Artificial Intelligence Bias in Health Care: Web-Based Survey

Area of Science:

• Medical Artificial Intelligence
• Healthcare Technology
• AI Ethics

Background:

• Growing investment in artificial intelligence (AI) for medical applications like diagnosis and treatment.
• Previous research addressed AI transparency and bias reduction, but developer perceptions remain understudied.
• Limited understanding of AI developers' knowledge and views on bias in healthcare AI.

Purpose of the Study:

• To survey healthcare AI specialists regarding their perceptions of bias in AI algorithms.
• To investigate developers' awareness and utilization of bias prevention measures.
• To assess variations in perceptions based on demographics and work environment.

Main Methods:

• A web-based survey with up to 41 questions was administered in German and English.
• Analysis included demographic data, technical expertise, and perceptions of fairness and bias.
• Participant data from those with medical AI experience and complete questionnaires were analyzed.

Main Results:

• 151 AI specialists completed the survey; 67% were male, with a median age of 30.
• 31% rated their AI projects as fair, 34% as moderately fair, and 12% as barely fair.
• Key bias contributors identified: lack of fair data (68%), insufficient guidelines (49%), and knowledge gaps (45%).

Conclusions:

• AI bias perception among developers is moderately fair, but gender minorities reported low fairness ratings.
• There's a need for increased focus on minority and women's perspectives in AI development.
• Strengthening knowledge and guidelines for preventing bias in healthcare AI is crucial.