In health care, many AI developers are white men. HSC is trying to diversify the field
READ MORE
Life and death in Fort Worth’s 76104
People in Fort Worth’s 76104 ZIP code on average won’t see their 67th birthday. What is causing the lowest life expectancy rate in Texas? What can be done to help? Read the Star-Telegram’s investigation:
Expand All
No doctors, no grocers, no help. Death comes early in these Fort Worth neighborhoods.
Life in a Fort Worth food desert, where fresh produce is 2 bus rides from home
The birth of hope: How 2 Fort Worth officers changed the lives of fatherless boys
They saw poverty and sickness, so these Dallas doctors moved into the neighborhood
The University of North Texas Health Science Center at Fort Worth will lead a massive new federal project to train a more diverse workforce in the world of artificial intelligence and health care, the National Institutes of Health announced.
The Fort Worth campus will serve as ground zero for the nationwide project to tackle an urgent question facing the country’s health care system. Congress has allocated $100 million over the next two years to fund the program, for which HSC has been named the “coordinating center,” said Dr. Jamboor Vishwanatha, who will lead the center.
Vishwanatha, the director of the Texas Center for Health Disparities, and his peers are tasked with using computers to make health care more equitable, and to prevent bad data or biased algorithms from making health disparities worse than they already are. That mission becomes increasingly more urgent as machine learning touches more and more aspects of our daily lives, from an Apple Watch or a Fitbit that monitors our movement today to pacemakers of tomorrow that could be powered by algorithms to respond to an individual patient’s needs.
“The biggest concern is: Yes, the technology is moving in that direction, but if the algorithms that are being developed and the people who are developing those algorithms, if they are not diverse, then eventually what will happen is you will have advances in the medicine, but it is not applicable to many of the groups,” Vishwanatha said.
Where do we encounter artificial intelligence in health care?
Artificial intelligence and machine learning algorithms power many of the tools we use everyday, from search engines like Google to the software behind ride-sharing apps like Uber.
The average person will probably only experience machine learning as it affects their health care through some kind of device or medical implant.
“Anywhere that there’s a device that’s making inferences about your body, that is almost certainly going to be AI powered,” said Shiri Dori-Hacohen, assistant professor in the Department of Computer Science & Engineering at the University of Connecticut.
Because artificial intelligence systems can use reams of data to understand and analyze complex problems, they have the potential to tackle some of the thorniest problems in health and health care. But because these systems are designed by humans, they also have the potential to be biased, and to potentially exacerbate the problems they are designed to address.
“Because we live in a biased world, our machine learning approaches are learning from data that we give it. And if the data is biased, because the world is biased, then naturally, our algorithms are also going to learn biased predictions,” said Dori-Hacochen, who is not affiliated with the NIH project. “There’s no magic there, [AI systems] can’t magically be better than the humans that are training them.”
Poorly designed systems in other industries have had disastrous effects: The news organization ProPublica found that an algorithm designed to predict a criminal defendant’s likelihood of committing future crimes was only correct about 20% of the time, and the algorithm was “particularly likely to falsely flag black defendants as future criminals.”
Step one: Recruit a diverse workforce
The coalition HSC will lead is charged with training and recruiting a more representative workforce to study and develop algorithms. Currently, the field is dominated by men, with few women of any race or ethnicity and few Black or Hispanic scientists of any gender entering the field. Of new artificial intelligence PhD graduates in 2019, just 3.2% were Hispanic and 2.4% were Black, according to a survey conducted by the Computing Research Association. That same survey found that 45% of graduates were white and 22.4% were Asian, with almost 25% of graduates of an unknown race or ethnicity. And for the last 10 years, women of any race or ethnicity accounted for 18.3% of new artificial intelligence and computer science PhD graduates.
HSC will be the core of a network of institutions, Vishwanatha said. The school will team up with private companies, community organizations, and other schools to create a consortium. The consortium will involve historically Black colleges and universities and tribal colleges so that even if those institutions don’t have formal machine learning curriculum, they can take courses and access data through the infrastructure created through the project.
The project will initially focus on partnerships with colleges and universities, but ultimately Vishwanatha said they hope to expand the curriculum to the K-12 grade level. All of the data and courses will be available to the public as well, so that even any student can learn about artificial intelligence regardless of the academic resources available at their school.
“It could be a community college, it could be a small tribal college in a rural area, but you will still have the same access as anyone in a large city to be able to use this technology,” Vishwanatha said.
Step two: Use artificial intelligence to address disparities in health care
Once the consortium has developed the educational tools to train the future leaders of artificial intelligence, those students will tackle some of the biggest research questions in health, using electronic health records and other data sources to tackle disparities in health outcomes.
Health disparities are visible across a range of different identities, based on gender, sexual orientation and identity, race and ethnicity, and geography. In Fort Worth, residents of the 76104 ZIP code have the lowest life expectancy in all of Texas, according to an investigation published by the Star-Telegram in September 2020. In 76104, the average life expectancy was 66.7 years in 2019, almost 12 years younger than the national average.
Vishwanatha and other leaders in the AI community are hopeful that carefully designed systems can actually address disparities like the ones experienced by people who live in 76104. A diverse workforce and more comprehensive data could potentially use social determinants of health to prevent people from getting sick in the first place.
One potential Vishwanatha outlined is a partnership between medical clinics and a community to better understand what factors are contributing to a health disparity. For example, at local health clinics, patients’ medical information is usually recorded through their electronic health records, which can be anonymized for researchers to analyze. But if patients agreed to offer additional information—like their access to fresh food and opportunities to exercise, their income level, and stressors in their lives—researchers could potentially pair that information with their health records. Access to such information, however, would require communities to trust that the consortium could handle such information responsibly.
“We need to collect that data and link it to the electronic health record,” Vishwanatha said. “And the only people who can provide that is the community.”
This story was originally published October 1, 2021 at 9:45 AM.
