Data science seems like it has a feedback issue. If you design an experiment poorly, it's easy to never know you learned the wrong thing. If you're building an

Data science seems like it has a feedback issue. If you design an experiment poorly, it's easy to never know you learned the wrong thing. If you're building an ML model and it's less accurate that what an expert would build, you might never know either. When you build many other kinds of products, it's easier to know whether they're working. Data products don't seem that way. I wonder if that could lead to weird "market for lemons" sorts of things.