AI Can Now Debate with Humans and Sometimes Convince Them, Too | Hacker News Reader

AI Can Now Debate with Humans and Sometimes Convince Them, Too

scientificamerican.com

37 pointsfurcyd14 comments

This comes as no surprise giving the repetitious, almost regurgatory nature of public debates these days. I can't remember the last time I heard a new argument made up on the spot. The Chinese Room thought experiment applied to debate wouldn't need an infinite book, you could get by with at most a pamphlet.

These days?

A good chunk of all arguments are a variant of the myth of the golden age, which has been around since the Greeks.

There's a saying that only three books have ever been written: voyage, love and war. I suspect the number of rethorical arguments is the same order of magnitude.

That's true in the abstract, but in the concrete there are unique lines of argument, contingent on details specific to a topic, that get invented shortly after the topic arises. It is at that level that you can literally have an indexed pamphlet to guide you verbatim through 90% of dialogues.

Injecting contextual facts into 500BC rhetorical forms is a little more human-level than repeating other people's filled-out Pnyxian madlibs, an activity that is a little below machine-level.

We should stop calling deep learning models AI, it gives unnecessary hype and misunderstanding of the capacities of the things, as well as a feeling that researchers aren't being honest about the capacities of the things they are developping

AI hasn't meant human-level general intelligence for decades now. I think that ship has sailed, if it was ever at port.

ML is part of AI, even if you are going by purely academic definitions.

I absolutely agree that universal approximator != intelligence, but we're just arguing semantics, here.

I’m curious what you mean.

Wouldn’t a class of universal approximators necessarily include some intelligent ones — for any measure of intelligence?

There’s obvious scaling issues, though.

The entire question is predicated on how you define "intelligence". Personally I subscribe to the view that we should avoid using the word altogether when referring to computers.

Sorry to disappoint, but those questions are bottomless pits of debating semantics.

You personally believe we shouldn’t use the word “intelligence” in these contexts?

Okay — then why even make the comment in the first place?

You clearly meant something by “intelligence != universal approximator”, but you’re unwilling to provide any semantics to back that up.

Go ahead and define it: how do you mean that statement?

I think you’re just too embarrassed to admit you said a personal view that’s wrong.

I'm going to apply the HN rulebook and assume you are in good faith even though I'm almost certain that you are not.

I'm using "intelligence" in the sense of "thing that a layman who doesn't know what compact convergence or lebesgue integrable means would recognize as intelligent".

A layman would not call such an agent, by itself, intelligent.

I'm advocating avoiding the use of "intelligence" precisely to avoid this kind of conversation.

I am replying in good faith.

Can you give an example of a "thing that a layman who doesn't know what compact convergence or lebesgue integrable means would recognize as intelligent" which can’t be modeled by universal approximators? Not just due to scaling.

I think you’re advocating avoiding the use of “intelligence” as a duck of your comment about intelligence and approximators just being wrong.

Would love to see the debate where it convinced 9 people. Only video included in the post is the one where it lost. They really did their best to hype this up. Describing it as “historic” and telling people to clap made me cringe. From my perspective, its strengths were in presenting testimonies. It fell flat when making a joke and when attempting emotional appeals.

I was more impressed with the human on stage, than the AI. Maybe it is because I am getting used to being amazed with things that computers and AI can pull off, but if Harish Natarajan was the one who truly impressed me.

Stay tuned for the first computer virus aimed at hacking its user.

Reply on news.ycombinator.com