Travis Khachato...: You are listening to Sagecast, the podcast of Pomona College, where this season we're talking with faculty and alumni who are taking on challenges and tackling the difficult. I'm Travis Khachatoorian. Marilyn Thompso...: And I'm Marilyn Thompson. Recently, Chinasa Okolo, class of 2018, came to campus to talk with students studying computer science. She's working in the Brookings Institution, researching the rapidly growing field of Artificial Intelligence. Travis Khachato...: Chinasa is taking special interest in Africa and how leveraging AI can advance global health. Here's that conversation. Chinasa, welcome to Sagecast. Chinasa Okolo: Thank you. Very happy to be back. Travis Khachato...: It looks like your main area of research is Artificial Intelligence. I'm very curious as to what led you down that path? Chinasa Okolo: Yeah, so I would say my interest just grew as a computer science major here at Pomona. I really got to learn more about AI and also the implications of AI through a senior project that I did for my major. Through that year, I had the opportunity to work with Professor Alexandra Papoutsaki. She was the first professor in the department to really focus on human computer interaction, and I had the opportunity to work with the system that she developed throughout her PhD. And so basically the system was focused on eye tracking, but to track eyes, it first had to track the face. And so when I tested it out on myself, I found out that it actually wasn't really tracking my face super well unless I happened to be in direct lighting. And so during that time also, there was a lot of research that came out surrounding the implications of face tracking and also premier work from Joy Buolamwini that showed bias in facial recognition systems. And so that work also had an impact on my journey in AI, and then also just how I chose to pursue the rest of my senior research as well. Marilyn Thompso...: What is AI? How would you explain AI to people who read it all the time, and then either jump immediately to Chat GPT, or go, "I know it's coming, I know it's here. I know it's important. I'm not quite sure what it is"? Chinasa Okolo: Yeah, so I would classify artificial intelligence as being a method of computer science that aims to make machines, like computers, phones, watches, for example, think and behave like humans. And the goal is really just to solve a variety of problems that include learning, identification, planning, reasoning and decision making. Marilyn Thompso...: A lot of what we read and hear in the US media right now about AI is either about how it's going to put us a lot of work, or it's going to do all the students' homework for them, or it's going to be doing ever more dangerous disinformation. Yet you're doing things that seem a lot more positive. Tell us about your work. Chinasa Okolo: So my work as a PhD student really focused on community healthcare workers. I refer to them also as CHWs as I talk, how these workers in rural India understand and perceive AI. And so this is really just important, just because in India there's a lot of technological innovation that's going on. A lot of it's from the government and also from different nonprofits and big tech companies that are based within the country. And so these applications are really targeting the work of community healthcare workers. But there wasn't, at the time that I did this research or started this research, wasn't a lot of knowledge about how do these workers understand AI and also how would they perceive its benefits and implications for real world use. And so I really became interested in addressing these questions as well. And then from this work, we found that, one, they had very low knowledge about AI, and then also that they had lots of misconceptions about how AI works. And we found this troubling, just because in the case that AI produces an incorrect prediction, we don't want them to go along with these. We want them to have the authority to actually basically kind of rebut against what AI is saying. And so that's how I also got into AI explainability, because we figured if we can make AI explainable to community healthcare workers and also just other users like them who have low levels of knowledge about AI, then we can make AI more useful and feasible in these environments that they're working in. So, yeah. Travis Khachato...: That's very interesting, because I never really thought about the development of AI turning into the world of haves and haves nots, and I would imagine just seeing doctors and healthcare workers in rural India really shined a light on that. Can you explain a little bit more about how we're going to build a more equitable AI landscape? Chinasa Okolo: Yeah. So I think there's definitely multiple parts to this. A lot of work recently has come out showing that the grunt work, the data work of AI is really being done primarily by low paid workers that are based in the global south or just in developing regions, let's say. And so people don't really realize that to have AI systems, you need a lot of data, but at first this data has to be collected, then processed, and which means labeling, cleaning it. But again, a lot of this work is being done by people who are not, one, who are being unfairly compensated for this work, and also being subjected just to very monotonous days of just sitting at a computer labeling data. And so, again, these people don't have a lot of knowledge, or strong technical knowledge about AI, and so their perspectives are not really included in the actual development. And from my knowledge, from what I understand, a lot of AI development is really focused more so in the west. And so by the west, I mean United States and Europe, et cetera. There's also a lot of development being done in China as well. But again, the perspective of people that live in the majority of the world is being left out. And so we've seen already that AI systems have been biased against underrepresented minorities within Western context like the United States, but also there are wider implications for its use just based on these present day challenges. Marilyn Thompso...: So you incorporated ethnography as you went out into rural India. What were some of the things you found? Were healthcare workers incorporating AI into diagnostics? What were some of the things, and how is that going to be something that we'll see around the world? Will it really enhance medical care, for example, and make healthcare workers more effective because they can reach more people? Chinasa Okolo: Yeah, definitely. And so my work really focused more so on prototyping, and so we weren't working with real world AI systems yet, but in India, there are actual systems that have been deployed and been rolled out, especially when it came to helping counteract the effects of COVID. And so for my work, first we examined pneumonia diagnosis, and then my most recent work, we actually traveled to India, focused on jaundice diagnosis. And so, again, these were both prototypes, and so really found out that while the community healthcare workers didn't have a lot of knowledge about AI, they were very enthusiastic about using it. And even though we mentioned to them that our work was just in very early stages, that they wanted to actually have these applications and even mentioned that they wish they could have AI for COVID, for example. And so other things as well is that they understood that even though AI is kind of being seen as something that can do their work for them, they mentioned that they still want to be, and they actually have to be involved, because, one, their patients may not necessarily accept AI, even though it may seem like a flashier, a very cool thing, and especially when it comes to very sensitive conversations, India is a very patriarchal society, and so it's hard for women to have discreet conversations about breastfeeding or just family planning and birth control. And so community healthcare workers are usually women, and so they have these conversations with women in their communities about this, and these women don't want to have to have such conversations with the an AI chat bot, they want to talk to an asha, that's what they call community healthcare workers in India, about these very sensitive conversations. And so there's still overall a very human aspect needed within AI, and these community healthcare workers realize it. And so I think that was just one of the most surprising things, and also I would say very endearing things I've seen within my work in India. Travis Khachato...: I'm curious how you ended up choosing India to focus your work, and what your experiences were like working in that country? Chinasa Okolo: Yeah. Let's see. So my advisors, I was kind of [inaudible 00:08:45] my PhD, but the people I primarily worked with, they both went to the University of Washington, Udub, for their PhDs, and they were in one of the premier labs for information and communication technologies for development, also called ICTD. And so this work really just focuses on improving the information and communication technologies for use in the developing world. And so for their work, they had been in India, all over different countries in Africa, doing very similar things. They weren't really working... one of my advisors did some AI projects as well, and so I think that interest just really came naturally. One, because they had the connections and experience, and then also just because it was just easy to integrate myself within that work and also kind of carve at my own path. And also one of my advisors is Indian himself, and so really helped because I don't speak Hindi, so he's super helpful. And also one of my lab mates is Indian, and he's very helpful in just the planning and the implementation part of my research project. And so interacting with the community healthcare workers, interviewing them about their experiences, even though I was in charge of the design. And overall the experience was very different from anywhere I've been. I actually do see a lot of parallels between India and Nigeria. I landed in Deli, and so I was like, it smelled very similar. I think I should say a smell of air pollution. But yeah, it was very interesting. I was in super rural India, but I think one thing that surprised me was that there was a very strong internet or phone connection. Because I was in upstate New York and sometimes an internet connection or my connection to my cellular provider could be very bad. So it was so interesting seeing that difference. And also my advisors did prepare me, or try to brace me in advance, because they mentioned that you'd probably be the first black person these people ever see. And so for me, it was kind of a little bit of celebrity-ish. People were taking photos with me and stares and all that stuff, but I became comfortable with it. And also my lab mate was very helpful in showing people who were trying to take pictures with me when we visited some... we visited, it wasn't a mosque, but it was just a temple. So I had a good experience overall. I got to visit a couple cities within the country, and overall I definitely would love to come back. Marilyn Thompso...: Now you're also researching AI in Africa and the impact of generative AI in Africa. First off, remind us of what generative AI is, and then how is it being accepted and used within Africa? Chinasa Okolo: Yeah. So generative AI has really come into the public purview over the past year, especially with the advent of tools like Chat GPT. So essentially it's the use of large models that have been trained on different kinds of data, whether it be image, text, videos, and the goal is to produce novel content. And so we're most really familiar with chat GPT, which is a text model. And also there are other models from OpenAI, like DALL-E, which are text to image models. And so this work is really preliminary, I just got it accepted as a poster at a conference, and so looking forward to building on the work. But basically I've seen a lot of it being used to help produce news articles, especially when it comes to things like elections. There also have been some cases that I'm investigating about the use of AI to produce... there was a case, presidential candidate in Nigeria, there was allegedly a faked call that was created, that was trained on his voice and a voice of a religious figure in Nigeria. And so I'm still investigating that, but there's also been cases of AI being used to generate marketing images as well. And so I'm really interested in the misinformation aspect of that, just because there are lots of implications for using these kinds of images in media. Travis Khachato...: Yeah. You mentioned Nigeria a lot, so I want to take a step back. Because I read that your parents are immigrants from Nigeria, correct? Chinasa Okolo: Yes, that's correct. Travis Khachato...: How has that shaped your perspective in everything that you've done? Chinasa Okolo: Yeah, definitely. I would say it's definitely had a very big impact on me, just wanting to acknowledge my heritage. Being in America, I think you kind of get swept up in the culture here, because it dominates here and also just around the world in general. But I think honestly, being at Pomona was the first time I was surrounded by other Nigerians, especially those that were born in the country. My best friend at Pomona was Nigerian born, and also just students from other African countries. I was also very involved in the Pan-African Student Association as a co-president during my senior year, and also just throughout my time at Pomona. And so just wanting to have that connection to the African continent was really important to me. And overall, I would say my interest in healthcare also kind of grew out of that, just because growing up I would always hear my parents talking about how family members were suffering back home, and a lot of the suffering that came due to the healthcare system and just being unable to get adequate healthcare. And so I think that also kind of drove my interest in healthcare, along with my parents also wanting me to be a medical doctor, but now I'm a doctor of computer science. So yeah, I hope that... Travis Khachato...: A different kind of doctor, helping doctors. Chinasa Okolo: Yes. Marilyn Thompso...: Yeah. How do you see AI transforming medicine? Chinasa Okolo: Definitely in a lot of different ways, both within rural context like India, which I focus on, and also in more developed context. So in rural healthcare, I would say there's also rural context within the United States and other "developed countries". So I think it really could serve as a conduit in just improving general access. And so telemedicine really arose throughout the pandemic, and I kind of see AI supplementing existing telemedicine services, by allowing people to get diagnostics done through meeting with their healthcare provider, and also just AI being used, let's say just to help in healthcare allocation or just directing people to relevant healthcare services. In the real world, or the "developing world", I see AI being used to supplement community healthcare workers, and also just people having access to diagnostic devices on their mobile devices or AI powered diagnostics on their mobile devices, just really to help them be able to, not really self-diagnose, but be able to take an image, get a diagnosis from that app, and then also being able to connect to a doctor or a medical professional, and just helping them kind of reduce the time needed to take off of work, travel far away to a better hospital, et cetera. And so there's many opportunities there. Travis Khachato...: I remember 2015, 2016, back when you were studying AI as a computer science major, just AI was not there yet. And it was kind of discussed as, in terms of my perspective, something that wasn't necessarily going to overtake society for a number of years. All of a sudden there's been exponential gains in AI capabilities. Has that surprised you as to how fast it's grown, or have you always anticipated this? Chinasa Okolo: Yeah, so I can't really say I anticipated it, but it doesn't really surprise me. I think just technology in general has seen this development over the decades. Well, I'm only a couple decades old, but just understanding how it's grown, especially when it comes to microchips or in GPUs and also just computers in general, just having access to lots of different functions just from one smartphone. I think that's just really impressive within itself. But AI overall, I think one thing that's really surprised me is how knowledgeable or just informed the public has become of its presence. And so I think Chat GPT is definitely probably the tool that probably has the most responsibility to play in terms of that public awareness of AI. And so I think it'll just be interesting to see how it continues to evolve over the next couple of years and decades. Marilyn Thompso...: So you're also working in helping to formulate AI governance. Tell us about your new job? Chinasa Okolo: Yes. So I recently started as a fellow at the Brookings Institution. So the Brookings Institution really focuses on just public policy in general. Within Brookings, I'm situated within the Center for Technology Innovation. And so we basically cover all aspects of technology, but a strong focus is on AI. And so later on in my piece, she got really interested in questions about AI governance, especially how it pertains to emerging markets, like in India or just Africa in general. And I wanted to really, my research showed me a lot of connections, and I also started some work with the African Union in terms of helping to develop a continental strategy for AI. And so I figured that post PhD, it would be really interesting to figure out how I can center my career on some of these questions and also address some of the questions domestically within the US in terms of how can we help, I would say, address some of the concerns about AI, especially when it comes to more so legislative efforts and governing and regulating these tools. Marilyn Thompso...: What do you see as some of the ethical issues that need to be addressed? Chinasa Okolo: Oh, yeah. There's so many. Within healthcare, we see in a lot of use cases in terms of biased healthcare algorithms that, let's see, are likely to recommend black patients for lower levels of care, versus patients of other races. Within the United States, the TSA has also been very active in adopting these facial recognition technologies, but again, work from Joy Buolamwini a researcher I mentioned earlier, and also like Timnit Gebru, have really shown that these technologies are more likely to be biased against women, and then they perform worse on women of darker skin tones. And so just on my way coming to Pomona, I saw that TSA is using facial recognition in the checkpoints. I use a service called Clear. They're only scanning my eyes, but I mean, there's also concerns with that as well. And so I think there has to be more public conversation about these technologies. And then also just an education. We've seen that Chat GPT has become very popular for students to help supplement their writing, and also just concerns about people being falsely accused of using these technologies. And so I think there needs to be more public conversation, and just general participation and helping to understand how we can advance these conversations. Travis Khachato...: How do you work to lessen the biases in these AI technologies as they emerge and become part of our daily lives? Chinasa Okolo: I guess in terms of bias, honestly, there will always be some kind of bias. I think there's just better ways in terms to alleviate how these AI tools are actually being deployed, especially when it comes to things like I mentioned, like TSA and other use cases as well. And data is also a very big part of this. I think there has to be more meaningful participation from people that have different kinds of expertise, like computer scientists or software engineers are not the only ones that have the expertise needed to contribute to data collection or just AI development in general. There needs to be people with legal backgrounds, from psychology, anthropology, sociology, et cetera. And so I think these voices are really needed in the development of AI. Marilyn Thompso...: What worries you about AI? Chinasa Okolo: Honestly, not many things. I think the biggest thing for me is that it'll serve the interests of large corporations and not really serve the needs of people who are marginalized. I think my main concern is AI actually exacerbating existing disparities, especially when it comes to things like policing, healthcare, education, economic opportunity. And so I think it's really important that, again, people from marginalized backgrounds be included in conversations in terms of how AI's developed and governed as well. Travis Khachato...: I guess on the flip side of that, what most excites you about AI? Chinasa Okolo: Yeah, so many different things. I would say some fields that really intrigue me, especially pertains to interpretable MO. And so I mentioned earlier, my research really focuses on AI expandability, but interpretability is really focused on just making sure that the models themselves are understandable, and just I guess interpretable to humans alike. And so I think it could actually solve some of the issues that we see with explainability, because honestly, explainability can be kind of seen as a band-aid solution for the lack of interpretability or the black box that we're seeing with these kind of models. And then other fields, like tiny MO really focused on expanding computing access of machine learning models to smaller devices like smartphones or smart watches. And I think this really would be helpful, especially when it comes to healthcare and improving how we can run advanced machine learning models on device so people can have better access to healthcare. Marilyn Thompso...: So your watch telling you that your heart is acting up, is that what you're talking about? Chinasa Okolo: Oh yeah, most definitely. But also just for community healthcare workers, the Cloud is there, but some places that don't have good connectivity to data, you can't really rely on a Cloud. And so if you can run these machine learning models on your device and it can save you a lot of time and also provide you with more accurate results as well. Marilyn Thompso...: So what contribution do you want to make in the near future and longer term? What do you envision your future in this field to be? Chinasa Okolo: Yeah, so many different things, but I would say just, one, opening up the conversation on AI governance to regions that have been underrepresented, especially within the African context. I've been fortunate to be able to work with the African Union, but honestly, the conversations that we've seen around AI regulation have really been focused on the United States, other places like Canada, the EU has definitely been a leader, but there are these countries that have economies that will make impactful contributions to the world, and so also they should be included in these conversations. Travis Khachato...: How impactful can these technologies be on developing countries? Chinasa Okolo: Yeah. Again, so many different ways. But I think one, in terms of helping to advance traditional industries like manufacturing, agriculture, especially agriculture, just because a lot of African countries or countries within a global south rely on this domain. And so there's lots of needs to help predict what crops will do best in a certain season, or for example, to quickly identify diseases that are affecting crops, just to ensure that they don't, I would say, kill an entire yield. And so I think that's really important for AI. AI could be really important in agriculture. Travis Khachato...: And I know this whole conversation has kind of had this overarching theme, but why is it important that your research focus on the global south to you personally? Chinasa Okolo: Yeah, so again, just having that personal connection to Nigeria and also just seeing very similar challenges throughout other countries within Africa and also other regions like India and in Latin America. And not to say no one is really addressing these concerns, because there are thousands of researchers and initiatives that are focused on AI in the global south, but I happen to come from a privileged background in terms of being educated at Cornell and Pomona, has provided me with a platform to kind of address or highlight these issues more prominently. And so also now being at Brookings as well. And so I think I really want to get these conversations started and hopefully ensure that other people that actually have these real life experiences of growing up in the global south and developing countries can meaningfully contribute and advance AI development. Marilyn Thompso...: Is the field of computer science becoming more diverse? And how will that help the field to represent more people? Chinasa Okolo: Yeah. So I haven't looked at the stats recently, but I believe so. Just understanding engineering departments at US universities are becoming more gender balanced, and when it comes to underrepresented minorities, the degrees are increasing. And so I would say yes to a certain extent. I think for me, when it comes to the actual AI development or just computer science in general, that you can have more minorities or just more people with marginalized backgrounds in these fields, but if their voices aren't being elevated and they're not given the opportunity to take leadership positions and actually meaningfully contribute to technical development, there's not going to be any change. And so you have to be really intentional about how you include these people. Travis Khachato...: And I would imagine a lot of your job or when you're out there is educating people on what AI is, and I'm sure there's a lot of misinformation or fears about AI. Is that what you're experiencing a lot when you're talking with people? Chinasa Okolo: Yeah, most definitely. And I would say it's really sad to see a lot of this misinformation or these just very wild conspiracies are being spouted by people who are actually very involved with technical development of these systems, or are the leaders or have done foundational work in AI and machine learning. And so I think one misconception is that AI can kind of take over the world. It's going to cause mass destruction or lead to the World War III. But the thing is, as I mentioned previously, humans are still very involved in AI development, whether it starts from collecting data, to training the models and actually choosing where to deploy them. And so humans are involved in every stage of the AI lifecycle. And so it's really important to understand that humans have control over AI. AI will never take control on its own. Marilyn Thompso...: How do we keep from having blind spots in AI because we don't have enough data for them to learn, to avoid things like not having enough different types of people and nationalities and diversity represented? Chinasa Okolo: Yeah. So again, as I've mentioned that it's really important that to include people from those backgrounds. And it may not necessarily seem feasible to do so, but again, large companies like Google and Microsoft have resources, and fortunately they have expanded their footprints into underrepresented regions throughout the world. So Microsoft and Google both have research labs within India. They also have research labs within South Africa, Kenya, Nigeria, and Ghana as well. And so there's been lots of efforts from their end, but I also want to definitely recognize the efforts of local grassroots initiatives. There've been a lot within the African continent, especially Deep Learning Indaba, Ghana NLP and [inaudible 00:28:37], they've been really focused on training African researchers, one, to collect their own data and also be intentional in how they train AI systems that cater to local languages. And so they've actually done a lot of work in filling this gap from systems, from companies like Microsoft and Google. Travis Khachato...: So when you think about your future in AI, do you anticipate it being more about public policy or boots on the ground grassroots movement? Chinasa Okolo: Yeah, I would say definitely a mix of that. For right now, I'm really focused on public policy, just because there's just so many conversations happening right now and we've seen an urgent need to regulate how AI is being developed, deployed, and just used in daily life. And so I think right now it seems like a good point for me to focus my career on that, but I'm really interested in the future and just founding my own thing, especially when it comes to doing more so grassroots AI initiatives. I'm definitely interested in being based somewhere on the African continent, still trying to figure out where. Nigeria is an obvious choice, but I really like places like South Africa, and I had the chance to visit Rwanda earlier this year with the African Union and had a really great time there. And they also have a thriving tech scene. And so yeah, I just definitely want to contribute any way I can. Travis Khachato...: Chinasa, it's very comforting to me personally that somebody with your kind of information is not pessimistic about the future of AI. I want to thank you for sitting down and hanging out with us at Sagecast. Chinasa Okolo: Yeah, thank you. Marilyn Thompso...: Thanks so much. It's been great. Chinasa Okolo: Thank you for the invite, I appreciate it. Marilyn Thompso...: Wow. AI is a topic we'll no doubt be discussing and trying to wrap our heads around for years to come. It certainly fits our theme of tackling the difficult. Travis Khachato...: Thanks for being with us for this episode of Sagecast, and thanks to our audio engineer Eric Tyron, and KSBC for hosting our conversations. Until next time.