The data will draw on everything from census findings to driving habits gathered from vehicle sensors to—arguably most consequentially—medical records.

Sifting the literature for real-world challenges thwarting adoption of clinical AI across medicine, a team of biomedical engineers and computer scientists has identified and fleshed out an exemplary use case.

Many parents would let their children be read to by robots as long as the device didn’t project a little too much lifelikeness.

Along with AI and machine learning, the list may include virtual and augmented reality, 3D printing, robotics and other technologies currently changing healthcare delivery.

Healthcare AI has potential not only for neutralizing its inherent algorithmic bias but also for personalizing its outputs to help humans address health inequities.

Because they learn as they go, machine learning models for drug discovery have to be continuously re-trained for changing conditions in drug production processes.

Black-box AI should be barred from reading medical images in clinical settings because machine learning, like human thinking, tends to take diagnostic shortcuts. 

When the network interpreted complete videos from 42 consecutive patients, it boosted detection rates by as much as 50% over physician-alone reads.

Upon examining a skin lesion they suspected of being malignant, few dermatologists—only 8%—would hold back from performing a biopsy if an AI tool disagreed, classifying it as benign.

The system hit 88% accuracy at optimizing stimulation settings, as confirmed by brain-response patterns on neuroimaging as well as visibly observable symptom improvement in patients with Parkinson’s disease.