What if AI Debated Every Document I Opened by Default?

It seems like Perplexity (or something similar) could do this, and it would solve a lot of problems

Today, OpenAI released “AI in America OpenAI's Economic Blueprint.”

It lays out a fundamental case for US government efforts to support AI development by

• Reducing regulation

• Providing more financial support to the industry

• Collaborating to solve obvious downsides (bioweapons development, child abuse)

• Supporting energy development.

As I read it, I thought some of the claims were vague in some areas, but I didn’t have time to research them all, so I thought I’d ask Perplexity for help.

I first asked it, What are some problems with this argument?

It gave me this response

I then asked it to unpack what it called a “false equivalence.”

Similarly, I asked it to critique this claim

It responded.

And what are some problems with this argument?

This got me thinking —

• Why can’t every article/pdf be put through a claims evaluator where I can click on a claim and see counterarguments or critical questions?

• Perhaps I could click further and see a response.

This seems like something Perplexity could do.

When I open an article in some “system” (for lack of a better word), why can’t I see the critiques of the claims and the counter responses?

This would

(a) Improve the quality of original arguments (people wouldn’t want weak claims that so easily challenged.

Imagine that before sending out an argumentative piece, you were first incentivized to test it against an inevitable AI challenger? It seems like you’d be inclined to make better arguments.

(b) Help individuals see and be more open to weaknesses in ideas presented to them.

© Make people more capable of engaging in dialogue about important issues rather than simply repeating talking points made by individuals and institutions whose “side” they were already on.

(d) Enable people to see alternative points of view without changing the “channel/station” and/or alienating friends and loved ones.

(e) Solve the problem of AIs being able to “manipulate” us, as the alternative side would immediately be present

(f) Get rid of a lot of bullshit

(g) These dialogic interactions would produce robust and unique synthetic data sets for continued model training (follow up with Anand Rao)

Also, it seems that I should be able to engage something like a Notebook LLM podcaster on the argument map, asking clarifying questions and having them help me through the discussion

If I had a system where I could read anything digital that I opened (including in my Kindle)

__