@pedrogonzlez- It isn't about that kind of perfection. Sure, robots will malfunction. But it is about known human biases. We know, for the sake of our brains we take shortcuts in our thinking. We can only juggle so much informaiton. We can only include so much data in our thinking. This saves energy and for the most part, it works for people. Think of it like "just good enough" product creation.
Humans can untrain themselves from these biases in short bursts and think very deeply, but for most functions it isn't necessary.
However, a robot or AI doesn't have these energy issues. A robot or AI doesn't need biases or shortcuts. So we can, in fact, program an AI to think more clearly than a person does on most decisions.
think about it this way: If i asked you to pick between two cars you were thinking of buying, what kind of informaiton would you look at? You might look at price, the color, the size of the seats. Research says you'd also look for the cupholders. You'd see how comfortable it was. You'd basically pick 8 or 10 things that were important to you in a car.
What would happen if I knew you were buying a car and dumped all that informaiton on you plus the blueprints of the car, the names of the people who designed it, the names of the facotiry workers, the detailed specs of the factory machines that built it, the moleclar structure of the paint on the car, the source code for the infotainment system, etc, etc, etc. You would likely ignore this information. It would be too much for your brain to handle.
It would not be too much at all for an AI to handle. But if a robot said to you, "don't buy car A, because I saw a flaw in the molecular structure of the pain" you'd laugh at it. That's the kind of imperfection/perfection we're taking about.