The married team of Gari Clifford and Rachel Hall-Clifford broke new ground when they began to create advanced healthcare technologies in partnership with the people using them.
For many years, Clifford, a biomedical engineer and professor of biomedical informatics, and Hall-Clifford, an anthropologist and associate professor of sociology and global health, have been working together with Indigenous Mayan midwives in Guatemala to pioneer low-cost, AI-based devices that can be used to address a range of problems for pregnant women and babies.
About half of Indigenous Mayan women deliver at home with the help of midwives who have few resources to screen for complications. Clifford, who trained in physics and engineering, spent more than a decade working with Hall-Clifford and Indigenous populations — listening, gathering data and searching for technological improvements that would make a meaningful difference in the lives of patients.
Using this co-design approach, the team developed an AI algorithm accurate enough to identify fetal growth restrictions, using an inexpensive device plugged into a low-cost smartphone. The tool allowed the midwives to spot and record abnormalities in pregnant women.
But as much as these technologies can help, Clifford and Hall-Clifford warn of the broader issues of exporting advanced technologies like AI from the high-income countries that created them to resource-poor countries, without working to adapt them to local cultures. They recently elaborated on their concerns in this Q&A.
Q: You’ve said we may be at an inflection point for AI in health care. Can you explain?
Prof. Gari Clifford
Rachel Hall-Clifford: In the Emory Co-design Lab for Health Equity, we recognize that tools themselves are agnostic in terms of outcome. Pen and paper are technologies too. AI in and of itself is neither a savior nor a villain, just a tool that we have to very intentionally apply equitably. That's our goal.
Q: Why does it matter that the data used to train AI health models mostly comes from high-income countries?
Gari Clifford: Because the world’s population is very diverse. There are so many different things that influence your likelihood of developing a particular disease. It can be diets, genetics or socioeconomic environment, for example. It can be behaviors, belief systems, height or weight. All of these vary enormously across the planet. When you collect data only in one place, the likelihood of that algorithm working in a completely different place is significantly lower. We've seen this time and time again.
Q: You’ve also warned about the dangers of what you call data colonialism. What is that?
Gari Clifford: This is really part of the larger issue of the neocolonialism of the new technologies. Data is sometimes sourced from the Global South, but usually in an exploitative manner. AI often requires expensive high-quality labels, and human labor can be much cheaper in the Global South. So we end up with the data being generated in the Global North; the Global South is used to label the data, but the product is designed for, and benefits the Global North. To add injury to insult, the Global South not only misses out on the vast majority of the benefits of the technology, but also bears a disproportionate burden of the negative effects, through climate change and e-waste, among other issues.
Prof. Rachel Hall-Clifford
Rachel Hall-Clifford: It's the water we swim in, in terms of the way the world works, that we anticipate innovations coming from high-income countries. We see this same pattern of inequality in health, especially when we're looking at the Global South or the global majority. We want to push back on who gets to reap the benefits of innovation. How can we create opportunities for everyone to be receiving care and pushing it out from traditional health facilities, but also who gets to be involved in the actual innovation? That's really where co-design comes in.
Q: AI requires a huge data infrastructure that’s beyond the means of many low-income countries. One solution you’ve proposed is called edge computing. How does that work?
Gari Clifford: In edge computing, all the intensive computing is going on at the edge of the healthcare system. It's in your hand at the moment you're meeting the patient. All the AI that's going on in your phone is edge computing. Traditionally, the tech companies have tried to push data up to the cloud and then do the compute there, where they get to control the data and keep it to themselves. But the amount of resources that are being consumed to process this data is becoming prohibitive. Training ChatGPT4 had a carbon footprint of about 10,000 round trips between Paris and New York and around 22 million liters of water (to cool the data center, for example). And this isn’t a one-off. But we don't need to push the data to the cloud if we leverage edge computing — your phone's already paid for, so the computing can happen in your pocket. You don't have to spend energy pushing large amounts of data up to the cloud, moving it around and storing it.
Q: You’ve both spent many years in Guatemala partnering with Indigenous women to develop a smartphone app for individual health. What’s the next stage?
Gari Clifford: In February, we should be launching a new phase of our co-design process of the safe+natal system, where we're going to look at how midwives respond to all this AI on the phone. It'll tell the user how to use the device, such as, “The volume's too loud; the volume's too quiet; please move the probe around.” The app then tells you how much good data you've collected and if it’s enough. We are also looking ahead to implementation of safe+natal in pilot sites around the world.
Rachel Hall-Clifford: I think we've been in this period of excitement and thinking of AI as a gold rush. Now I start to see a lot of thinking that it's sort of the bogeyman of our times. The reality is it's a neutral tool that we need to apply thoughtfully and ethically, just like any tool. I hope the Co-design Lab and our work can be part of shaping how we do that.