(C) Alec Muffett's DropSafe blog.
Author Name: Alec Muffett
This story was originally published on allecmuffett.com. [1]
License: CC-BY-SA 3.0.[2]


BRILLIANT IDEA: let’s bring about the end of the internet by embedding locally-running LLMs into Web browsers!

2024-02-06 21:42:48+00:00

I’m having {a, yet another} discussion about software regulation and how it’s an illiberal and misconceived approach to try regulating, export-controlling, or otherwise restricting access to expressive speech any kind of software, even if it is an ostensibly malicious tool, because the concept of dual use which applies to practically everything on the internet.

This led to a diversion on code and data being speech, and becoming increasingly human speech, which may lead in future to regulatory confusion…

…perhaps it should be difficult to constrain free-expression, even when that expression happens to be computer code which might be used for malicious intent. Code is clearly a form of speech even if it is not regarded as protected free expression in a variety of jurisdictions.

It’s going to get really interesting in the next decade or so, simply because LLMs and AIs interface with the world using natural language which is going to be undeniably “speech”, so we could end up with language which is ok to utter to a human being but would be regulated in some fashion if it is uttered to a computer.

I had a discussion similar to this on Bluesky earlier today: Imagine that somebody embeds a locally-running LLM/AI into a web browser, so that it can read web pages for you and summarise them for you, perhaps translates them for you, or provides some sort of assistive mechanism for visual impairment. Using that web browser you visit the Facebook group which includes “dating reviews” of various named individuals and how they performed as a “date”. Then you tell your web browser to:

"Click through the pages of this Facebook group for me; aggregate all of the names and the sentiment and opinions which relate to them, and correlate and reduce the names to remove duplicates. Produce a CSV of all names and map the associated sentiment for each name, normalised in the range 1 to 3."

Presto, you have star ratings of single men (or: women, nb, whatever) in New York; tell me that’s not going to terrify somebody?

Such terror means that somebody is going to try and regulate the functionality and basically attempt to enforce that “human beings have to do all of the hard work”, because it would be technically possible (though practically infeasible) to do this with pen and paper. That which would be legal for a human to be tasked with would be illegal for a computer to do.

If this does not sound relevant right now, consider that it is exactly the same problem as the popular “artificial intelligence will be used to create a tidal wave of misinformation that will overwhelm governments and newspapers” – (bogus) story of the moment, pushed by people who imagine that “scale changes everything, but a misinformation sweatshop staffed by minimum-wage workers does not require urgent legislation”.

Yet this is also why it is a bad idea to attempt to regulate the shape of code: because nobody in their right mind wants to prevent the development of assistive technologies.

Dual use is everywhere.
[END]

[1] URL: https://alecmuffett.com/article/109116
[2] URL: https://creativecommons.org/licenses/by-sa/3.0/

DropSafe Blog via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/alecmuffett/