Explainable AI, where each step of the process is understood and can be independently audited, is a very far world from Gen AI. How can the benefits of GenAI be harnessed in a regulatory context?
Explainable AI usually necessitates using labelled models, and avoiding the use of more complex neural nets and GenAI algorithms that can't be explained. However, missing out on the tremendous potential of the latest, powerful models, isn't the way to go.
The Latest AI Voice relies heavily on prompt engineering - ensuring the promts are tightly scoped and relevant to avoid the hallucinatory issues with multiple conversation threads. How can this be tested reliably over time and monitored for false positives and false negatives. Only a rigorous testing framework can provide the data integrity and quality assurance needed for medical grade assurance.
