Vertical AI, and the duty to explain

This is the aspect I think will matter more and more as AI moves into tightly regulated industries.

What vertical AI is

Picture the difference between a general practitioner and a deep specialist. The generalist knows broadly. The specialist understands one field deeply and knows they must be able to answer "where is the basis for this diagnosis". Vertical AI is a model trained and operated for one specific field such as medicine, finance or law, where every recommendation has to be traceable to a source and compliant with the rules. That quiet work is what people call compliance and traceability.

The seriousness of this direction shows in the numbers: as of 2024, the United States Food and Drug Administration had cleared more than 950 medical devices with AI built in, most of them in diagnostic imaging. Each of those devices had to demonstrate its safety and its basis, not merely that it worked.

How this applies to running a business

I find the lesson applies to every business, including outside the regulated fields: once AI touches a customer's entitlements, the most important question is not "did it answer well" but "what is it basing that on, and can we trace it back".

An AI assistant that makes a recommendation without being able to state its basis is hard to use for anything consequential. So when choosing a tool for sensitive work, favour one that allows you to trace the source, record the reasoning, and verify, rather than one that simply produces a tidy answer.

In the important work you plan to hand to AI, can you trace back why it recommended what it did?