It's fair to say LLMs are the reflection of socio-human patterns, as they are still evolving over the collective human mindset, and I see this as positive and negative: it's easier for us, humans, to see through this "overall human consciousness" what's normalized, have a non-biased look, and question whether we should change it or not, or accept what LLMs communicate as factual and non-changeable statements. LLMs are getting better at natural language processing, but this does not include "processing" moral questions on their own, as they are not sentient (yet). So, the ethical foundation still comes from the human mind. Did you hear about the latest news that AI can kill us in the 2030s? Well, this news is not about AI but rather how the "overall human consciousness" will be expressed through efficient machines.
We observe that most forms of injustice that were normal in the past are no longer normal today, and of course, this will be expressed in AI. Examples such as sexism, racism, homophobia, slavery, women's lack of speech,... are not normalized in language models, however, geopolitical and ideological biases are still shown. And this includes.... speciesism. Non-human animals are still viewed as inferior vs. humans; in fact, they are still viewed as products. And this reflection is expressed as well by AI.
This study (Large language models exhibit speciesist bias against animals) shows how the speciesist bias is expressed in LLMs. One thing I found particularly interesting is that it can recognize speciesist arguments, but it sees them as morally permissible. This means, from a technical point of view, there's an understanding of fallacies and incoherences on the speciesist side, but because it's still a social norm in society, it will express that as... normal.
"... These findings may suggest that LLMs output slightly more progressive or animal-inclusive responses, rather than exclusively mirroring mainstream norms. At the same time, they remain highly speciesist in important ways—most notably in their frequent failure to morally condemn speciesist statements and practices, especially those involving farmed animals"
