The scientific method requires the formulation, testing, and modification of hypotheses using systematic observation, measurement, and experiment.
Essentially, it combines the Rationalism of Descartes with the Empiricism of the philosophers Locke and Hume.
The scientific method has worked reasonably well in Medicine, which in practice usually blends knowledge of the pathophysiology of disease with evidence-based medicine.
An example of this in healthcare? Most doctors will, in practice, combine expert opinion and knowledge of pathophysiology with the understanding provided by the larger empirical studies to reach a diagnosis or optimum treatment approach for a given person.
But, in the subset of AI called Deep Learning, this may be about to change.
Why? Because – unlike Expert-based Machine Learning – Deep Learning needs no a priori hypothesis about a given disease’s prognosis to being meaningful. Instead it just needs lots and lots of data.
In other words – theory doesn’t count ( at least much).
Whether or not this is a good thing or bad thing has yet to be determined – but one thing is for sure: it is going to be a thing.