50% women, which is a little bit unique, and again, purposeful. And what we found is that after this three-year time frame, there was a 60% improvement in employability and income rates for women who went through the Sama intervention versus who did not. And that was really exciting. Really, really exciting to see that that basically means that our hypothesis that women have barriers to entry and employment is indeed true, but that the purposeful hiring model at the beginning of their careers actually makes a difference in them not only having greater income but continuing to stay employed.
0:11:46.5 Rob: When you look at the impact these studies showed with women specifically, was that surprising or just delighting?
0:11:54.7 Wendy: It was delighting. We did a lot of community surveys before launching this program, because let’s not just assume we know what people need, that’s the entire point of financial independence, let’s hear from them what is needed, and what we found is that, in particular, it was just really challenging for women to get into the right networks, to get jobs. And so, this was the model between that and knowing that the impact that women can have when they are lifted up, but I think what was surprising is not just the income levels, but I think what surprised me was the employment levels. I mean, 60% is significant. So, yeah, I was delighted, [chuckle] to answer your question.
0:12:34.3 Rob: Of course, the ability to do a job and the ability to get a job are vastly different skills. This was reflected in the report, as researchers found subjects who did not receive employment placement assistance had a much harder time finding work, even if they had the same technical training as others. So, does that mean mere up-skilling is not enough?
0:12:57.2 Wendy: I think what that speaks to is that while training is incredibly important, that action, the purposeful hiring, is really kind of what moves the needle. I’m all for providing the training, and I think what we found is that what we’re trying to do is, in addition to the core skills training, is gonna be, how do we help people match themselves to jobs, and there’s some pretty incredible organizations out there who are specializing in this area, so we don’t assume we’ve gotta figure it all out. [chuckle] We are working with partners to help us on things like the skills matching. But the key thing, you have to make those purposeful choices, so, oftentimes, I talk to our customers about the ethical AI supply chain, or really how you use impact criteria as part of how you make your buying decisions. And I think that’s something that’s incredibly important, because I know we’re going to be a proof point and move tens of thousands and over time, hundreds of thousands, of people out of poverty, but imagine the world’s biggest corporations are spending trillions of dollars in procurement. Imagine what they can do if they make those same purposeful hiring decisions. We can leave millions and tens of millions of people out of poverty.
0:14:08.6 Wendy: We’re doing this at Sama, but I would love for every company in the world to take the same approach, and if you’re not in a position to make these purposeful hiring decisions, well, work with suppliers, use social impact criteria as part of how you make your buying decisions. So the more we can get this out there, that, “Hey, the model works,” makes a meaningful difference, it also has tremendous business benefits as well. What happens for us, we have incredible retention, and that has created more value in our AI platform. So there are many, many different reasons to take this approach, and from a social mission standpoint, we can build incredible technology, create incredible value from our products, and we can change lives at the same time.
0:14:51.5 Rob: What do you foresee in terms of additional education beyond data labeling? When you look at further development for their up-skilling, do you foresee that being an offering over time?
0:15:00.8 Wendy: Oh, yeah, absolutely. I love that you mentioned that. I hate to sound buzz wordy, if you will, but I mean, data is the new code. The skills that are being built here aren’t just for, “Okay, I can label”; it’s really about building not just those critical thinking skills, but the way that we move our workforce up the value chain is that they are building analytical skills. So gone are the days where our workforce would come in and be like, “Oh, is there a dog in this picture?” or something like that. They’re doing way more sophisticated work, doing everything from training machine learning models and driving very sophisticated training data sets, to evaluating and quality sharing training data sets that have been produced by automation, all the way to identifying what’s missing, do we have a representative and complete data set? So it’s really about moving from labeling and taxonomies and workflows to analytics and beyond. So, yeah, there’s a lot of exciting work ahead, and beyond the analytics and management, yes, data science, that is the next frontier.
0:16:08.7 Rob: If data science is the next frontier, then any company seriously deploying it has an awesome responsibility, not just to make a great product, but to utilize this technological position to up-skill the next generation of talent, and do so in such a way that ensures representative, more diverse, more creative future for our industry.
0:16:31.6 Wendy: Next time on How AI Happens:
0:16:38.5 Speaker 3: There’s a pretty substantial rules-based expert system that sits on top of this to help manage some of the downsides of the inherent biases we have in the data.
0:16:53.0 Wendy: How AI happens is brought to you by Sama. Sama provides accurate data for ambitious AI, specializing in image, video, and sensor data annotation and validation for machine learning algorithms in industries such as transportation, retail, e-commerce, media, med tech, robotics, and agriculture. For more information, head to sama.com.