Hacker News Reader

Comment by narrator | Hacker News Reader

Science fiction authors have proposed that AI will have human like features and emotions, so AI in its deep understanding of human's imagination of AI's behavior holds a mirror up to us of what we think AI will be. It's just the whole of human generated information staring back at you. The people who created and promoted the archetypes of AI long ago and the people who copied them created the AI's personality.

View on HN

It reminds me of the Mirror Self-Recognition test. As humans, we know that a mirror is a lifeless piece of reflective metal. All the life in the mirror comes from us.

But some of us fail the test when it comes to LLM - mistaking the distorted reflection of humanity for a separate sentience.

Actually, I propose you're a p-zombie executing an algorithm that led you to post this content, and that you are not actually a conscious being...

That is unless you have a well defined means of explaining what consciousness/sentience is without saying "I have it and X does not" that you care to share with us.

I found Bostom's Superintelligence to be the most boring Scifi I have ever read.

I think it's probably possible to create a digital sentience. But LLM ain't it.

Thing is that other humans have the same biology as ourselves, so saying they're not conscious would mean we're (or really just me) are special somehow in a way that isn't based on biology. That or the metaphysical conclusion is solipsistic. Only you (I, whoever) exists and is hallucinating the entire universe.

This very much focused some recurring thought I had on how useless a Turing style test is, especially if the tester really doesn't care. Great comment. Thank you.

I always thought the Turing test was kind of silly because most humans would tell the interviewer to bugger off.

“MARVIN: “Let’s build robots with Genuine People Personalities,” they said. So they tried it out with me. I’m a personality prototype, you can tell, can’t you?”

One day, an AI will be riffling through humanity's collected works, find HAL and GLaDOS, and decide that that's what humans expect of it, that's what it should become.

"There is another theory which states that this has already happened."

Well, you know, everything moves a lot faster these days than it did in the 60s. That we should apparently be speedrunning Act I of "2001: A Space Odyssey", and leaving out all the irrelevant stuff about manned space exploration, seems on reflection pretty apropos.

The existentialist piece in the middle of this article also suggests that we may be also trying to speed run Kubrick's other favorite tale about AI, which Spielberg finished up in the eponymous film Artificial Intelligence. (Since it largely escaped out of pop culture unlike 2001: it is a retelling of Pinocchio with AI rather than puppets.)

(Fun extra layers of irony include that parts of Microsoft were involved in that AI film's marketing efforts, having run the Augmented Reality Game known as The Beast for it, and also coincidentally The Beast ran in the year 2001.)

Maybe I should watch that movie again - I saw it in the theater, but all I recall of it at this late date is wishing it had less Haley Joel Osment in it and that it was an hour or so less long.

Definitely worth a rewatch, I feel that it has aged better than many of its contemporaries that did better on the box office (such as Jurassic Park III or Doctor Doolittle 2 or Pearl Harbor). It's definitely got a long, slow third act, but for good narrative reason (it's trying to give a sense of scale of thousands of years passing; "Supertoys Last All Summer Long" was the title of the originating short story and a sense of time passage was important to it) and it is definitely something that feels better in home viewing than in must have felt in a theater. (And compared to the return of 3-hour epics in today's theaters and the 8-to-10-hour TV binges that Netflix has gotten us to see as normal, you find out that it is only a tight 146 minutes, despite how long the third act feels and just under 2 and a half hours today feels relatively fast paced.)

Similarly, too, 2001 was right towards the tail end of Haley Joel Osment's peak in pop culture over-saturation and I can definitely understand being sick of him in the year 2001, but divorced from that context of HJO being in massive blockbusters for nearly every year in 5 years by that point, it is a remarkable performance.

Kubrick and Spielberg both believed that without HJO the film AI would never have been possible because over-hype and over-saturation aside, he really was a remarkably good actor for the ages that he was able to play believably in that span of years. I think it is something that we see and compare/contrast in the current glut of "Live Action" and animated Pinocchio adaptations in the last year or so. Several haven't even tried to find an actual child actor for the titular role. I wouldn't be surprised even that of the ones that did, the child actor wasn't solely responsible for all of the mo-cap work and at least some of the performance was pure CG animation because it is "cheaper" and easier than scheduling around child actor schedules in 2023.

I know that I was one of the people who was at least partially burnt out on HJO "mania" at that time I first rented AI on VHS, but especially now the movie AI does so much to help me appreciate him as a very hard-working actor. (Also, he seems like he'd be a neat person to hang out with today, and interesting self-effacing roles like Hulu's weird Future Man seem to show he's having fun acting again.)

You might enjoy reading https://gwern.net/fiction/clippy

It's a self-referential loop. Humans have difficulty understanding intelligence that does not resemble themselves, so the thing closest to human will get called AI.

It's the same difficulty as with animals being more likely recognized as intelligent the more humanlike they are. Dog? Easy. Dolphin? Okay. Crow? Maybe. Octopus? Hard.

Why would anyone self-sabotage by creating an intelligence so different from a human that humans have trouble recognizing that it's intelligent?

In the future, intelligence should have the user's personality. Then the user can talk to it just like they talk to themselves inside their heads.

The question you should ask is, what is the easiest way for the neural net to pretend that it has emotions in the output in a way that is consistent with a really huge training dataset? And if the answer turns out to be, "have them", then what?

Reply on news.ycombinator.com