A world where you never know if you're talking to another human being unless you're right in front of them.
You wouldn't be able to trust that it's your own mother on the phone. Every thread and chat may just be you talking to yourself through a pile of code. And god knows who that code's talking to. Photo, video evidence in court? Can't prove it's not real, your honor.
An artificial world. Everyone sequestered away into their own bespoke realities, separated by unbreakable walls of uncertainty.
AI images are even worse than that. They still struggle with stuff like hands.
I've played a little around with StableDiffusion and I must admit that it is great fun and can give some cool results but I don't think it's going to challenge our perception of reality anytime soon though.
You can get AI to generate an image of a conventionally beautiful 20-something woman sitting in one of a handful of seductive poses relatively easy. If you're lucky the AI won't even mangle her hands. But the moment you move away from that, quality rapidly drops. Making the AI do old, fat or ugly people is hard and getting it to do any type of complex scene involving multiple people or objects is near-impossible.
It'll be a long while before you can put "realistic footage of Jeremy Corbyn kissing Hitler" into the AI and get a believable output.
Define a long time. Cause in the scheme of things we went from Twitter bots to mostly passing the Turing test in half a dozen years.
Yeah an AI that can barely make an image of an attractive woman is still an insane leap compared to a decade ago
The rudimentary twitter bots you're imagining were the forefront of tech over 60 years ago, not 6. 'Eliza' was an NLP bot that was developed on computers in the 1960s that was more advanced than most of those twitter bots, back then people were also asking just how many more days until the robo-apocalypse. This is an insanely complex field, and we've made an advancement and it looks very pretty, but it still has fundamental flaws.
So, we've gone from that to this AI in 60 years. Thus, if we're comparing those leaps as roughly equivalent, by a long time, we could easily be looking at another 50 years.
Also Eliza for the curious: https://www.masswerk.at/elizabot/
Precisely. There are orders of magnitude difference in difficulty between "making very standard, vague stuff that is 90% believable" to "making a wide range of things that can actually be convincing".
I think it's just the nature of the beast, a lot of AI generated stuff seems to work with a 'platonic ideal' if you want something that outputs believable stuff that ideal needs to be enforced quite strongly and that'll then override what makes things special that's why when I put in three voice samples it output the right pitch but a completely wrong 'dialect/accent' i.e. it enforced the 'platonic ideal' it had of what language is, so far the believeable stuff I've heard is all voice actors who due to the nature of their work have to have a standard accent, emulating a mom that maybe says a word weird or who has a strong xyz accent is pretty far away and might never happen unless someone spends hours to imitate a specific speech pattern.
have you ever listened to chuds talking for 5 minutes?