Artificial Intelligence (AI) in Health and Trust: the Lack of Explainability

The key to a great story is not who, or what, or when, but why. (Elliott Carver, Tomorrow Never Dies)

  • With respect to the deployment of AI models, perhaps the most pressing ethical issue pertains to their non interpretable or so-called black box algorithms, where the inner logic of these algorithms remains hidden to their developers. (Hidden layers of artificial neural network in a machine learning algorithm)
  • This lack of transparency can reduce the trustworthiness. And from an ethical principle perspective, the disclosure of details about medical treatments to patients is a fundamental tenet of medical ethics.
  • It thus requires that physicians themselves be able to grasp at least the private mental inner workings of the devices they use. So for AI models to be ethical, they must be explained, and developers must be in capacity to communicate to their end users, whether they be permissioned or patients, the general logic behind the model’s output decisions.

AI solutions: unforeseen consequences

  • Communicating with patients about the use of these technologies, also will have the benefit of increasing their trust and acceptance. What is the use of collected data discussed is an ongoing concern.
  • As more diagnostic and therapeutic interventions become based on machine learning, the autonomy of patients in the decisional process about their health and the possibility of shared decision making can be undermined.
  • Consider for example, an automated triage tool, where payers consider recommendations of this triage tool are preconditioned for reimbursement and would refuse to cover treatments that the model recommends against.
  • Given the importance of personal contact and interaction between patients, between physicians and caregivers, we need to be able to guard against that. And it’s paramount for us to integrate AI models into patient-centered, supportive, and empowering delivery systems. While it’s clear that the clinical use of AI models will inevitably transform existing models of healthcare delivery of dermatology practice, healthcare providers must promote a North Star game of human-centered use of AI; one that’s fully aligned with data protection requirements, that minimizes the effect of bias and that achieves transparency.

Justin Ko, MD. Ethical Considerations in AI. 8th World Congress of Teledermatology, Skin Imaging and AI in Skin diseases – November 2020


Scroll to Top