How we talk about generative AI - Intro

A few months back, i had a conversation with a friend about the impact of how we speak about LLMs (large language models) and current mainstream machine learning technology. From the very beginning of the LLM hype people have been writing about the problematic aspects of personifying the interaction with generative AI tech (trying to not use the word "chatbot" here, as it's already a personified word).

A couple of weeks after that conversation, Emily M. Bender and Nanna Inie published an article on exactly that topic: How to talk about "AI" without adding to the anthropomorphization. Their article is a start and an invitation to expand on the topic.

But first things first. Personification and anthropomorphization are, how should i put it, very common ways of humans to phrase and elaborate what's on our mind and in turn shapes the way we perceive and categorize our world. In all the languages i know, and maybe/probably in all human languages, there are not enough words to reference and describe all of the things and ideas and feelings and social aspects that we want to communicate on.

Coming from the literature side, the concept of personification is part of many more ways of figurative speech, such as metaphors, alliteration, allegory, etc. I must admit i was quite surprised to learn that there seems to be a difference in definition between personification and anthropomorphization. When checking back anthropomorphization on Wikipedia, i was greeted with the following disclaimer note:

This article is about the attribution of human traits, emotions, or intentions to non-human entities. For the representation of a thing or abstraction as a person, see Personification.

The personification entry shows the same in reverse.

"Attribution of human traits to non-human entities" vs. "representation of something as a person". Yes, there's a difference and maybe in an analysis the directionality is relevant. However, i would argue that in general, it doesn't really matter. It's about distinguishing and connecting human and non-human semantics.

So, anthropomorphization and personification is very common. Yet, the way we read and speak about LLMs is deeply framed by the ideological background of the first and foremost us-american LLM makers. Think TESCREAL. On the other hand, what i often read to be specifically human, is framed by the ideological background of the artistic genius or the artistic inspiration as something divine.

I hope to find the time to write more about all the various aspects of this topic.