Artificial intelligence (AI) is widely defined as the capability of computer systems to perform tasks that would normally require human intelligence, such as learning, reasoning, and making decisions. Regulatory frameworks around the world similarly characterise AI as machine-based systems that can operate autonomously and produce outputs, such as predictions or recommendations. These definitions highlight that AI is more than just software automation – it is a type of technology capable of inference and complex decision-making. 

Past and present

In clinical dentistry, AI software analyses patient data, detects patterns in images, estimates risks, and assists in treatment planning. Where early systems like MYCIN relied on simple rules, today’s AI uses deep learning trained on large datasets and is now integrated into practice management platforms.

A major challenge is the "black box” problem: AI systems can produce results without revealing their reasoning. For example, a radiographic algorithm might identify a lesion but not explain how it did so, making error detection difficult. 

The response is the “glass box” model – AI systems designed for transparency and explainability that allow clinicians to understand and communicate the AI's reasoning.

Choosing a system

Responsible AI use begins with choosing systems that adhere to regulatory and data protection standards. In dentistry, AI provides various outputs such as diagnostic (disease detection), predictive (risk assessment), simulative (treatment visualisation), and assistive (highlighting areas of interest). 

AI-powered imaging tools can now detect caries, periodontal bone loss, and fractures, with some studies reporting up to 90% accuracy. In certain cases, AI sensitivity for identifying caries even exceeds that of experienced clinicians. However, system performance varies and heavily depends on the quality, representativeness, and curation of the training data. Therefore, clinicians should inquire about validation studies, performance metrics in similar populations, and governance systems. 

Human oversight

AI-generated findings should be interpreted alongside patient history, examination, and other evidence. The “human-in-the-loop” model means the clinician retains final responsibility, and AI serves as a decision-support tool rather than a replacement for professional judgement. 

However, human decision-making is prone to bias, time pressure, and fatigue – weaknesses AI aims to address. To prevent human error from compromising robust systems, some suggest a 'human-over-the-loop” approach – system-level governance with audits and periodic reviews. Clinical reasoning for accepting or rejecting AI output should always be documented. 

Informed consent

For consent to remain valid in AI-enabled care, patients should be informed when and how AI has contributed to their assessment or treatment planning. This does not require technical detail, but it does require clarity in communication. 

Patients should be made aware of:

• The role of AI in their care

• Its potential benefits and limitations

• The possibility of false positives, false negatives, and data-related bias

Crucially, patients should be reassured that responsibility for diagnosis and treatment decisions remains with the clinician. The use of more explainable (“glass box”) systems may support clearer, more meaningful patient communication.

Documentation

Clinical records should clearly document AI use, including:

• The name and version of the AI system employed

• Its role within the clinical pathway

• The nature of its output

• The extent to which its output aligned (or conflicted) with clinical assessment

• The rationale for the final decision

• Confirmation that AI use was discussed with the patient

Such documentation supports clinicians in dentolegal matters and will likely become the global standard where AI is used in clinical pathways and treatment planning.

Looking ahead

The global AI in dentistry market is expanding rapidly, with estimates suggesting a compound annual growth rate of around 20 per cent over the coming decade.

By 2030, AI-assisted diagnostics are likely to become standard, with regulatory and ethical frameworks hopefully keeping pace. As AI literacy increases, adoption may shift from caution to confident collaboration. 

Despite progress, essential aspects of care remain human: judgement, communication, empathy, and what Aristotle called phronesis – practical wisdom shaped by experience and moral judgment – qualities that are not easily reducible to algorithmic rules. 

References

The INFORMED and RECORDS frameworks were developed by Raj Rattan to support safer, ethical AI-enabled dental practice, embedding the principles discussed above. Find out more by visiting the  Dental Protection AI: Safer Practice Framework page.