We use anthropomorphic for machines all the time, a cars engine doesn’t “run”, a computer doesn’t “sleep”, a surveillance camera doesn’t “see”. We just use these words because it’s a close approximation of what it is doing and it’s simpler to understand for normal people, why are LLMs different?
I think the difference is that we have a very intuitive understanding that computers or cars or cameras aren’t alive, while we can chat with LLMs and many people are genuinely convinced they’re “alive” in some sense. It’s important not to anthropomohphize LLMs so we can have a healthier relationship with them as tools, and not think of them as analogous to friendships or romantic relationships, etc.
I guess but it gets to the point of being annoying, saying shit like “the statistical text predictor machine predicted text in a way that I deemed to be wrong” doesn’t quite roll of the tongue as “the clanker fucked up”. I don’t think people that are prone to AI psychosis are going to be deterred by a media outlet using slightly different wording to explain LLMs.
Don’t get me wrong I’m glad they didn’t drink the techbro kool aid but I don’t think this is going to have a huge effect on anything.
To expand on what Elixir_Fan said: See the ELIZA Effect. This is not a new problem and it does indeed arise when people stop seeing the software as a tool.
We use anthropomorphic for machines all the time, a cars engine doesn’t “run”, a computer doesn’t “sleep”, a surveillance camera doesn’t “see”. We just use these words because it’s a close approximation of what it is doing and it’s simpler to understand for normal people, why are LLMs different?
I think the difference is that we have a very intuitive understanding that computers or cars or cameras aren’t alive, while we can chat with LLMs and many people are genuinely convinced they’re “alive” in some sense. It’s important not to anthropomohphize LLMs so we can have a healthier relationship with them as tools, and not think of them as analogous to friendships or romantic relationships, etc.
I guess but it gets to the point of being annoying, saying shit like “the statistical text predictor machine predicted text in a way that I deemed to be wrong” doesn’t quite roll of the tongue as “the clanker fucked up”. I don’t think people that are prone to AI psychosis are going to be deterred by a media outlet using slightly different wording to explain LLMs.
Don’t get me wrong I’m glad they didn’t drink the techbro kool aid but I don’t think this is going to have a huge effect on anything.
To expand on what Elixir_Fan said: See the ELIZA Effect. This is not a new problem and it does indeed arise when people stop seeing the software as a tool.