Anthropomorphizing language can be cute when applied to your favorite car, but it helps to muddy the discourse when applied to tech sold as "AI", especially given all the boosters and AGI-cult members peddling their nonsense about imminent artificial minds. New from me & Nanna Inie on Tech Policy Press -- how to spot & revise away from anthropomorphizing language applied to "AI":
https://www.techpolicy.press/we-need-to-talk-about-how-we-talk-about-ai/
@emilymbender
THANK YOU! I've been saying this to everyone, even professors, for years now. Most of them do not have the remotest idea what the machine is actually doing, and so it's very easy to slip into the space of believing the metaphor, at least at a de facto level, since there is no real knowledge beyond the metaphor. I don't even call LLMs and text and image generators AI anymore, because it implies that they're doing much more than they really are, or that they "could" do more, if we just continue to give them infinite access and processing power.
@emilymbender It cannot be cute when applied to any car, but the point _absolutely_ stands.
@emilymbender Excellent article.
At the moment, I am wondering why this #anthropomorphizing #AI is going viral at a time when many people are #alienated from #nature, attribute little or no #intelligence to animals, or ridicule the environmental personhood of nature.
We are quicker to attribute personality to an LLM than to a wild boar; we attribute non-existent intelligence to machines, but not to a bee (who undoubtedly demonstrates a kind of intelligence).
What does that say about us humans?
@emilymbender Spot on! Extra thanks for using/attributing imagery from Better Images of AI instead of the usual blue tint robot stereotypes or the candy comic junk from the banana machine.
@emilymbender
Great article. Thanks for sharing. The ubiquitous chat interface is a huge problem. And in the that regard I’d like to ask, do you know any project that aims to provide a different interaction model for language models?
@davidcoronel I don't, but also I don't think LLMs are worth "interacting" with at all -- so why?
Unless you mean something like spelling correction or code completion? (Which can be done with much smaller models built on curated datasets.)
@emilymbender
Just curious about systems that use language models without anthropomorphism since I have an intuition that it is possible to find/build such systems.
@emilymbender I think your point on accountability is something that needs to be talked about more. I agree that a lot of the anthropomorphizing language serves to distance companies from being accountable for the output of AI models, and I think that's a big part of why I feel uncomfortable when people refer to AI in that way
@emilymbender Excellent read, thank you for this! I will carry a printed copy and a stapler with me at all times. It will make my conversations about AI so much shorter and easier on my part.
@emilymbender This is just something rattling in my brain. I Worked ages ago in the mental health consumer space; had some great conversations in the office about how people talked initially about doctors in a way that made them seem like mystical powered wizards that could learn and adapt to fix all their problems, and the other extreme, usually after a setback, where they're lied to. The language sounds very similar, maybe its time to shift towards person centric AI like the industry did
@emilymbender "Artificial" is a fair adjective, as in, meat nets and matrix maths are not the same resolution, but do share delightfully parallel diseases like hallucination, on very different scales. My computer dies too, don't cry, just needs charge. But as a long time unix admin, I am *happy* to replace "agent" with "probabilistic daemon", and readily demur, if saying "probabilistic automation systems" pisses Sam Altman off as much as "Stochastic Parrot", I will march with Dr Bender again.
"AI" is an alter-ego for Tech Bros.
@emilymbender thank you for the article. I wish I would have known about before I flew out to visit University of Washington last year. Loved the people and the campus! Thank you for all that you do!
@emilymbender There is word for seeing human faces in non-living things: pareidolia....
Maybe the AI craze (or rather: the craze using AI not for what it was intended) is the latest form of pareidolia.
@emilymbender Appreciate this take. I have admittedly always felt uncomfortable with non-human things being called she, her, or using human names, be they boats or Alexa. I think it’s a way of reinforcing relationships that don’t actually exist, which this article excellently illustrates. Thanks for sharing.
@emilymbender Excellent article, thanks!
Part of the issue is that LLMs often work best when they are prompted to act as a person in a specific role. "You are a skilled X" is a common system prompt type. This kind of roleplaying setting seems to trigger outputs that resemble what a human in the same situation could produce, which is what the user wanted (or at least the best the model could do, within its limitations). So the anthropomorphism cuts deep into the behaviour of the model.
@emilymbender Thank you for this article! 😌
@emilymbender excellent piece (and well-timed for me, as I’m updating my reading list for my “Writing about AI” class…I think I’ll be adding this!)
