I guess I was wrong about AI persuasion

https://dynomight.substack.com/p/persuasion

by paulpauper • 11 months ago

8 2 11 months ago

dynomight.substack.com

jnxx• 11 months ago

What I think he is wrong with is that LLMs are, despite being named as "Artificial Intelligence", not intelligent .

I guess that some people confound the source of information with its medium. A book might contain Newtons equation of motion, Gauss' works, Maxwell's equations, or an introduction of Einstein's theory of special relativity. But it is not the book that is intelligent, it is the people that wrote up the ideas.

You can see that confusion between the medium and authority by the holy scriptures of the large religions which were among the first human organizations that employed written words for their purposes - for example, muslim and Jewish people treat their respective foundational books as holy themselves.

With LLMs, it is clear that when you average the writing of many more or less intelligent people, the result might appear more or less intelligent. In the same way as common sense, sometimes, does make sense. But it is still not the medium that is intelligent, but the authors of the input.

But what the blog author is probably right about is persuasion by influencing communities which in turn influence individuals. That is exactly how Googles or Facebooks practices that violate privacy - like offering free emails accounts which use data from personal mails for advertising - became widely accepted even if it is clear they are not good for individuals.

I think that dimension of persuasion might be often overlooked when people, for example, think about the Cambridge Analytica scandal:

https://longform.org/posts/the-great-british-brexit-robbery

https://www.europeanpressprize.com/article/great-british-bre...

https://ia601000.us.archive.org/28/items/civ53332855/great-b...

amenhotep• 11 months ago

He never actually mentions LLMs once. It is explicitly a post about a hypothetical implicitly artificial being with 300 IQ that can communicate at 10,000x speed. LLMs are an interesting background to this as an illustration of how people might end up interacting with it, but no more.

I guess I was wrong about AI persuasion

Related Stories