Slashdot Mirror
Artificial Intelligence Is Killing the Uncanny Valley and Our Grasp On Reality (wired.com)
rickih02 writes: In 2018, we will enter a new era of machine learning -- one in which AI-generated media looks and sounds completely real. The technologies underlying this shift will push us into new creative realms. But this boom will have a dark side, too. For Backchannel's 2018 predictions edition, Sandra Upson delves into the future of artificial intelligence and the double edged sword its increasing sophistication will present. "A world awash in AI-generated content is a classic case of a utopia that is also a dystopia," she writes. "It's messy, it's beautiful, and it's already here."
"The algorithms powering style transfer are gaining precision, signalling the end of the Uncanny Valley -- the sense of unease that realistic computer-generated humans typically elicit..." the article argues.
"But it's not hard to see how this creative explosion could all go very wrong."
"The algorithms powering style transfer are gaining precision, signalling the end of the Uncanny Valley -- the sense of unease that realistic computer-generated humans typically elicit..." the article argues.
"But it's not hard to see how this creative explosion could all go very wrong."
If you can't tell it's AI generated just train another AI to recognize that it was generated.
Whether the world is doomed or not is immaterial; what matters is whether you will wake up one day to find it doing things you don't like (especially to you).
What if you're working for a big company, and your boss Skypes you from his house, and says it somewhat irregular, but there's an important invoice that needs to be paid right now. He's e-mailing you the invoice right now, and he assures you it's legit and urgent.
Well, it's not your boss but a foreign hacker who used a couple of facebook photos to fake a live conversation.
We live in a world where Warner Bros. just spent millions trying to shop out Superman's mustache to spectacular failure. Disney's Grand Moff Tarkin was even worse. If these titans of the entertainment industry can't pull off a canny reproduction with their hand-crafted flagship products then I really think we'll have to worry about defeating Skynet before we'll have to worry about 'AI' defeating the uncanny valley.
Currently, not a lot of people are used to put doubt into video (or even real-time face-to-face video) because the technology to fake it realistically enough has only started very recently to become cheap enough to be a worthy try for an attacker: And it will still be a little bit more time until it start getting used in real-time (basically once " ${price of renting cloud GPU time to run the neural net} ${money that can be made in such attempts}" ).
Once awareness is raised, society will eventually adapt and only the most gullible will fall for the tricks while our successor on /. will wonder why not more people are using whatever authentication is the most common for video chatting.
A bit like how a couple of decades ago, every body was aware of signature forging and wouldn't trust a simply hand written note, but would fall for attempt at phone-calls social engineering (i.e.: impersonating a general role by being a good actor, back at a time where the phone quality would barely let you recognise a voice reliably).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Or it might be the end of comments sections. Consider this scenario:
Someone develops and publishes a comment-bot AI. It's not a general-purpose AI, but you can configure it with a position to promote and point it to a site, and it will then start posting unique comments promoting the view, and posting rebuttals to anyone who opposes the view. It's not going to pass for human in a conversation, but in single posts it'll appear human most of the time.
First thing that happens? Joe's Pizza unleashes a hundred instances to tell the world how great their pizza is. AI spam. But this is hard spam to get rid of, because it's constantly changing: This AI learns how anti-spam measures work. CAPTCHA tests get even more annoying for a while. But that's ok: The internet is used to spam. Joe's Pizza gets a lot of hate.
Then an election rolls around. Say, a US presidential election.
Suddenly, millions of instances appear - half of them promoting the Republicans, and half the Democrats. Comment threads all over the internet become fifty-pages of almost fully automatically generated text, flooding out any human voice. Both parties deny such underhanded techniques, of course - and perhaps even truthfully, as fingers are pointed to independent pressure groups or the governments of other countries as a possible source.
Meanwhile, the Church of the Easily Offended gets their running. They set a few thousand running - their job is to identify 'inappropriate' material - anything that offends their religion, or standards of decency or of clean language - and submit reports or write angry letters to site operators. In an amusing irony, the church website shortly has to close their own comments section because of the millions of bots now searching the internet for church comments pages and posting about why Islam is the true religion.
In the end the only option is to drop anonymous comments entirely, and tie any comments into verified accounts established with proof of identity.