AI and the need for justification (to the patient).

Area of Science:

• Medical Ethics
• Artificial Intelligence
• Health Informatics

Background:

• Shared decision-making in healthcare requires treatment choices to be justifiable to patients.
• Black-box AI, common in healthcare, often lacks transparency, making decisions difficult for patients to understand.
• Current AI models may not adequately align with individual patient values and preferences.

Purpose of the Study:

• To argue that black-box AI's lack of algorithmic justifiability undermines patient-centered medical care.
• To demonstrate how algorithmic transparency alone fails to ensure patient-directed justifiability.
• To introduce and advocate for a hypothetical 'Justifiable AI' model for improved shared decision-making.

Main Methods:

• Conceptual analysis of algorithmic justifiability in medical AI.
• Critique of black-box AI's limitations in aligning with patient values.
• Introduction of a hypothetical Justifiable AI model framework.

Main Results:

• Black-box AI decisions are difficult for patients to reconcile with their personal values due to a lack of explicit rationale.
• Algorithmic transparency does not sufficiently address the challenge of patient-directed justifiability.
• The proposed Justifiable AI model explicitly incorporates normative and evaluative considerations.

Conclusions:

• Justifiable AI models are preferable to black-box alternatives for fostering shared medical decision-making.
• Development of Justifiable AI is crucial for enhancing patient trust and adherence to treatment plans.
• Explicitly modelling patient values in AI is essential for ethical and effective healthcare applications.