unbibium: (kuribo)
[personal profile] unbibium
Today, there are powerful AI models that you can download and run on a home computer. Some people think that sufficiently powerful AI models are "sentient" or have wills of their own and we'll have to give them rights... so, what does that imply?

25 years ago, when the Sims came out, I read about people who'd torture their Sims by walling them in or removing the exit ladder from a pool or similar. That makes me wonder if, today, there are sadistic teenagers out there downloading image recognition models and feeding it gross-out images, thinking they're performing torture on a helpless digital animal. I wonder if they'll feel guilty about it later, or whether it'll inspire them to go out and torture non-digital entities.

This reminds me of the people who learn to lucid dream, but call it "reality shifting" as if they're actually traveling to parallel universes. I read a post from someone who committed an assault in their dream, and instead of interpreting it as a nightmare that they had to shake off, they thought it was an actual crime they committed against a real person somewhere in the multiverse, for which they had to atone.

We do have to be careful about how our beliefs shape our behavior, because the world gets more challenging every day.

Date: 2025-02-02 12:05 am (UTC)
mmcirvin: (Default)
From: [personal profile] mmcirvin
The Sims was a spinoff of SimCity, which in turn got inspiration from a cautionary fable by Stanisław Lem about Trurl the constructor giving an exiled tyrant a tiny simulated kingdom to play with, whereupon his sage friend Klapaucius explains that Trurl's perfection is such that he can't say he hasn't given rise to actual suffering beings.

I suspect that LLMs are not in any useful sense sentient in part because the world of "experience" they're built on is just a corpus of text rather than any sort of effort to live in a world. But I also think that we could eventually get into a gray area where it's really impossible to tell, and I will also observe that the moneybags techbro types trying to create "AGI" seem to have zero compunctions about the possibility of creating sentient slaves, whether they can do it or not.

And it's also true that the greatest damage done by these systems today seems to lie in the disconnect between what people think they can do, and what they actually can do. Humans actually seem to be dismayingly easy to fool in this regard because we are instinctively social beings and pick up easily on things that seem to be social cues.

Profile

unbibium: (Default)
unbibium

July 2025

S M T W T F S
  12345
6789101112
13 141516171819
20212223242526
2728293031  

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags
Page generated Jul. 23rd, 2025 06:20 pm
Powered by Dreamwidth Studios