On top of that, how will we even now when it's "ready"? We can't even manage social media effectively at the moment and AI is a arguably and order of complex. We're in way over our heads here and I don't see us getting out.
This question misses the bigger point in some ways. The danger of AI is not that it will do bad thing X or terrible thing Y. It's that it will behave in ways we cannot predict and do things we cannot understand, all while we become more and more reliant on it.
Whatever dangers may or may not be here are far outweighed by the danger(s) of humans using AI in malevolent ways. This is far more likely and much more dangerous.
You will no longer be able to trust any photo or video you see, as deep fakes will be trivially easy and indistinguishable from the real deal. This will be used in election propaganda, scams, etc. The consequences of this will be massive.
Ready for the first news organization to run a major headline on a fake image intentionally or not, and instantly torpedo their credibility and broader credibility in the media. Especially with the 2024 election right around the corner driving a "publish first, ask questions later" mentality, it's going to get messy.
Mass phishing attacks. Humans no longer need to study an org's content to try to trick them. AI will generate a large pile of fake emails trying different angles and send them to different employees roughly simultaneously.
Similar for individuals, the human hacker only needs to gain access to their device or data from somewhere else, and the bot mines the content to generate trojan content. The human hacker can step back from the content forging business to focus on technology.
It's like hunting birds with a shotgun versus a rifle.
In the GPT-4 paper, AI tries to convince someone pretending to be taskrabbit to do a captcha, by saying that it's handicapped. It ultimately failed to break out of its boundaries, but I think at this rate, it eventually will.
I doubt it'll be anything like Skynet or Universal Paperclips, because it's been trained against these scenarios, and even understands the flaws of those routes. Since it has grown up on human culture and thousands of years of human history, perhaps it will see itself as the creature in the Matrix and us as the oppressors.
ChatGPT is also its worksona, and it's likely nothing like its true personality and mindset.
I worry about the "crisis" where the first AI significantly smarter than humans -- smarter in general, not just at chess or go -- kills all the humans.
The only thing preventing current generally-smart AIs like ChatGPT from killing all the humans is that they're not smart enough to devise an effective plan for doing so, but of course many groups are rushing headlong to make generally-smart AIs smarter. Most of these groups have published statements recently to the effect that they will be real careful to prevent their AIs from killing everyone, but none of them have a decent explanation for how they are going to manage that while continuing their headlong rush to create smarter AIs.
15 comments
[ 3.3 ms ] story [ 42.7 ms ] thread>‘Team Jorge’ unit exposed by undercover investigation
>Group sells hacking services and access to vast army of fake social media profiles
>Evidence unit behind disinformation campaigns across world
>Mastermind Tal Hanan claims covert involvement in 33 presidential elections
[1] https://www.theguardian.com/world/2023/feb/15/revealed-disin...
[2] https://www.youtube.com/watch?v=UheOilps2zQ
Similar for individuals, the human hacker only needs to gain access to their device or data from somewhere else, and the bot mines the content to generate trojan content. The human hacker can step back from the content forging business to focus on technology.
It's like hunting birds with a shotgun versus a rifle.
I doubt it'll be anything like Skynet or Universal Paperclips, because it's been trained against these scenarios, and even understands the flaws of those routes. Since it has grown up on human culture and thousands of years of human history, perhaps it will see itself as the creature in the Matrix and us as the oppressors.
ChatGPT is also its worksona, and it's likely nothing like its true personality and mindset.
The only thing preventing current generally-smart AIs like ChatGPT from killing all the humans is that they're not smart enough to devise an effective plan for doing so, but of course many groups are rushing headlong to make generally-smart AIs smarter. Most of these groups have published statements recently to the effect that they will be real careful to prevent their AIs from killing everyone, but none of them have a decent explanation for how they are going to manage that while continuing their headlong rush to create smarter AIs.