Deepfaking Democracy: Why AI Threatens News And Global Elections In 2024

Forbes

Deepfaked news segments that appear to be delivered by well-known journalists and TV networks are going viral on social media. Often times, the view counts of deepfaked segments largely outnumber those of real news stories. We take a look at how this application of AI has dramatically improved in quality, and why that is a grave concern as 4 billion people around the world will cast ballots in regional and national elections this year.   0:00 Why 2024 is a consequential year for democracy 1:26 Deepfake news quality is improving dramatically 4:26 How manipulated media is detected 6:39 Spread of deepfake news VS real news 8:50 Why deepfakes are protected by the First Amendment 11:00 How deepfakes can be used in targeted ways to impact elections 12:16 What is needed for defamation lawsuits against deepfake creators 13:11 Deepfakes can be a double edged sword for political campaigns 13:48 How can citizens protect themselves from deepfaked information? 14:45 Why the U.S. government should stay out of debates over truth   Read the full story on Forbes: https://www.forbes.com/sites/alexandralevine/2023/10/12/in-a-new-era-of-deepfakes-ai-makes-real-news-anchors-report-fake-stories/?sh=19112da657af  Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1  Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:  https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript  Stay Connected Forbes newsletters: https://newsletters.editorial.forbes.com Forbes on Facebook: http://fb.com/forbes Forbes Video on Twitter: http://www.twitter.com/forbes Forbes Video on Instagram: http://instagram.com/forbes More From Forbes:  http://forbes.com  Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.

Transcript

00:00 [crowd cheering]

00:08 -2024 will be a record year for elections around the world.

00:12 Over 4 billion people, more than half of Earth's population,

00:16 are expected to cast a ballot.

00:18 Seven out of 10 of the most populous nations

00:20 are going to the polls, and many elections

00:23 will be in countries consequential to the news cycle.

00:27 Taiwan held its presidential elections on January 13th,

00:31 which saw William Lai of the Democratic Progressive Party

00:34 win with over 40% of the vote.

00:38 Lai's election is expected to make relations

00:40 between Taiwan and mainland China more antagonistic.

00:43 Both Ukraine and Russia,

00:47 who remain locked in war with one another,

00:49 have scheduled elections in March.

00:53 And US elections in November are bound to draw

00:55 intense international attention

00:57 in what is shaping up to be a rematch of the 2020 elections.

01:01 -Democracy is still at risk.

01:03 This is not hyperbole.

01:05 -Many academics, political analysts,

01:07 and think tanks expect 2024 to be a major stress test

01:11 on the concept of democracy itself.

01:13 And one particular variable

01:14 that will further complicate this test

01:16 is the rise of AI tools and the ability

01:19 to create convincing, deepfake news content.

01:26 -We're entering an era in which our enemies

01:28 can make it look like anyone is saying anything

01:31 at any point in time,

01:32 even if they would never say those things.

01:35 Moving forward, we need to be more vigilant

01:37 with what we trust from the Internet.

01:39 That's a time when we need to rely on trusted news sources.

01:44 -Deepfakes and cheapfakes are not new.

01:47 But with the explosion of AI that was ushered in

01:50 by the introduction of chat GPT just over a year ago,

01:53 we saw deepfakes proliferate.

01:56 And the types of deepfakes that we've been looking at,

01:58 which are these fake news segments

02:01 using the real likeness and real logos of real news outlets

02:05 and the faces of real broadcasters, are seemingly new.

02:09 And they are particularly problematic right now

02:11 as we are heading into a really high-stakes election

02:15 and also as we are in the midst of a war.

02:18 We have seen deepfake news segments

02:21 from top prominent anchors at all sorts of outlets,

02:25 ranging from CNN to CBS and beyond.

02:30 -Truth or fake, you're beginning with a story

02:33 of a video on social media

02:35 where President Zelensky appears to surrender to Russian forces.

02:38 What's that about?

02:40 -A false video of President Zelensky

02:41 was diffused yesterday,

02:43 where he's apparently making an announcement,

02:46 giving up to Russian forces.

02:48 This video was diffused on a hacked Ukrainian news website

02:52 called Ukraine 24.

02:54 I've been seeing it sort of come in and out

02:56 for several years now,

02:57 but in really seeing it consistently in high quality,

03:02 I would say in the last 12 months.

03:04 -It's only gotten worse.

03:07 Much worse.

03:10 There's no tomorrow.

03:13 -There's two main reasons for it.

03:15 One is that the technology to create deepfakes of news anchors

03:19 has just gotten better.

03:20 But two, and I think this is also important,

03:22 is that most of the major social-media companies

03:25 have eviscerated their trust and safety teams.

03:28 And that's not just Twitter, by the way.

03:29 That one's easy.

03:30 But it's even the Facebooks of the world,

03:32 the YouTubes and the TikToks.

03:34 And so, as a result of that, when people create fake content,

03:37 it's much, much easier to distribute.

03:40 So, remember that when we're talking about deepfakes,

03:42 there's really three parts to it.

03:44 There's the underlying technology,

03:46 the bad actors who are misusing these technologies,

03:49 but then there's the spread of that.

03:51 And the spread of that technology

03:53 is not an AI question.

03:54 That's a social-media question.

03:55 All three things have now lined up.

03:57 The technology's getting better.

03:59 Bad actors are figuring out that you can monetize

04:01 or abuse this content.

04:03 And the social-media companies have fallen asleep

04:05 at the wheel again.

04:08 -Using video from the CBS News archives,

04:11 Chris Ume was able to train his computer

04:14 to learn every aspect of my face and wipe away the decades.

04:20 -This is how I looked 30 years ago.

04:23 He can even remove my mustache.

04:25 -There are two approaches to detecting manipulated media,

04:28 what we call proactive and reactive.

04:30 So, the reactive is sort of my bread and butter here

04:33 as an academic at UC Berkeley.

04:34 What we do is we take an image, an audio, or a video,

04:37 we run it through a battery of tests,

04:38 and we try to figure out if it's been manipulated

04:41 or AI-generated all after the fact.

04:44 So, stuff gets online, some fact-checker contacts us,

04:47 we analyze the content,

04:49 and we eventually tell the fact-checker,

04:50 and they eventually set the record straight.

04:52 And meanwhile, the whole world has moved on

04:53 and gotten defrauded to the tune of millions of dollars.

04:55 So, the reactive stuff is good, if you will, as a post-mortem.

04:59 But at the speed at which social-media moves,

05:02 half-life of a social-media post can be measured in minutes,

05:05 you're not there fast enough to deal with the damage.

05:08 The proactive techniques, the way they work,

05:11 is that if you pick up your phone and record something

05:15 or you are in the business of generating AI content,

05:18 you can inject into that content,

05:20 whether it's real or AI-generated,

05:23 a digital watermark that is cryptographically signed,

05:26 and then downstream, your browser or a piece of software

05:30 can read that watermark and say,

05:31 "Nope, I know that this is AI-generated,"

05:34 or in fact, that it is real,

05:35 and you can do that instantaneously.

05:37 This only works when you have good players.

05:39 So, when Adobe decides it's gonna put watermarks

05:42 into its content, well, great, I trust Adobe,

05:44 but a lot of bad players out there,

05:46 and a lot of this code for creating deepfakes is open source.

05:50 So, if you have open source and you've got some code

05:52 in there for inserting a watermark,

05:53 well, the bad guy's gonna go in there and remove that code

05:56 and we're off to the races.

05:57 So, the watermarking absolutely are going to play a role here

06:01 but they will not, in and of themselves, solve the problem

06:04 because there's always ways around this technology

06:06 and there's open source and there's bad actors.

06:08 But I'm super supportive of that for the big players

06:12 like Adobe, OpenAI, and MidJourney,

06:15 and maybe we lop off half the problem.

06:17 - I'm being replicated by a stand-in actor

06:22 to read this monologue, and then my appearance

06:25 and voice is changed using artificial intelligence.

06:29 So, now that I've got your attention,

06:32 let's send it back to the real Brian Sullivan

06:34 as we dive deeper into this revolutionary technology.

06:38 - Some of the more prominent clips that we had found

06:40 were actually from a TikTok and YouTube creator.

06:43 His name is Krishna Sahai.

06:45 Some of his most watched and most viral videos

06:48 were news segments with CBS anchors

06:52 that interviewed him as the only survivor

06:56 in a school shooting.

06:57 - We are now interviewing the only survivor

06:59 of the recent Texas school shooting,

07:01 high school student Krishna Sahai.

07:03 - In one example, he was the shooter.

07:05 In one example, he was a survivor.

07:08 And across all the examples,

07:09 he was mocking the school shooting itself.

07:12 - We are now interviewing the only survivor

07:14 in the recent school shooting, TikToker Krishna Sahai.

07:17 - Dude, that must have been absolutely traumatizing.

07:19 What was going through everyone's head, man?

07:21 - Bullets, duh.

07:23 - One of the most interesting pieces of this

07:25 and troubling pieces of this is that in many cases,

07:28 the deepfake news segments that we found

07:30 were getting more views and more virality

07:33 than actual news segments from those same outlets

07:37 that were posted to their blue check verified

07:40 social media accounts around the same time.

07:42 One example that we found was from Face the Nation.

07:45 This YouTube and TikTok creator had a segment

07:47 that was actually one of his more innocuous segments

07:50 that was about a group of kids jumping in an elevator

07:53 and the elevator crashes down and then they owe

07:57 this building more than half a million dollars in damages.

08:00 - Over $560,000 in damages liable

08:03 after TikToker Krishna Sahai destroys elevator.

08:07 - So it's not the most threatening example,

08:09 but what was so fascinating about it

08:11 was that it was viewed more than 300,000 times

08:14 and it used again the Face the Nation logo

08:16 and on the Face the Nation's social media account on TikTok,

08:21 the post from the same day only garnered 7,000 views.

08:25 So when fake news segments from a creator

08:29 that uses the outlet's logo or anchors from that station

08:32 is in fact getting more eyeballs than actual news clips

08:37 from the actual outlet's blue check social media accounts,

08:40 you can see how that could become extremely problematic

08:43 and deter people from actually following

08:45 what is considered real news.

08:47 - So even deepfake technology is protected

08:54 by the First Amendment,

08:55 lying is protected by the First Amendment.

08:57 I could make false statements

08:59 and I am not going to be punished

09:01 unless that false statement carries

09:03 some additional harm with it, direct harm.

09:07 Usually harm that is perpetrated against an individual.

09:10 And frankly, when you're in a situation

09:13 talking, making political statements,

09:15 that's the strongest protection the First Amendment gives.

09:19 So a general statement of a political nature,

09:22 which I think a lot of deepfakes

09:24 we're seeing in this coming year will be,

09:26 are actually protected even when they're lies,

09:29 unless there is some direct harm

09:32 that is inflicted upon an individual

09:34 or even to some degree, a small segment of society.

09:38 And what we're talking about here

09:39 are things like defamation.

09:41 If I say something or if I use a deepfake

09:43 in a way that makes a false statement about you,

09:47 harms your reputation,

09:48 that would be something that is now

09:49 outside of the First Amendment,

09:50 not protected by the First Amendment.

09:52 While we do have a collective media literacy problem

09:55 in this country where people don't know the difference

09:57 necessarily between news and opinion,

10:00 or even within news,

10:01 the difference between a good source

10:04 and a not so good source,

10:05 or an outright lying source that has an agenda of its own,

10:08 we're getting better with that.

10:09 I think people collectively, and this is anecdotal,

10:11 people collectively are getting better

10:14 at identifying, separating truth from falsity.

10:17 It's harder when you bring in video

10:20 because they don't know the same tells

10:22 that we're already being trained to look for

10:24 in printed or online information.

10:27 So there's a level of validity,

10:28 of veracity to something that they see in video

10:31 and they go, "Oh, it's video.

10:32 It's really hard to fake that."

10:34 And it's happening so much.

10:35 And frankly, it's mostly being perpetrated

10:37 by people who want to take advantage of you.

10:40 We know that in the 2020 and 2016 elections,

10:43 a lot of the misinformation during the election period

10:46 was coming from overseas,

10:47 from places we aren't going to be able to get to punish.

10:52 And I think that's probably what's gonna happen again,

10:53 which is what makes it so difficult to combat.

10:58 -You said this is the most comprehensive report

11:00 that we've gotten about the 2020 election

11:02 and foreign interference by the intelligence community.

11:05 And it does make clear

11:06 that this massive Russian influence campaign

11:09 was designed, orchestrated by Putin,

11:11 to denigrate Joe Biden

11:13 and to support the re-election of President Donald Trump.

11:16 I wonder -- -I've heard people say,

11:18 "Look, disinformation, deepfakes,

11:19 they can't change an election."

11:21 And I don't think that's true

11:22 because if you look at the last two election cycles,

11:25 the difference between one candidate or another

11:27 in terms of the electoral vote

11:29 came down to some 80,000 votes in a handful of states.

11:33 You don't have to move tens of millions of votes.

11:36 You have to move tens of thousands of votes.

11:38 And not only that, I know where those votes are.

11:41 If I'm the bad guy trying to interfere with your election,

11:44 I know exactly what states,

11:45 I know exactly what towns, what localities,

11:47 and I know how to find these people on social media

11:49 and manipulate them.

11:51 And that, I think, should worry us.

11:53 And you need a series of defenses.

11:55 And so you need a series of proactive defenses

11:57 and a series of reactive defenses.

11:59 And you need better corporate responsibility.

12:01 And you need some liability.

12:02 And you need some regulation.

12:04 And you need good consumer protection.

12:06 And so, you know, when you put all those pieces together,

12:09 I think we can start to trust things

12:10 that we see online a little bit more.

12:12 -It's possible, it's difficult, both in a legal sense

12:19 and a practical sense, to bring a defamation lawsuit

12:22 against someone based on their creation of a deepfake.

12:26 So let's just say you created a deepfake about me.

12:30 I have to show, very specifically,

12:32 that not only you lied, you harmed me in some way,

12:34 and specifically, you harmed my reputation.

12:36 Ah, but beyond that, I have to show a number of other things.

12:39 I have to show us, specifically,

12:42 that you made a materially and substantially false

12:46 assertion of fact about me that was published

12:50 and harmed my reputation,

12:52 and that you did it with some level of fault.

12:55 -This ability to create so many images so rapidly,

13:00 it's an incredibly powerful tool.

13:01 -New artificial intelligence technology

13:03 makes it easy to create fake images

13:05 that can look very realistic, like these created by artist

13:08 and online trust and safety expert Tim Boucher.

13:11 -I think, absolutely, we are going to see the campaigns

13:13 use it against their opponents.

13:16 We also can see campaigns using it to bolster their own opponent,

13:19 to create images of them looking more heroic or taller,

13:22 for example.

13:23 But here's the other place that the candidates can use it.

13:27 Imagine now there's a hot mic of a candidate

13:30 or a sitting president saying something inappropriate

13:32 or illegal.

13:33 They don't have to cop to it anymore.

13:35 They can say it's fake.

13:37 And so they can also deny reality.

13:38 So the deepfake technology is a double-edged sword.

13:41 You can create harmful content,

13:43 but you can also dismiss real content

13:45 by simply saying it's fake and muddying the waters.

13:48 I think the most important thing right now is to remind people

13:51 to think before they share,

13:54 to be a bit skeptical of what they are consuming,

13:56 and to really try to pay attention to the source.

13:59 If the source is an authoritative news outlet, great.

14:03 I don't think we can rely anymore on which accounts

14:06 are blue-check verified accounts and which aren't,

14:09 because now we know that many of the social media platforms

14:12 allow people, any person, to purchase verification

14:16 where the bar is significantly lower for verified accounts.

14:19 But I think that you should really be focused on

14:21 where the news is coming from, who is posting it,

14:24 what their motive may be,

14:25 and what sorts of perspectives they are including in the clip.

14:29 I think all of those things are able to help us,

14:32 especially in a very fast-moving news environment,

14:34 better calculate what is worth sharing versus what isn't

14:37 and help us better understand what we are consuming.

14:44 We do not want the government to step into these discussions

14:48 about truth and falsity.

14:50 We don't want the government to step into decisions about,

14:52 this is my opinion, this is your opinion,

14:54 this is the right opinion.

14:56 We want the government to stay out of it.

14:57 We have a concept in the United States

14:59 called the marketplace of ideas.

15:02 And what it says is we have a place

15:04 where we can all come together and buy and sell

15:06 not products or goods, but thought, ideas, values.

15:10 And in that marketplace of ideas,

15:12 we not only get to test our own thoughts

15:14 and maybe change what we believe,

15:17 we may be able to change other people's opinions,

15:19 but we're also reinforced in what we believe.

15:21 And the whole point, though,

15:23 is we get to sort it out for ourselves.

15:26 We know that phrase that exists,

15:28 a lie travels halfway around the world

15:30 before the truth can put its shoes on.

15:32 I think now the lie travels around the world

15:34 once, twice, three times

15:35 before the truth even wakes up, right?

15:39 We've dealt with technological changes before

15:40 throughout our nation's history.

15:41 The First Amendment remains unchanged.

15:43 It's been there all along and it's worked.

15:47 And I believe in the First Amendment

15:50 and I believe it's gonna work this year as well.

15:52 Thanks, just one last thing while we're rolling.

15:56 How do I know you're not deepfaking me right now?

15:58 That's one of my worries.

16:01 I mean, we're sitting here about deepfakes.

16:03 We could turn, I don't even know if I see this.

16:05 How will I even know whether you're deepfaking me?

16:07 (laughing)

16:10 (silence)

16:12 (silence)

16:14 (silence)

16:16 [BLANK_AUDIO]

Category

Transcript

Comments

Recommended