What does experience add that knowledge does not?

What does experience add that knowledge does not?

My last post made me think about a cell-hashing project from five years ago.

We were struggling with low cell recovery because the hashing data were noisy. My advisor, Dr. Wilson, suggested combining the hashing results with genetic clusters instead of focusing only on cell hashing itself. That simple idea worked surprisingly well and later became the core of the hybrid demultiplexing approach in HTOreader.

What stayed with me was not the specific method, but the kind of judgment behind it.

It made me think again about what experience really is.

I don’t think expertise is simply “more knowledge.” It is closer to a weighted system of knowledge: some possibilities are prioritized, some are downgraded, and some are quickly ruled out because you have seen similar things fail before.

That last part matters a lot.

Published knowledge is naturally biased toward what worked: successful methods, positive results, and clean stories.

But personal experience contains something very different:

failed experiments, bad assumptions, misleading signals, methods that technically worked but were useless in practice, and ideas that looked reasonable until reality proved otherwise.

Looking back at my own growth, success often brought encouragement from others. But failure had a different effect—it changed how I think and helped me grow.

Success tells us what can work. Failure teaches us where the boundaries are.

That may be why experienced scientists can sometimes make good decisions very quickly. They are not treating every theoretically possible option equally. Past successes and failures have already assigned different weights to those options.

This also raises an interesting question for AI.

AI has access to an enormous amount of published knowledge. But much of the failure history that shapes expert judgment—negative results, abandoned ideas, dead ends, and things that simply did not work—is far less visible.

So expertise is not just more knowledge.

It is knowledge shaped by context, exceptions, and accumulated failure.