An artist’s illustration of artificial intelligence (AI). This illustration depicts language models which generate text. It was created by Wes Cockx as part of the Visualising AI project l...

Why AI’s Biased Algorithms Deserve A Closer Look

Artificial intelligence has the potential to transform healthcare by delivering applications that improve patient care and boost operational efficiency. As with any new technology, however, it must be deployed responsibly. Many AI users are familiar with its propensity to hallucinate or generate false information. While the damage from that can often be mitigated by refining the prompt and careful fact-checking, a larger issue lies in algorithmic biases– a problem that remains largely underestimated and unchecked by the medical community.

Why? Scale. With each passing year, more algorithms are being applied in the healthcare industry via artificial intelligence technologies. The global AI healthcare market was valued at $19.54 billion in 2023 and is projected to grow from $27.69 billion in 2024 to $490.96 billion by 2032

Biased Algorithms Perpetuate Racial Disparities

For all their utility, AI tools are drawn from our existing body of knowledge. Their outputs will only be as strong as their inputs allow. If the algorithms used by healthcare decision-makers are exacerbating disparities in care between White and non-White patients, now is the time to take a closer look at our existing body of knowledge and how it feeds into those algorithms. A key question, then, is whether we can trace biased outcomes directly to the use of algorithms in healthcare.

According to one systemic review conducted in December 2023, the answer is “sometimes.” The authors identified 17 studies that examined the effect of 18 algorithms on racial and ethnic disparities in health and healthcare. More than half of them found that the algorithms they reviewed either perpetuated or exacerbated existing disparities.

The study went a step further by identifying strategies to mitigate racial and ethnic biases associated with algorithms. These solutions revealed why a more than one-size-fits-all approach will be needed to solve the problem. Types of mitigation strategies included removing a race or ethnicity input variable from the algorithm, replacing race or another input variable with a different measure, adding an input variable, recalibrating the algorithm with a more representative patient population, stratifying algorithms to assess Black and White patients separately, and using different statistical techniques within algorithms.

The American Medical Association (AMA) has called for eliminating the misuse of race in clinical algorithms and implementing strategies to remedy any associated harms. However, a survey conducted in early 2023 found that 12 out of 68 regional or specialty-based medical societies had yet to consider taking action on this issue.

Next Steps for Healthcare Leaders

Progress in this area has been disappointingly slow. Healthcare leaders must scrutinize the algorithms feeding into the processes at all levels of their organization to reduce and eliminate bias. Offering the highest quality care to patients regardless of race should be a high priority for the industry, and AI solutions can — and should — help us overcome our natural biases with proper human oversight.