Ray Kurzweil and Mitch Kapor’s “Long Bet” on the Turing Test | Hacker News Reader

Ray Kurzweil and Mitch Kapor’s “Long Bet” on the Turing Test

twitter.com

21 pointsaabhay17 comments

> This is often overlooked by those computer scientists who correctly point out that it is not impossible for computers to demonstrate creativity. Not impossible, yes. Likely enough to warrant belief in a computer can pass the Turing Test? In my opinion, no.

I never understand this creativity argument against AI. How is creativity different than adding some randomness to the answer generating? I would say that computers are perfectly capable of that.

It’s interesting. Creativity might not be different in effect to “adding some randomness”, but it is different in cause. If the effect is indistinguishable, this distinction may seem sentimental. But take the idea to its limit. If an A.I. told you that it loved you, that would be practically different as a mere effect of added randomness than as an expression of a cause. A piece of text stripped of human context is no different than one made by a bona-fide human. But language without human context, would not be human. Two people saying the exact same thing may have entirely different meanings. Etc. And of course, this is still extremely abstract. People aren’t “contexts”.

> How is creativity different than adding some randomness to the answer generating?

Well, that puts you pretty deep into questions of the philosophy of art, the inner workings of AI, and what the person was predicting about those things when they wrote it.

If I train an art-generating ML model exclusively on pictures of cats and dogs, we might expect it to only be able to generate pictures of cats, dogs, and perhaps cat-dog hybrids.

In a sense it's able to create things outside of the bounds of the training set, as there were no cat-dog hybrids in the training set. But in another sense, it's unable to create things outside those bounds, as it can't create pictures of horses or astronauts.

Some would say the latter is needed for true creativity, boundless possibilities being a defining property of art.

In the case of current image generating ML models, their training set wasn't limited to cats and dogs - but it was limited to the art you could scrape from the web in ~2022.

And of course Kapoor was writing this in 2002, so was presumably judging the artfulness of a hypothetical image generation system.

> How is creativity different than adding some randomness to the answer generating?

Deterministic systems are clearly not creative. But randomness is a necessary, but not sufficient, requirement for creativity.

Adding random noise to an image makes it grainy (like TV static from back in the day), it doesn't lead to new masterpieces.

There's a difference between creativity and initiative. Aleatoric music has been a thing for over 200 years. But you still need a human to roll the dice, as it were. I would like to see a computer do something unprompted before I am sold on AI being creative

I'll save you a click

https://longbets.org/1/

Kurzweil argues that in 2029, a computer will pass the Turing test. His arguments:

Hardware: Moore's law. It might come to three-dimensional computing.

Software: Interestingly, he sees this as more difficult.

> It is correct to point out that achieving the "software" of human intelligence is the more salient, and more difficult, challenge.

Currently, if a LLM will pass the Turing test, it would actually be the case that the software is actually very simple.

Kapor's argument against that are basically that he thinks it's just too difficult to imagine:

> it becomes extremely difficult even to imagine what it would mean for a computer to perform a successful impersonation, much less to believe that its achievement is within our lifespan.

This will be interesting. AI's excellent grammar disarms people into believing the AI has common sense in novel situations when, in fact, it doesn't. Think of it like this: a photo of a scene can fool people into thinking it's an actual scene up until the actual scene rotates a bit but the picture doesn't follow. Interesting times for the Turing Test.

The real question is whether many humans will pass it.

Indeed

What's the current state of Turing test?

People who have been too lazy and uncaring to leave may not realize that Xitter no longer allows you to read threads if you’re not logged in. So if there’s interesting content here, it’s effectively walled off for an increasing number of people. Some of us no longer even have an account there.

Mastodon and Threads allow you to read threads without logging in. I’m sure Bluesky does too, not that it matters.

https://nitter.net/denizen_kane/status/1709800563485946156

Just change twitter.com to nitter.net. Or nitter.poast.org

Good content is rare to find on Twitter these days, so not much is lost from not being able to link to it anymore.

It's a shame when content is posted to hn that I can't read without some sort of sign up. It's rare that I'm prepared to do that for some random story on an aggregator.

There doesn't seem to be anything in the guidelines about avoiding pay walled or sign in walled content in submissions, but it does make hn less valuable as a resource when a high proportion of links require more investment than is reasonable to read.

Bluesky doesn't even allow you to read posts until you log in, but it does leak enough in the meta tags when it's posted to, say, discord.

Is the X pronounced like "sh", "z", "ks", or "eks" ?

Reply on news.ycombinator.com