Health researchers in South Dakota and across the U.S. want to use artificial intelligence (AI) to do things like cure cancer, predict the onset of Alzheimer’s and diabetes earlier, and diagnose and address disparities in the impact of kidney disease.
But first, they need data.
Lots of data.
Data housed by government agencies, hospitals and other health care facilities, drugmakers, research institutions and insurers.
Two leaders with the National Institutes of Health (NIH) told lawmakers this week that connecting researchers with data to train and refine AI medical applications also must be done in a way that safeguards citizen privacy and includes data from underserved and vulnerable populations.
The NIH representatives spoke virtually on Wednesday to members of the Legislature’s Study Committee on Artificial Intelligence and Regulation of Internet Access by Minors.
“Data is what drives artificial intelligence,” said Susan Gregurick of the NIH. “We need data relevant to individuals and to patients in real time and in high quality.”
The NIH has put nearly $1 billion into research and development of machine learning and AI medical research since 2019, Gregurick said, with $296 million in spending in 2023 alone.
A University of South Dakota professor named Bill Harris has been awarded multiple NIH grants over the course of his career. His latest award, for $506,000, supports research into an AI-powered model that looks at patterns of fatty acids in blood samples to predict a patient’s odds of developing Alzheimer’s disease about four years sooner than doctors can now.
Dr. Harris is a professor at the University of South Dakota School of Medicine, but is pursuing the AI research through his company, OmegaQuant.
Another South Dakota project falls under the NIH’s “Artificial Intelligence/Machine Learning Consortium to Advance Health Equity and Researcher Diversity,” program, or AIM-AHEAD. That project has South Dakota State University Professor Semhar Michael looking into the use of machine learning to ferret out disparities in health outcomes for populations affected by end-stage kidney disease.
The SDSU project, which includes partners from Dakota State University, Sanford Health and other researchers in and outside of South Dakota, was awarded a two-year grant worth just over $1 million.
Gregurick of the NIH said that the SDSU project is one of 274 around the country funded through the AIM-AHEAD initiative.
Gregurick said AI research in health care is still in its beginning stages. The NIH hopes to see more “multimodal” AI projects in the future. Such efforts would seek to integrate AI-informed data produced by analyzing things like blood or tissue samples with other sources like voice recognition data. Voice recognition data could help train AI models to detect changes in speech patterns in hopes of triangulating the trajectory of cognitive decline and offer earlier interventions for Alzheimer’s and dementia patients.
Privacy concerns
But none of the research can be done ethically unless there are guardrails to protect patient data, according to Lyric Jorgenson, the NIH’s associate director for science policy.
Anonymizing data to scrub names and other personally identifying information will be important, as will gaining the consent and support of patients at the point of data collection.
“We do want to understand the risks of sharing information, especially with people who shouldn’t have access to it,” Jorgenson said. “Think of it as putting information in a safe, and only your family has access.”
She said communication with patients and communities will be key – the NIH is funding outreach and education efforts on AI and data collection, in addition to research into data use – but basic data security practices also need some adjustment.
Putting data in the cloud, versus putting it on a flash drive that can be passed from person to person, will be part of that.
Data in the cloud can be walled off and provided only to those who’ve been vetted, according to Jorgenson, and a cloud dataset’s manager can see who is accessing the data and when. Data that’s downloaded and saved onto a hard drive or flash drive can’t be tracked and managed with that level of precision, she said.
Sen. Mike Rounds, R-South Dakota, told state lawmakers that the U.S. is in a better position to protect privacy than countries like the United Arab Emirates, Saudi Arabia or China, the last of which collects data through surveillance systems to monitor its citizens.
“Do you want AI to be developed in a place like that?” Rounds said.
Rounds to state lawmakers: Encourage AI development in South Dakota
Rounds has been a leader in the Senate on AI issues, part of a bipartisan group of lawmakers that’s met with tech industry heavyweights in hopes of informing other elected officials about the benefits and potential pitfalls of the technology.
AI research, particularly in the areas of data quality, represents an opportunity for South Dakota’s younger generation, he said, as an AI model is only as good as the data it’s trained on.
Schools like Dakota State University, which has invested heavily in cybersecurity and other data science programs, are well-positioned to do that work.
“These huge databases have got to be accurate,” Rounds said.
Pumping funds into medical research might seem expensive, he said, but so is managing a host of diseases that AI research could prevent or cure with the right investment.
“It’s not inexpensive, but compared to what we pay to try and limit and prevent the illnesses to be cured, the investment is miniscule,” Rounds said.
Lawmakers on the study group wanted to know what, if anything, they could do to help move the research along.
“You talked about not slowing down the progress of AI, but at the same time, there’s got to be policies and oversight managing the implementation of this, and probably training as well,” said Rep. Chris Karr, R-Sioux Falls.
The main goal, Rounds said, should be to “stay ahead” of AI development. Supporting education programs that focus on AI systems and maintaining a business-friendly atmosphere could help position the state to take advantage of the new technology, Rounds said.
“I would do everything I could to incentivize the development of AI databases here, in all different areas,” Rounds said. “It becomes a nexus for other things to happen.”
Companies looking to develop data centers or AI hubs, Rounds said, are “going to go where doing business is as simple as possible.”