What is it about language? And why does a chatbot destroy more than it "generates".
Having wondered for the longest time about the consequences of using Generative AI as a replacement for our own language, I thought it best to finally resolve the irony in the situation of using Generative AI without my own thoughts on this phenomenon being adequately expressed.
Large Language Models are less generative, more destructive (and I’m not just talking about the data centers) - this is equal parts about cognitive decline and decline of the traditional interwebs.
If you look at the original purpose of the internet - it was hardly meant to be used for means other than information and scientific communication - and in that sense, the proliferation of Large Language Models is one that shows it can satisfy, but in reality, betrays these goals. If, for millenia, humans had been sharing their information through different forms of writing, it is necessary to look closely at what it is that unifies the structure of this activity such that it merely reinvents itself at every turn.
Songs are not sung before they’re written.
Movies are not produced before they have a script written.
Heck, even our destinies are “written”.
So, where then do we place Generative AI and it’s style of writing then? It creates such horrid by-products of what the internet had archived for the last couple of decades and in this way betrays the actual purpose of KNOWLEDGE itself!
I’m not a linguist but a few things that occurred to me are that:
- Language is a key part of communication. And with regards to this function, there is a very popular theory that I have often seen reproduced in textbooks - it is the “Transaction Model of Communication”.
Image source: Transaction Model of Communication under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
- What LLM’s do is - readily scrape the internet for context, and learn from large datasets - completely blind to intrinsic meaning - things that are often not understood or expressed at all or even with the least bit of concurrence in semantics.
For example, there’s this podcast I listen to about Urdu poetry - and in each episode, the host brings up different kinds of interpretations of a single word in different contexts. They recently did an epicode with the many meanings of the Urdu word “Dagha” - urge you to look at it here - https://www.thequint.com/podcast/urdunama-podcast-the-many-meanings-of-daghaa-berrayal-in-urdu-shayari
The next step - LLMs provide an answer, or perform a coding task or some such automation, with complete unawareness of the latent meanings in a conversation or the logic that does not get expressed merely by creating mathematical representations.
This is thereafter implemented and acted upon by the immediate recipient of that communication without any awareness of what led the LLM to actually produce that answer. This happens simply because we don’t think of the world as a mathematical continnum of linguistic variables. Even if a completely open-source, open-weight LLM would create an imperative for transparency - we are still left looking at best at a proficient liar.
Which gets me to my next point - asking an LLM is like asking a person who has never been held accountable in their life - because this LLM persona just keeps coming up with different answers if you come up with a new way of phrasing your question! You can call it sycophancy but this is so much worse, because it brings in a whole new level of unpredictability to the already vexatious conundrum brought upon by the unnecessary mathematicisation of language.
Next up, and worse of all, is the fact that it cooks up answers that it just scraped from the internet like a schoolboy who mugged up every answer in an English Literature workbook.
For all these reasons, I aim to limit my use of Artificial Intelligence. Not to mention the way in which it swallows up an industry and then throws up a SaaS product built on an exploitative subscription tier.
In the end, all I can hope for is that we continue our millenia old tradition of socialised knowledge creation through writing and information sharing. That chase is better than whatever the hell this catch in the muddied waters of our dwindling environment is.
Till then, thanks for reading.
