Artificial Intelligence (AI) appears as a transforming force in dentistry, promising optimization of diagnosis, medical triage and work flow automatization. The present paper analyzes the potential of Machine Learning systems, especially of the convolutional neural networks (CNN), of surpassing human performance in imagistic interpretation, due to their capacity of processing data sets at a scale inaccessible to a single clinician. However, this performance is critically conditioned by the quality of the training data. The study investigates the fundamental risk of the “algorithmic bias”, in which an AI pattern trained on data containing human errors of diagnosis will learn, reproduce and amplify systemically such errors. On the other side, the AI potential of reducing the error rate when trained on “gold standard” data sets validated by experts is discussed. The conclusion reached is that an ethical and efficient AI implementation in dental practice does not mean substitution of the physician, but the development of a human-machine synergy, in which the clinician performs the final supervision and validates contextually the recommendations of the algorithm.
Keywords:- algorithmic bias
- Artificial Intelligence
- assisted diagnostic
- convolutional neuronal networks (CNN)
- dentistry
- human supervision