Science Experiments vs Demonstrations

Vancouver's Science World is a great place to take kids. There are all sorts of cool installations and a constant rotation of science activities: The electricity show, the balloon show (culminating with a hydrogen balloon being set on fire) and a bunch more.

What's neat is that at the beginning of the electricity show, the presenter made it clear that this was a demonstration, not an experiment. She explained the difference like so:

"In an experiment, you don't yet know what will happen. In a demonstration, you do."

That's a neat way to think about when your idea does and doesn't need a prototype or proof of concept: If you don't know what will happen, you need one. This is why common startup advice about fast iteration of minimum viable products doesn't quite fit with ideas that rely on data and AI. With a non-data product, "just start building so you can show something to customers" is great. The most important question there is whether people care enough, and you can only answer that by showing them. Here, a POC would be more like the science demonstrations at Science World. No matter how awesome to behold they are, you as the conductor are not learning anything new.e

But with a data or AI product, the first question is whether it can even be done: If we mix these data sources and build those models, can we make such and such predictions with this much accuracy? Who knows? And if the answer to that makes or breaks your product, it's your number one risk and therefore needs to be experimented on, as quickly as possible.