I would be interested in the kind of model which will tell you something is wrong when you show it a black square and call it a black square, then show it a whi

I would be interested in the kind of model which will tell you something is wrong when you show it a black square and call it a black square, then show it a white square and call it a black square. Not the kind of model which would average the results and adapt to nonsense.