Inequalities In Healthcare AI: Old Wine, New Skins?
Let’s face it.
Access to healthcare has been complicated by deep-rooted inequalities for ages. How easy, affordable, appropriate and effective healthcare has been determined by race, social-economic status, religion, gender and other societal factors. And it has made it harder for anyone not mainstream to access the health services they need and deserve.
AI has entered the Chat!
We are well and truly in the era of AI. going by predictions going round, AI will have a profound effect on healthcare. Making diagnosis easier, cheaper and faster, selecting optimal care plans, writing prescription without the pesky interactions we encounter every so often. The benefits are immense.
That is if you are part of the demographic whose data is being used to train the models. Unfortunately, the available datasets are very thin on marginalised groups. Maybe because they don’t interact much with healthcare. There are many known reasons.
Training AI models are becoming cheaper by the day. In fact as of today, they many off-the-shelf models you could use. The problem is the data you feed the model. Is it representative of all people in the population? More often than not, it’s not. But the model will still be generalised to everyone.
Determinants of health are intertwined with one’s genetic makeup and environment. If the data used on AI doesn’t take in the ethnic makeup, the environment they reside in, their diets, habits etc, then it’s fair to say the recommendations cannot apply.
What are the drivers of Inequalities In Healthcare AI
- Biased algorithms:
AI systems are trained on large datasets, which may contain biased or incomplete information. If these datasets are not representative of diverse populations, the algorithms may produce biased results (Obermeyer et al., 2019). For example, if a facial recognition system is predominantly trained on lighter-skinned individuals, it may have difficulty accurately identifying individuals with darker skin tones, leading to misdiagnosis or inadequate treatment recommendations for underrepresented minorities.
- Limited access to technology:
Underrepresented minorities often face barriers to accessing healthcare services and technology. AI-based healthcare solutions, such as telemedicine or health monitoring apps, require access to smartphones, stable internet connections, and digital literacy. At the very least, working healthcare systems collect adequate data. Limited access to these resources can prevent minorities from benefiting from AI-driven healthcare innovations, further widening the healthcare gap.
- Lack of diversity in AI development:
The lack of diversity among AI developers can contribute to biased algorithms. When the development teams lack representation from underrepresented minorities, they may overlook or fail to address the unique healthcare needs and concerns of these populations (Benjamin, 2019). This can result in AI systems that do not adequately cater to their requirements, leading to unequal healthcare outcomes.
- Ethical concerns:
AI in healthcare raises ethical concerns regarding privacy, consent, and data security. Underrepresented minorities may be more hesitant to share their personal health information due to historical mistrust or fear of discrimination. If AI systems rely on extensive data collection, the reluctance of these populations to participate can result in biased or incomplete datasets, leading to inadequate representation and potentially exacerbating existing health disparities (Mittelstadt et al., 2016).
- Lack of transparency and accountability:
AI algorithms can be complex and opaque, making it difficult to understand how they arrive at their decisions. This lack of transparency can make it challenging to identify and address biases within the algorithms (Char & Shah, 2016). Without proper accountability and transparency, underrepresented minorities may be disproportionately affected by biased decisions made by AI systems.
To mitigate these issues, it is crucial to prioritize diversity in AI development teams, ensure representative datasets, and promote transparency and accountability in AI algorithms. The battle for more inclusive healthcare is still on to address the barriers to access faced by underrepresented minorities. Additionally, involving diverse communities in the design and implementation of AI-driven healthcare solutions can help ensure that these technologies address their specific needs and concerns (Nundy & Montgomery, 2020).
References:
– Benjamin, R. (2019). Race after technology: Abolitionist tools for the new Jim Code. Polity Press.
– Char, D. S., & Shah, N. H. (2016). Bias in artificial intelligence algorithms. Journal of the American Medical Association, 316(24), 2673–2674.
– Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of algorithms: Mapping the debate. Big Data & Society, 3(2), 2053951716679679.
– Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447–453.
– Nundy, S., & Montgomery, T. (2020). Waking up from the AI dream: AI, health inequality, and the need for equity-centred innovation. Health Affairs Blog. Retrieved from: https://www.healthaffairs.org/do/10.1377/hblog20200416.577800/full/
I’m fascinated by the possibilities of AI in healthcare in resource-poor settings. Follow my twitter https://twitter.com/joemdoc
Join our newsletter mailing list to get fresh news on laboratory medicine practice and policy!