Considerations When Implementing AI

The benefits that AI will bring -- and has already brought -- to healthcare are endless! However, as with any technology, there are elements of AI that serve as precautions, like:

• In the security and handling of patient information that is both sensitive and confidential; healthcare providers will need to make sure that the AI vendors they work with are committed to transparency and have an understanding of how to protect data from breaches.
• The information these AI tools use in their knowledge bases should be continually updated in order to ensure the timeliness of information. Medical research is highly dependent on the newest discoveries, and AI tools should give responses that reflect these developments.
• AI models don't offer much interpretability and are a black box of sorts -- they don't often give an explanation for their decision making, and as AI models are able to give more complex and dynamic answers, it becomes more and more important for users to understand and correct any errors these technologies make
  • If "the training process for these models is not transparent, it is impossible to know why the models succeed on some questions while failing on others" (Omiye et al., 2023). 
• The information these models are trained on have to be as objective and fair as possible-- medical research as it exists today reflects bias and adversely affects underrepresented populations; if AI models are trained on research that already perpetuates disparities in healthcare, then the answers it produces will also exhibit these biases.
  • Based on research that already exists, "models also perpetuate false conclutions about racial differences on such topics such as skin thickness and pain threshold" (Omiye et al., 2023).
• Along with ensuring the timeliness of information, healthcare providers and users should make sure that the information AI models are trained on are reliable and as unbiased as possible. 
• Generative AI is unique in its ability to build strings of text based on previous strings of text, but this poses an equally unique possibility of "hallucinations"-- which refers to responses to prompts that may not be accurate. 

(Arshad et al., 2023)

At the end of the day, it's important to recognize that "algorithms make crucial decisions affecting human lives. AI ethics advocates ensuring algorithms are fair, unbiased, and transparent. There should be a mechanism to understand the algorithm's underlying decision-making logic" (Khalid et al., 2023).

For this reason, it would serve us well to understand AI in the context of clinical practice. The American Medical Association (2024) emphasizes the term "augmented intelligence," highlighting the human element in medicine that remains integral to clinical encounters that prioritizes patient health and safety and the achievement of the Quintuple Aim.