On AI Preview (again)…

Yesterday, I opined that we should extend the concept of “preview” to show how AI might interpret our content.

With responsive web design, we’re exposing our content and design to variable VISUAL containers, so we preview in different configurations. With AI, we’re exposing our content to variable COGNITIVE containers, so this should be previewed/tested the same way (so I theorized, anyway…).

I implemented a rough POC tonight. This was pretty simple with Optimizely CMP, since it has a framework for requesting previews from multiple services. When you request preview, the system can poll every channel and allow you to select from the previews provided by systems that responded. I just added “AI Summarization” as another preview supplier.

(Also helpful: we have an AI abstraction layer, as we’re quite committed to the idea of “Bring your own AI” for customers who want to use a specific AI engine.)

In this screencap, you can see that I’ve configured “previews” for three different questions someone might ask AI about the content. These previews are shown right alongside the variable visual previews (“Opti”, “Dark Mode”, etc.)

Not bad for something quick and dirty. I think the idea has merit.