Apparently Wendy’s is trying to do this and I’ve heard of stores here in Sweden being on the same wavelength. I see the merits of surge-pricing for grocery stores, fast food restaurants and taxi-like(apparently Über has been doing this for some time) but why would you use AI?
I think you – as a matter of practice – need to be able to explain why your prices are what they are. Saying “It’s arbitrary” isn’t going to cut it and saying “The AI says so” is pretty much equivalent. I don’t think surge-pricing is hard to do. You know when you have peaks in load over the day and you can set prices accordingly. “The price of milk is $3 because between 16:00 and 19:00 we see the majority of people in the store which necessitates the highest staffing.” <- That seems like a reasonable rationale.
I’m curious though, what happens when you put a carton of milk in your cart at 15:58 when the listed price is $2 and go to the checkout at 16:02 and get charged $3? That’s not great legally and even from a PR-perspective it won’t go over well. You could say that prices change ever whole hour which might be reasonable with respect to PR but that still leaves the law and I’m not so sure we pass that hurdle this way.
You also need to avoid maximizing algorithms even if they are deterministic. They are better than AI because you can follow the rationale but it’s not going to look great to charge old people more because they are slow(for instance). Just because it yields more money doesn’t mean it’s going to work in the long run or… be legal.
AI is fine for generating images or text where a human can say “Yeah, that’s fine” but prices for stores? No, not a good idea. Similarly I’m quite perplexed by the use of machine learning in self-driving cars. Surely when people are injured because the car drove itself into a wall or a crowd you will want to establish why. But that’s no something you can get from machine learning. It’s an opaque process that yields something that provides opaque solutions. Why would you want that somewhere safety-critical? Even if it turns out machine learning yields results that are five times better than any human driver, the question why is still going to pop up when mistakes are made.