What are the three main ethical risk areas when AI is used in medicine?

Transparency and comprehensibility (the black box problem), humanity and empathy (bias and discrimination in the algorithms), and access to the technology (avoiding two-class AI medicine).

1. Transparency and comprehensibility. The black box problem: you know what data went in and what came out, but not how the system got from one to the other or why it reached its result. Add the potential for AI errors. The proposed remedy is to keep a person involved: human in the loop or human on the loop (see the next card).
2. Humanity and empathy. Algorithms can carry biases (systematic distortions) that discriminate against groups. This demands strict attention to the quality of training data and transparency from AI vendors about how models are built.
3. Access to the technology. Avoid a two-class AI medicine in which some people can see a doctor while others are automatically diagnosed and treated by a computer.

Each risk maps onto a principle: the black box threatens informed decisions (autonomy), bias threatens equal treatment (justice), and two-class medicine threatens fair access (justice again).

Go deeper:

• Wikipedia: Explainable artificial intelligence — the research field that tries to open the black box.
• Wikipedia: Algorithmic bias — how training data turns into discrimination, with medical examples.
• Wikipedia: Two-tier healthcare — the general concept behind "two-class medicine".