Validation and Evaluation as Essentials to Ensuring Safe AI Health Applications

Area of Science:

• Health Informatics
• Artificial Intelligence
• Medical Technology Evaluation

Background:

• Artificial Intelligence (AI) is rapidly advancing in health informatics.
• Current AI validation lacks the rigor of other health technologies.
• Health AI can introduce and perpetuate biases, posing safety and efficiency risks.

Purpose of the Study:

• To revisit health informatics principles for AI evaluation.
• To establish scientific principles for AI evidence production.
• To address the need for rigorous validation of health AI.

Main Methods:

• Review of existing health informatics principles and techniques.
• Application of the Precautionary Principle for AI evaluation.
• Integration of continuous quality improvement methods.

Main Results:

• Need for robust evidence and evaluation trails from AI developers.
• Call for policy makers to mandate rigorous AI evaluation.
• Emphasis on balancing agile evaluation with thoroughness.

Conclusions:

• Scientific principles are needed for health AI evaluation.
• Agile and pragmatic methods are essential for evolving AI.
• Continuous evaluation is key to prevent perpetuation of errors and bias.