4. AI in dentistry: ethical and regulatory considerations

AI can increase dentists’ diagnostic accuracy but may also (increase/decrease) invasive treatment decisions

Mertens et al, 2021

increase

what are some ethical risks of AI in dentistry?

• reading scans, tx planning tools, EHR - pt info

• biases: AI models can learn and amplify biases

• impacts: group representation may see worse outcomes ot degraded accuracy (subgroup disparities)

the ethical risk of impact in terms of group representation is especially important for which models?

ML (still present in transformer-based generative AI models including LLMs)

what are some issues with transparency and over-reliance in AI in dentistry?

• limited explainability: predictions w/o meaningful explanation, no reason

• risk of over-reliance: decision SUPPORT

• automation bias: AI output can influence clinical decisions even when the output is incomplete or incorrect

• suggestion: treat it as a piece of evidence requiring professional review, not a true diagnosis

how can AI be used in dental practice?

• diagnostic and clinical workflow - image, pathology detection, tx recs; FDA reg, etc

• documentation workflows - summarize, scheduling, coding, automation

“no name” does not mean de-identified

HIPAA:

• Safe Harbor (removing 18 specific identifiers)

• Expert Determination (person with appropriate knowledge and experience certified re-ID risk is sufficiently small)

how to evaluate AI workfow?

• regulatory and institutional approval

• data classification and privacy

• security and data handling

• ethical review

summarize the key takeaways of this lecture:

• AI can improve care, but it also introduces ethical, privacy, and regulatory risks

• "No name" does not mean de-identified. FDA, HIPAA, institutional policy, and DUA rules may overlap

• Evaluate risk across an entire process, not just a software or tool

• AI supports clinical judgment; it does not replace it. We remain responsible for patient care decisions and patient data.