This is me rambling.
Imagine walking into a large retail store a few years from now.
The store’s AI is watching sales across every aisle. It knows which products people pick up, which displays make them stop and where they tend to gather. The cameras already hanging from the ceiling give it a view of how people move through the building. Connect that behavior to inventory, pricing and point-of-sale data, and the store can start changing itself in real time.
That could mean a flash sale when an aisle goes quiet. It could mean a small discount on something people keep looking at but don’t buy. It could also mean raising the price of a popular item because a crowd has formed around it.
These don’t need to be huge price changes. A few cents here and a dollar there, repeated across thousands of products and hundreds of stores, becomes real money fast.
During Christmas, maybe the system notices people piling into the toy section. It lowers prices on a few popular toys to get families into the aisle, nudges premium items upward and makes sure another shipment of whatever everyone wants is already moving toward the store.
From the retailer’s point of view, that’s beautiful. Pricing reacts immediately. Inventory goes where demand is heading. Fewer sales disappear because something ran out at the wrong location.
But let’s change the situation.
There’s a heatwave. People start buying more bottled water. The AI sees demand climbing and does exactly what it was designed to do: it raises the price.
That may be great optimization. It may also mean the person with the least money pays more for water on the day they need it most.
The AI isn’t being cruel. It doesn’t hate thirsty people. It’s following its objective and finding money we left sitting on the table.
That’s what worries me about systems like this. We can tell an AI to maximize revenue, then act surprised when it discovers every unpleasant way to do that. “Increase profit” sounds like a clear goal until the system starts testing the moral boundaries we forgot to give it.
Retailers will need rules that aren’t optional suggestions. Essential goods may need hard pricing limits. Some customer data shouldn’t be used for pricing at all. Someone needs authority to stop or reverse decisions when optimization crosses into exploitation.
AI could make a retail operation dramatically more efficient. It could reduce waste, keep shelves stocked and help customers find better prices.
It could also become the fastest price-gouging machine anyone has ever built.
Same technology. Same data. Different boundaries.
That part isn’t the AI’s decision. It’s ours.
By Matthew Williamson · Originally published on LinkedIn · June 24, 2024 — read the original →