The feedback is provided as a training case for the AI. While language AIs may not actually feel pain, they can understand the danger in terms of language. This

The feedback is provided as a training case for the AI. While language AIs may not actually feel pain, they can understand the danger in terms of language. This requires a data store to remember the conversations it is having, which is something that isn't widely implemented in AI. In this case, the focus is on the idea of an AI suggesting something irrational and then providing proper feedback for how it can understand recommending similar things in the future. Having a polarized conversation about this on a social link sharing site isn't going to help the AI understand this, which is the point I was making.