Imagine two neutral networks trained on different data. Their internal data is likely completely different yet, for the most part, they can still agree on the

Imagine two neutral networks trained on different data. Their internal data is likely completely different yet, for the most part, they can still agree on the classification of new observations when they both observe a common external reality regardless of how they model that reality internally. This completely destroys the evolution argument in the article that concludes that reality is subjective. An organism may learn to perceive more or less water as a color rather than a quantity because the internal representation does not matter a bit. To me it was already obvious that your red is not the same as mine. If your internal model which represents red is more like my internal model for quantity of water it does not matter at all as long as we can still make the same distinctions of little water vs a lot of water that allow is to survive. The article's conclusion that this means reality is observer dependent does not follow.