Microsoft President Says Deep Fakes Are Biggest AI Concern (reuters.com) 52
Microsoft President Brad Smith said Thursday that his biggest concern around artificial intelligence was deep fakes, realistic looking but false content. From a report: In a speech in Washington aimed at addressing the issue of how best to regulate AI, which went from wonky to widespread with the arrival of OpenAI's ChatGPT, Smith called for steps to ensure that people know when a photo or video is real and when it is generated by AI, potentially for nefarious purposes. "We're going have to address the issues around deep fakes. We're going to have to address in particular what we worry about most foreign cyber influence operations, the kinds of activities that are already taking place by the Russian government, the Chinese, the Iranians," he said.
"We need to take steps to protect against the alteration of legitimate content with an intent to deceive or defraud people through the use of AI." Smith also called for licensing for the most critical forms of AI with "obligations to protect security, physical security, cybersecurity, national security." "We will need a new generation of export controls, at least the evolution of the export controls we have, to ensure that these models are not stolen or not used in ways that would violate the country's export control requirements," he said. For weeks, lawmakers in Washington have struggled with what laws to pass to control AI even as companies large and small have raced to bring increasingly versatile AI to market.
"We need to take steps to protect against the alteration of legitimate content with an intent to deceive or defraud people through the use of AI." Smith also called for licensing for the most critical forms of AI with "obligations to protect security, physical security, cybersecurity, national security." "We will need a new generation of export controls, at least the evolution of the export controls we have, to ensure that these models are not stolen or not used in ways that would violate the country's export control requirements," he said. For weeks, lawmakers in Washington have struggled with what laws to pass to control AI even as companies large and small have raced to bring increasingly versatile AI to market.
Code generated by GPT is worse (Score:5, Informative)
The neural network is created to make the text resemble the text you asked it for as much as possible, rather of having any sort of logic behind it.
So it's failure mode is to write a code that looks as close as possible as a working code, but still incorrect or unsafe.
Re:Code generated by GPT is worse (Score:4, Interesting)
Indeed. I expect criminal enterprises will start seeding subtly broken code so the next "AI" will learn to do it like that. Then, a few years later look for these vulnerabilities and there will be no way to fix them fast. Even without that, code generated now will have systematic problems and these may well result in systematic vulnerabilities as well that can then be found in a large number of entirely different software.
Re: (Score:2)
Hush! I'm in IT security, this is perfect job security for me for years to come!
Just Another Day (Score:2)
Why worry about deep fakes when real and corrupt politicians, scientists and businessmen will spread lies first hand.
Re: (Score:3, Insightful)
They could use GPT to automate the whole corruption and lying process
Train it to be as effective as possible in making a lie pass up as truth, or convincing someone to do the dirty deed you want for the least amount of bribe money
Re: (Score:2)
We can just look at the hands to see if it's fake or not.
Speaking of which: Don't Donald Trumps hands look a bit too small? Maybe he's been a deepfake all along.
Re: (Score:2)
There's no reason for the person that is only there to pose as some sort of king so the population can target someone instead of actually getting aware of the system should be real.
Re: (Score:2)
Re: (Score:2)
We can just look at the hands to see if it's fake or not.
You're out of the loop. Or trolling. Either way, wrong.
Re: (Score:2)
You are missing the point i guess. Now, politicians have the perfect excuse and explanation for anything you catch them doing even if you have proof such as pictures or video of them doing it; Deep fakes, the damn AI got so good, you can't even tell for sure it's a deep fake but I swear it is a deep fake! Not guilty, case closed, it was just another deep fake!
People will get so used to deep fakes that they won't believe anything they see. Everybody will become like people who say images from the Moon and th
Re: (Score:3, Insightful)
> You are missing the point i guess. Now, politicians have the perfect excuse and explanation for anything you catch them doing even if you have proof such as pictures or video of them doing it;
Trudeau has been caught in scandal after scandal. He has political connections that allow him to appoint family friends to head investigations against him OR coalition agreements to push laws through. What could someone possibly fake? It's not like fake videos of him doing blackface or lying would do anything.
Re: (Score:2)
Re: (Score:1)
> You don't have to name a specific politician.
I can name who I like.
> Your side isn't better than the other side
That's where you're wrong. Liberal, Bloc Québécois, Conservative, NDP, Green, PPC... I could randomly pick any side that isn't Liberal and it would be the better choice. Justin Trudeau is running the country into the ground.
Re: (Score:2)
Oh boy! Wake up! Those black faces pictures of Trudeau were obviously deep fakes for sure...
Chinese influence: deep fakes!
Aga Kan's island pictures: deep fakes!
etc. etc.
Re: (Score:2)
People will get so used to deep fakes that they won't believe anything they see. Everybody will become like people who say images from the Moon and the space station are deep fakes.
"Believe only half of what you see and nothing that you hear."
-- Edgar Allen Poe
If the majority become sceptical of online sources would save the world a lot of crap.
Personal experience counts for more. Thats a GOOD THING.
Politics should be personal, should be local and verifiable.
Doubt should be in everyones minds all the time.
Thats how science works. Fkin doubt it, fkin check up on it, don't believe the hype.
I can't see anything but good coming from people stopping believing the shit they see on the medi
Re: (Score:2)
Thats how science works. Fkin doubt it, fkin check up on it, don't believe the hype.
I can't see anything but good coming from people stopping believing the shit they see on the media and internet.
Because sometimes the shit they see is science, and they choose to ignore it because it doesn't fit with their worldview.
Barrier to entry (Score:5, Insightful)
Why worry about deep fakes when real and corrupt politicians, scientists and businessmen will spread lies first hand.
Because with deep fakes the barrier to entry is much lower.
Videos are currently considered more reliable than images, which are more reliable than 1st hand reports, which are more reliable than standard news reporting, which are more reliable than news reporting from unnamed sources, which are more reliable than opinions masquerading as news.
The reason videos are thought reliable is that fake videos are hard to produce: it takes a lot of time and/or effort and if your time would otherwise be spent making an income, fake videos cost money as well. You might consider, depending on the depth of fakeness, to write up some dialogue, make storyboards, hire some actors, rent a location (or go to a remote location), makeup the actors, and so on and so on. CGI is generally hard, maker-intensive, costs a lot of money in terms of energy and compute time, and can often be easily seen as fake.
A recent slashdot article [slashdot.org] told of a pizza commercial made by AI. I don't know what the workflow was, but I expect that the user asked ChatGPT to write up a script, then used the script with StableDiffusion to make storyboard images, then used some AI video process to turn the storyboard images into video clips. Then the AI sound program put a voice to the script.
The article was posted as a way to laugh at how completely insane and creepy the finished product was, but buried in the actual article was the comment "Definitely wasted 3 hours of my life making this today... Everything is AI from the VO to the video and images. Assembled in After Effects."
Three hours to make a full-blown 30-sec pizza commercial is, quite frankly, astonishing.
One could imagine a different person using ChatGPT to write up a script, mark up the sections viewed as "poor", and asking ChatGPT to rewrite those sections, and repeat until the entire script was viewed as high quality. Then do the same with the other AI tasks in the workflow. Instead of 3 hours, maybe take three 8-hour days to make a more presentable, less easily debunked video.
You can do this yourself - I've done it with poetry. Ask ChatGPT to write a poem in the style of someone, then review the poem, and selectively tell it to rewrite paragraphs or lines that you think are sub-optimal. You'll very quickly narrow down to something that you feel is really good.
There's a presidential election coming up in about a year, and we will probably have so many deep fakes that it will be impossible to tell what is real any more.
(Also of note: the leak of LLaMA in the beginning of March led to a bevy of breakthrough improvements over the succeeding 2 months. We can expect that AI will further improve at the same breakneck pace for the foreseeable future.)
Re: (Score:2)
Three hours to make a full-blown 30-sec pizza commercial is, quite frankly, astonishing.
That commercial was only full-blown in the sense that it had AIDS.
Seriously though, I do get what you're saying, and it is actually impressive. However
You can do this yourself - I've done it with poetry. Ask ChatGPT to write a poem in the style of someone, then review the poem, and selectively tell it to rewrite paragraphs or lines that you think are sub-optimal. You'll very quickly narrow down to something that you feel is really good.
Nah. I tried that, I wanted a poem about the Cummins in the style of ee cummings. It can't even give me the number of lines I ask for.
Re: (Score:1)
The Durham report(and the media's refusal to acknowledge their complicity) has already shown you don't need deep fakes if a there's a group willing to curate democracy in order to save it.
Re: (Score:2)
Re: (Score:2)
> "The Media" isn't homogenous.
Multiple local news stations say the same thing verbatim
https://www.youtube.com/watch?... [youtube.com]
Inconsistant (Score:2)
his biggest concern around artificial intelligence was deep fakes, realistic looking but false content.
Is it really though, because I just watched a video with Brad Smith proclaiming how awesome AI was, and that we should all embrace our new glorious AI overlords while stripping to reveal a large "I HEART AI" tattoo on his chest and rear.
Make up your mind Brad!
I agree (Score:4, Funny)
Re: (Score:2)
Could've been worse, the real Ron is a Deep Jerk.
Mr. Smith makes a good point (Score:2)
Re: (Score:1)
Back in my day, all the crazy people either hung out with or "were" Napoleon. Glad to see AI is expanding their horizon; they get new "friends".
Tough luck for politicians and celebs. You mad bro (Score:2)
Re: (Score:2)
Re: (Score:1)
Re: Tough luck for politicians and celebs. You mad (Score:2)
Yeah. I have been shouting at everyone that deepfakes are just Photoshop for video, but nobody seems to understand. Fuck them. Idiots. Soon, somebody will build a bot which takes whatever video is trending and generate and post a deepfake with inverse meaning. 98% of the country will be baffled as to why the president is flip-flopping on every issue and every sports team wins every game.
Re: (Score:1)
98% of the country will be baffled as to why the president is flip-flopping on every issue and every sports team wins every game.
Disagree, it'll be closer to 99%, but your reply made me laugh. Thanks!
Microsoft should know (Score:3)
Nake Fews! (Score:1)
That wasn't Brad Smith, that was a deep fake.
Microsoft President is an idiot (Score:5, Insightful)
Yes, deepfakes are a serious concern. No, they are by far not the biggest one.
Re: (Score:2)
Re: (Score:2)
He said "the biggest". Maybe learn to read?
Re: (Score:2)
What do you feel is the "biggest" concern?
Re: (Score:2)
Not that one.
Re: (Score:2)
In other words, you can't actually name a single thing that's a bigger concern. You just wanted to call him an idiot. Got it.
Re: (Score:2)
Nope, I just do not care to. Because you are obviously scum that argues in bad faith. Why would I feed a troll like you? Incidentally the question was _directly_ what the biggest one then was, not whether there was a bigger one. Makes it obvious where you stand.
Re: (Score:2)
You nailed it, I *am* scum of the earth and acting in bad faith. Glad you noticed!
Actually, I'm genuinely curious, but clearly you aren't able to take someone's question at face value.
Cryptographic signatures please? (Score:2)
For over a decade I've been wanting a web standard so news agencies and whoever could digitally sign their published articles so the users could trace the source back to its origin for content that needs to very trusted and verified...
Now it's probably too late with the current media landscape crumbling under the aggregators of this world, but still worth a try maybe?
Only for celebrities (Score:2)
The rest of us doesn't care.
Your faces are worth zilch, as is your Getty stock.
OH NO! (Score:2)
If anything, AI deepfakes prove that the populous will believe anything told to them by the authorized people of authority. Then take action based on that belief as they were trained to do. Except the trainers didn't account for someone faking their appearance and mannerisms, and now they have no way to correct the problem. Sucks for them. I guess they should have limited their order giving to