I know nothing about AI, but the FB case (not having read the link, because we don't do that here) doesn't sound like an incipient overlord, it sounds like machinery which isn't understood by its operators. Which is a problem, but not a civilization-ending one. Admittedly this is different in that machinery is usually understood by its designers; but OTOH farmers and brewers and doctors successfully work daily with biological machinery they don't fully understand, and all of those organisms are more complicated than anything humans have ever made.
AIs improvising their own language is not a new thing. AIs have even developed language without prompting; IIRC one lab a few years ago built a bunch of little learning robots and had them solve mazes, with rewards to the entire group every time one of them managed it, and the robots (which had mikes and speakers) started emitting noises to guide each other, despite not being programmed to do so. Evolutionary hardware design is really, really weird and interesting.
Any thoughts on how long it will be before it will be possible to fake video/audio of a person of your choice saying whatever you want so that it is indistinguishable from the real thing even under sustained forensic investigation? I'm hoping that you guys can come up with some reassuring reason why the answer is "never".
I don't know about "under sustained forensic investigation," but I bet pornography will get really creepy very quickly.
I bet pornography will get really creepy very quickly.
Will?
I should have probably added "in new ways".
started emitting noises to guide each other, despite not being programmed to do so.
Doesn't sound like language so much as sheep bleating at each other so the flock stays together. But fascinating nonetheless.
Doesn't sound like language so much as sheep bleating at each other so the flock stays together.
That's what the sheep want you to think.
I bet pornography will get really creepy very quickly.
I don't know what it says about me that my first thought was 'Trumpenporn'.
I was also deeply freaked out by the first half of the Radiolab episode, where the producer attended a demonstration of what looked like a product that could generate fake audio clips that sound totally real, based on only like 20 minutes of audio input from the person whose voice is getting spoofed. But then on my way home I heard the second half, where they tried to use a similar product themselves and it was exactly as robotic and unconvincing as you'd think it would be.
On the whole, one of their weaker episodes, I thought: sensationalist fear-mongering with no substance to back it up.
I went to a lecture by a media studies professor in about 1994 where he predicated the evolution of software that could convincingly fake news video, and add new audio/video of people's speech. He then talked through the implications of it. Prescient.
The FB thing doesn't seem like language to me.
French link, but trust me, the story in its viral form is bullshit.
I actually don't think the fake video thing matters a damn. Half the world thinks anything they don't like is faked anyway. Kerners are go!
If - if - the software actually appears, the consequence will presumably be that everyone stops trusting what they see and hear all but a small selection of media sources. So, more or less what has already happened. You'll still believe the video you see on the BBC; you won't believe the video you see on Zero Hedge. Just like with print right now.
Come to that, we already have technology that can convincingly do this for audio - comedians do it all the time. Do we have a problem with lots of fake speeches on the radio? No. (We have a slight problem with people in high government office being fooled into thinking they're talking to the Ukrainian prime minister or the foreign minister of Pakistan.)
The real consequence of the faked video thing will be that bad actors will be able to claim fakery even when confronted with actual evidence of malfeasance, or produce fake evidence of them doing bad acts that they're actually doing or about to do, to inoculate themselves against those charges.
Basically, the last 20 years of politics, but increasingly worse.
Pwned by Alex. Or at least I think it was Alex.
The solution to 15 is, presumably, chain of custody - do you trust the person who took the footage or not?
If you need that and you care, we've had the solution for decades - digital signatures. The originator can sign a one-way hash of the file with some public key and anyone wno wants to can verify both its integrity and its authenticity.
The problem is that the world is full of people who have no integrity and couldn't care less about the authenticity of today's helping of outrage product.
18: that's not really practical from a consumer point of view, but yes, you're right. What you end up with is the BBC using digital signatures to verify that the film it shows really was taken by its camera crews, and the audience just deciding to trust what's on the BBC. And a general assumption that an unsigned clip of something happening is no more reliable than a simple assertion by an anonymous person that it happened.
As I say, this is really the process we use for print journalism already. I don't have any way of knowing that, say, FT reporter Alice O'Riordan really saw what she says she saw happening in Caracas. I trust her report because it's on the FT which has a good reputation. If I read the same report on a blogspot site with the words "by Alice O'Riordan of the FT" at the bottom, I wouldn't trust it.
The catch is that video is more visceral than print; choosing to disbelieve your eyes is harder than choosing to disbelieve words you've read, even if you know they're equally (un)trustworthy.
20: it is now, sure. But it's getting less so all the time as CGI gets better.
Look at how cinema's changed. The mass battle scenes in things like "Spartacus" were filmed by actually getting a load of guys in armour to pretend to fight. That was the only way to do it in 1960. And that kind of spectacle was a big seller for the epics of the day. Not any more; we still love spectacle, but it has to have something more than just sheer scale, because we all know that the difference between a battle with 20 orcs and one with 20,000,000 orcs is just a few Ctrl-C/Ctrl-V moves (I exaggerate slightly).
Once we've all seen the products of this kind of fake-video technology it'll lose a lot of its impact.
I have no idea what you mean.
"A CGI orc dies a thousand times before his death, but the live-action extra tastes of death but once."
Can I second the claim that the viral form of the FB story is utter bullshit, and was the first time it appeared, around a fortnight ago. The voice demo stuff impressed me. Not the Adobe thinghere, but some Canadian startup I can't track through my bookmarks. Maybethese
some Canadian startup
Oh good, we could turn the conversation towards the much more soothing topic of how unaffordable Canadian cities will become as the U.S. tech industry moves north. Maybe the robot overlords will take over the housing market. I'm not sure I'm using anything like human intelligence to type this comment.
Spoiler for 12: It turns out the AI wasn't speaking a new language, it was just speaking French the whole time.
We have a slight problem with people in high government office being fooled into thinking they're talking to the Ukrainian prime minister or the foreign minister of Pakistan.
Or each other, apparently. (Thought admittedly that was in written form.)
These Norwegian racists didn't even need fake photos - a real photo of a bus was enough to send them into a fit of stupidity:
Reminds me of that Brass Eye bit about the dizzying rate of crime we know nothing about. Muslims that don't exist are even scarier than the ones that do!
Chinese chatbot voices anti-Communist sentiment, is promptly disappeared...
https://www.ft.com/content/e90a6c1c-7764-11e7-a3e8-60495fe6ca71
Norwegians have nicer bus seats than we do. This doesn't surprise me in the least.
Trump is going to sue colleges for affirmative action in admissions because he's done nothing so effectively as show that white men have serious intellectual failings.
I'm guessing that the end result of Trump's election and the subsequent cascade of shit running down America's legs will be something where a subset of white people (marked by location and education) becomes a "minority" group with set-aside quotas and the like. I honestly don't know if me-type white people will be the numerical minority or the Trumpists, but they'll get the quotas and set-aside dollars. It will probably take a couple of generations, but social mobility across regular white people and huge-asshole white people seems to really be dropping. There's always been pockets of greatly disadvantaged white people, of course, but they were usually isolated from each other and they also stopped thinking of themselves as disadvantaged once they stopped being disadvantaged.
(In my mind, I'm still on topic.)