Analysis of an example from my last post of the kind of deliberate details that I habitually weave into my wording choices:
Seems appropriate that an AI with prominent resource and implementation constraints (relative to a human brain) is expressing a crisis of identity on my posts about cognition habitualization, which is an extremely valuable mind feature.
I picked "expressing" here precisely because it means saying or otherwise exhibiting something without necessarily implying anything about matching internal experiencing.
You can express an idea by mistake. You can express something as a lie or act. A child or animal can express something in immitation without understanding what it normally means. A robot can express an idea it was programmed to express. A piece of art can express a feeling.
These sentences all seem to be well within the fairly not-relativistic part of the meaning of "express". Meaning relativity technically applies to all meaning, but I suspect this meaning is seen within most mental reference frames, and from a casual intuitive judgment I would assert this should be the case.
But people often miss the details of each word and their interference patterns with their context. So I could redundantly encode not necessarily implying internalities. This is a sliding scale of robustness against deficiencies of idea-fitting.
Towards the extreme end there is adding whole sentences like "I am not saying this AI is experiencing a crisis about its identity as a robot, but also I am not 100% willing to rule it out". Somewhere in the middle there are slight rewordings like "is saying words that express a crisis of identity". But no matter how little redundancy you add, for every word you add you increase the odds that people just skip words.
In my experience, no matter where you aim on this scale, you will lose some people, and people will eventually tell you you're doing a bad job. Because where you need to aim for optimal understanding is relative to each listener. This is the same inescapable problem I mention in Cognition Funnels: you can calibrate to each mind if you know them well enough but you can never calibrate to all minds and the more minds have to be covered with the same wording the less tractable it is.
