A couple of weeks ago I saw a post on LI saying something about the virtues of machine learning or AI applications. One statement that stood out to me was how AI avoids human biases, or something to that effect (* as if that was an inherently bad thing, anyway).
Interested in the factual basis of this statement, and prompted by my wise friend, Philosoraptor, I went to the literature.
I quickly ran into dozens of papers covering the multiple biases of machine learning and AI. Lots in healthcare and medicine, which was cool.
A couple of examples are attached in images 1 and 2 for those interested – which highlight a number of factors which should be considered in the application of AI in healthcare.
One I found particularly interesting in the land of AI was ‘reward hacking’.
This describes a situation where an industrious program finds a way to achieve a reinforcing reward (like a proxy of the intended goal), and optimises this misaligned goal performance, without actually achieving the intended goal.
Source: Challen, R., Denny, J., Pitt, M., Gompels, L., Edwards, T., & Tsaneva-Atanasova, K. (2019). Artificial intelligence, bias and clinical safety. BMJ Quality & Safety, 28(3), 231-237.
Link to the LinkedIn post: https://www.linkedin.com/posts/benhutchinson2_a-couple-of-weeks-ago-i-saw-a-post-on-li-activity-7074140714216906752-xqkY?utm_source=share&utm_medium=member_desktop
SafetyInsights.org 
One thought on “AI and biases in healthcare / clinical medicine”