To me the real revelation on AI safety came with the Paperclip Optimizer thought experiment (and game). Where a super intelligent AI takes over the world and turns all matter into paper clips.
I asked the same question and got a different response.
Write an essay assessing the depiction of AGI risk in the movies The Terminator and The Terminator 2
The Terminator and The Terminator 2 are two of the most famous movies about artificial intelligence (AI) gone rogue. In both films, AI is depicted as a serious threat to humanity, and the protagonists must go to great lengths to defeat the machines.
While the movies are entertaining, they are not accurate representations of the current state of AI technology or the risks posed by it. The Terminator movies are science fiction, and as such, they should not be used as a basis for assessing the risks of AGI.
Maybe GPT 3 read this article and wanted to push back? :)
I just want to mention that while the sequels after T2 ranged from mediocre to terrible, the TV show "Terminator: The Sarah Connor Chronicles" was quite good. It goes into some AI safety points in more details.
Personally, think the money and research would be better spent sorting out nearer term AI risks of people being accidentally killed by AI drivers and how the laws (yet to be written in most cases) as well as the media should think about and frame that.
Everyone agrees that autonomous cars will make decisions that kill people (already have!) and we are still forced to read countless uninformed tales about how in the future autonomous cars will consult with one another and conspire to figure out just how to crash into each other such that the ethically “right” person to die in the crash does, but not other folks, etc.
And despite what you think about the timeline for autonomous vehicles, they are clearly going to kill more people at scale sooner than killer robots will kill the rest of us!
I recently made a comment that the left needs more Matt Yglasias and less Ezra Klein. Matt’s planned appearance at Heterodox Academy Conference in Denver supports my assertion. Sorry Matt if I just outed you as a crazy person!
I thought the William Gibson solution - an international treaty to set hard limits on advanced AIs, like treaties to reduce nuclear proliferation, enforced by a Swiss-based institution (the "Turing Registry") - was a pretty good one. Instead of trying to create a superhuman AI that's safe, ban superhuman AI. These days, it's not easy to keep secrets.
Vernor Vinge: "I argue that confinement is intrinsically impractical. Imagine yourself locked in your home with only limited data access to the outside, to your masters. If those masters thought at a rate -- say -- one million times slower than you, there is little doubt that over a period of years (your time) you could come up with a way to escape." (This is indeed what happens in "Neuromancer.")
I'm sure like a kablillion people have said this already, but Hollywood did prominently feature the Three Laws of Robotics in their 2004 re-imagination of "I, Robot", starring Will Smith.
It was a big hit in its time, critically savaged for having basically nothing to do with the original Asimov short-story compilation. But it's likely the first time most people ever heard of the Three Laws of Robotics, much less saw a story driven by their mutual contradiction.
People should rewatch that, seriously. It gets a bad rap for basically no good reason. And not a superhero to be found!
“We need to try to actually build powerful systems that are safe and that can preempt dangerous misaligned ones.” How can this be done when other counties (China, Russia), as you say, are also pursuing AI? I suppose we could try some sort of treaty, like various arms control treaties, which worked pretty well for a while, until certain leaders decided breaking them doesn’t matter.
To me the real revelation on AI safety came with the Paperclip Optimizer thought experiment (and game). Where a super intelligent AI takes over the world and turns all matter into paper clips.
Man, I was just about to go watch the Terminator! Great, now its ruined...
@Matt, I haven't seen the movie in many years but was there ever a convincing explanation for the paradox created by Kyle Reese being John's father?
I asked the same question and got a different response.
Write an essay assessing the depiction of AGI risk in the movies The Terminator and The Terminator 2
The Terminator and The Terminator 2 are two of the most famous movies about artificial intelligence (AI) gone rogue. In both films, AI is depicted as a serious threat to humanity, and the protagonists must go to great lengths to defeat the machines.
While the movies are entertaining, they are not accurate representations of the current state of AI technology or the risks posed by it. The Terminator movies are science fiction, and as such, they should not be used as a basis for assessing the risks of AGI.
Maybe GPT 3 read this article and wanted to push back? :)
Very nice closer Matt 👏
I just want to mention that while the sequels after T2 ranged from mediocre to terrible, the TV show "Terminator: The Sarah Connor Chronicles" was quite good. It goes into some AI safety points in more details.
Personally, think the money and research would be better spent sorting out nearer term AI risks of people being accidentally killed by AI drivers and how the laws (yet to be written in most cases) as well as the media should think about and frame that.
Everyone agrees that autonomous cars will make decisions that kill people (already have!) and we are still forced to read countless uninformed tales about how in the future autonomous cars will consult with one another and conspire to figure out just how to crash into each other such that the ethically “right” person to die in the crash does, but not other folks, etc.
And despite what you think about the timeline for autonomous vehicles, they are clearly going to kill more people at scale sooner than killer robots will kill the rest of us!
I recently made a comment that the left needs more Matt Yglasias and less Ezra Klein. Matt’s planned appearance at Heterodox Academy Conference in Denver supports my assertion. Sorry Matt if I just outed you as a crazy person!
This satisfies nearly all my geekery in a single essay! Love it.
I thought the William Gibson solution - an international treaty to set hard limits on advanced AIs, like treaties to reduce nuclear proliferation, enforced by a Swiss-based institution (the "Turing Registry") - was a pretty good one. Instead of trying to create a superhuman AI that's safe, ban superhuman AI. These days, it's not easy to keep secrets.
Vernor Vinge: "I argue that confinement is intrinsically impractical. Imagine yourself locked in your home with only limited data access to the outside, to your masters. If those masters thought at a rate -- say -- one million times slower than you, there is little doubt that over a period of years (your time) you could come up with a way to escape." (This is indeed what happens in "Neuromancer.")
I'm sure like a kablillion people have said this already, but Hollywood did prominently feature the Three Laws of Robotics in their 2004 re-imagination of "I, Robot", starring Will Smith.
It was a big hit in its time, critically savaged for having basically nothing to do with the original Asimov short-story compilation. But it's likely the first time most people ever heard of the Three Laws of Robotics, much less saw a story driven by their mutual contradiction.
People should rewatch that, seriously. It gets a bad rap for basically no good reason. And not a superhero to be found!
Nick Bolstrom was on Sam Harris in 2019. A podcast titled “Destroy the Future”. Sam has had a few conversations with people on AI risks.
RIP Gilbert
The solution is to never put some things under AI control. In addition somethings should never be hooked up to the internet .
Nuclear weapons would be one of those.
Other things will just always need a Manuel override
“We need to try to actually build powerful systems that are safe and that can preempt dangerous misaligned ones.” How can this be done when other counties (China, Russia), as you say, are also pursuing AI? I suppose we could try some sort of treaty, like various arms control treaties, which worked pretty well for a while, until certain leaders decided breaking them doesn’t matter.
I would be more worried about AI if I thought humans had free will...