Fair enough. I hope I was clear that I do think that computer programs can understand. I just don't think that optimizing for solving the Turing test will generally yield a computer program capable of "understanding" or "super intelligence".
Whether, and to what extent, a given Turing test session probes understanding depends on what the interrogator chooses to ask. The training of ChatGPT does not contain any explicit bias towards demonstrating an understanding of anything, and it seems that a reasonably determined and detached interrogator should have little difficulty in identifying it on this basis.
