“The measure of intelligence is the ability to change.”   Albert Einstein


This letter to the editor in Nature is noteworthy as machine and deep learning are becoming more pervasive in decision support in clinical medicine.

The authors, members of the ‘Developmental and Exploratory Clinical Investigation of Decision-support systems driven by Artificial Intelligence (DECIDE-AI)’ group, describe a robust early and small-scale clinical evaluation stage to be interposed between the in silico algorithm development/validation and large-scale clinical trials evaluating AI interventions.

This laudable effort is proposed to eliminate the AI chasm between the data science and the clinical medicine that currently exists in many such projects. The authors then go on and present cogent arguments to support the need for this intermediary development stage and its adequate reporting.

The development-to-implementation gap should be narrowed not only with the aforementioned better guidance on the reporting of human factors and early-stage clinical evaluation, but a continual partnership from the inception of the project (think of this as “co-parenting” a child from conception, not another marriage).

So an even better concept is that of a “Design AI”- the data scientists actually round with the clinicians in the clinics or hospital units to jointly “conceive” ideas for projects with the clinicians.

Too many data science in clinical medicine projects are launched with too little clinical relevance or impact, so even if the model performs adequately and generalizes, the clinical impact is very small or even nonexistent.

To reengineer a project after it is launched initially by data scientists is sometimes futile in terms of capturing a sizable impact. In addition to the initial conceptualization, an ongoing working partnership is essential for the best feature selection and engineering possible for the project.

If clinicians can get involved from the inception of the project and have an opportunity to work on the project along with the data scientists, then there is ownership and accountability. Of course this strategy mandates the availability and commitment from the clinicians.

The full correspondence can be read here