Pending funding, the National Institute of Health (NIH) plans to launch the Bridge to Artificial Intelligence (Bridge2AI) program. Collaboratively managed by the NIH Common Fund, the National Center for Complementary and Integrative Health, the National Eye Institute, the National Human Genome Research Institute, the National Institute of Biomedical Imaging and Bioengineering, and the National Library of Medicine, the program seeks to provide comprehensive, high-quality and ethically sourced datasets to catalyze the widespread use of AI in the biomedical and behavioral research communities.
AI has the ability to transform the biomedical and behavioral science fields. Possible applications include informing clinical decision making, monitoring and predicting health needs in real time and analyzing how genetic information, cell characteristics, and social and environmental factors affect one’s health. Beginning to incorporate AI applications into the medical field has uncovered a number of issues with the current datasets. Most available datasets at the moment are incomplete, lacking context, diversity, and standardized collection conditions. As a result, the use of these datasets is leading to biased, unethical outcomes.
Over the last year the Computing Community Consortium (CCC) had a task force, the Computing Challenges to Humanity Ethics team, that discussed a number of the negative effects of using biased datasets in predictive medical algorithms. The ways in which the data is being collected proposes a number of issues. Specific instances include:
- A disproportionate amount of data collected between races causing broad, unfounded claims, such as all African Americans having a higher creatinine levels. This unfounded claim has led to a higher rate of undiagnosed kidney disease among African Americans and serious health complications. You can read more here.
- Minority women have a greater risk of complications during child birth due to inequitable care in hospitals. Medical algorithms determining the success and risk factors of a C-section wrongfully interpret the data to mean that minority woman are generally at a higher risk and are therefore barred from receiving one. You can read more here.
Bridge2AI seeks to overcome these complications by developing guidance and standards that ensure the eradication of inequities and bias to create ready-to-use AI datasets. You can read NIH’s announcement here and watch a video about the program on their YouTube page.
