You can find this blog post here: https://www.programmablemutter.com/p/large-language-models-are-cultural
The post itself does not need much of a summary, instead I recommend and advise the people who are interested to read the full post at the url posted above. Here are simply the 4 perspectives, paraphrased:
- Gopnikism (from an article by Eunice Yiu, Eliza Kosoy, and Alison Gopkin https://arxiv.org/abs/2305.07666 ): LLMs are not agentic intelligences, but powerful cultural technologies. Why? because they lack direct contact with the real world, which also affects how they can potentially learn from the world and innovate (create new ideas). LLMs possess “no direct feedback” and are “not updating”, but still they have transformative power.
- Interactionism: focuses on how humans are likely to interpret and interact with the outputs of LLMs. Here, the take is almost biological or evolutionist: “from birth onwards, human beings expect relevance from the sounds of speech” (Dan Sperber). It’s not so much about whether or not there’s agency in LLMs, but more about what we normally expect from speech.
In this regard, what Farrell quotes about LLMs is most interesting: that hallucinations from LLMs “may glide easily past people’s innate skepticism, … because hallucinations are highly probable and we do not expect that the machine will deceive us (why would it? no explicit intentions there). This means that “LLMs as mass contributors to our information ecosystems could disrupt its quality” (quoting Adam Sobieszek and Tadeusz Price’s 2022 article: “Playing Games with Ais: The Limits of GPT-3 and Similar Large Language Models”). Farrell’s asks particularly about “what kinds of feedback loops” we are “likely to see” between human inputs and LLM outputs, which is something that Baudrillard would call the inner dynamic of a “hyperreality.” - Structuralism: A descriptive term borrowed from Weatherby who seems to describe LLMs as systems that exist independently of a world and of humans: “Weatherby suggests that the advantage of structuralist and (much) post-structuralist thought, is that it allows you to examine LLMs in their own right, rather than in reference to something else” – if this is an accurate description of Weatherby’s position, then I can only be amazed: how can it be a flaw to see and understand something in its relations to other things? how can you even examine something “in their own right” without looking at its relations, its embeddings, its context?
Herbert Simon is also quoted here as a “structuralist” about the cultural significance of LLMs; where Simon seems to be identifying human languages with modelled languages, even claiming that “AI is simply generating culture.” Language is taken here simply as an “interface” between diverse things, speaking of a “computational semiology” without reference to the real world nor to language-users. Wonderful quote describing this out-of-the-world (mad?) belief: “The datafication of everything stands to become qualitative, to form a linguistic-computational hinge for other forms of data processing and the extensive text-first yet multimedia inter face of our daily lives and global social processes.”
Weatherby is quoted as a perfect addition: “We could call it spreadsheet culture in hyperdrive, a world in which all data can be translated into summary language and all language into optimized data with nothing more than a prompt. But where spreadsheets had limited functionality, LLMs act as universal translators in the same arena.” – “we have spent the last three decades making virtually the whole world into a giant spreadsheet”. Well, yes. Many people did that. And there is indeed immense utility. But this seems to be completely evading any normative questions of “ought” or “good/bad”? Was it good that we did that? should we do that? (Weatherby is quoted from this article: https://www.theideasletter.org/essay/our-spreadsheet-overlords/ )
Farrell is really trying to see the positive in the position of “structuralism,” but I can only see huge flaws; praises of reductionism instead of critical engagement and arguments.
4. Role Play: