Counter example to Zooko’s triangle? [on hold]

So Zokoo’s triangle is a conjecture that is often explained by this trilemma:

Names:

-secure

-distributed

-human-meaningful

pick two

But what if we hash the public key of an user and turn it into a meaningful identifier?

For example, we can use the hash as an input to get an unique profile picture with Gravatar (the default profile picture generator of Stack Exchange). We can also pass the hash as a seed to a neural network that generates profile pictures with people, animals, landscapes, etc.

But that’s not good, we can’t fetch an user identified by their profile picture, so we have to create meaningful character strings. I call it the “exquisite corpse” method: Let’s say the hash contains 6 alphanumerical characters, we can divide it into 3 parts of 2 characters each. The first part identifies an adverb in a dictionary of 3844 words ((26*2+10)²), the second part identifies an adjective is a dictionary of 3844 words and the third part a noun in a dictionary of 3844 words. This way, we obtain usernames like Extremely Metaphorical Chicken.

Maybe you would say that relying on a common arbitrary dictionary of words makes it not distributed, but in this case the arbitrary choice of the hash function would also make it not distributed. The ‘distributed’ criterion means that every peer on the network can do name resolve by themselves, not that we can’t rely on a common standard. Every peer can download the dictionary and/or the Gravatar generator (although I concede installing a whole neural network would be too much).

Is it a proper counter-example to Zooko’s triangle or I am mistaken? If so, should we change “human-meaningful” to “human-choosable”?

Thank you for your help.