When AI Can Fake Reality, Who Can You Trust? | Sam Gregory | TED

We're fast approaching a world where widespread, hyper-realistic deepfakes lead us to dismiss reality, says technologist and human rights advocate Sam Gregory. What happens to democracy when we can't trust what we see? Learn three key steps to protecting our ability to distinguish human from synthetic - and why fortifying our perception of truth is crucial to our AI-infused future.
If you love watching TED Talks like this one, become a TED Member to support our mission of spreading ideas: ted.com/membership
Follow TED!
X: / tedtalks
Instagram: / ted
Facebook: / ted
LinkedIn: / ted-conferences
TikTok: / tedtoks
The TED Talks channel features talks, performances and original series from the world's leading thinkers and doers. Subscribe to our channel for videos on Technology, Entertainment and Design - plus science, business, global issues, the arts and more. Visit TED.com to get our entire library of TED Talks, transcripts, translations, personalized talk recommendations and more.
Watch more: go.ted.com/samgregory
• When AI Can Fake Reali...
TED's videos may be used for non-commercial purposes under a Creative Commons License, Attribution-Non Commercial-No Derivatives (or the CC BY - NC - ND 4.0 International) and in accordance with our TED Talks Usage Policy: www.ted.com/about/our-organiz.... For more information on using TED for commercial purposes (e.g. employee learning, in a film or online course), please submit a Media Request at media-requests.ted.com
#TED #TEDTalks #AI

[@lornenoland8098]

“None of this works without… responsibility” If there’s anything that defines modern politics and media, it’s “responsibility” 🙄 We’re doomed

[@runnergo1398]

There are humans who have lived socialism type lives in villages and thrived for a long time. If only the entire human race was like that and could convince AI to also be like that. AI could end up being the best government ever. It will lack racism, greed, fear, hatred, anything we consider to be immoral. We give way too much power to a tiny small amount of people who just want to kill people for that power. Imagine an AI (a TRUE AI) that just wants us all to live in harmony while advancing science without hurting anyone.

[@ronkirk5099]

There will always be a segment of our society who will believe anything they see or hear without filtering out the nonsense, but even for the rest of us who are more skeptical and reasoning, it will get harder and harder to sort fact from fiction. Can more technology protect us from the technology? That remains to be seen.

[@callmepistol]

Don’t be naive. Everyone will be susceptible. It will be a war of susceptibility no one will be immune.

[@runnergo1398]

The future will be AI vs AI vs AI. The best AI wins.

[@gb8602]

and that during a time where critical thinking is discouraged

[@jhunt5578]

The AI detection tools need to be democratised. We can't have one agency that just says trust me bro it fake, or trust me bro it's real.

[@StaffyDoo]

This. Security by obfuscation is often more prejudicial than not - specially in the case you mention.

[@CircuitrinosOfficial]

But then the new AI models can be trained to beat them. They become useless as he said in the video

[@rumfordc]

@@CircuitrinosOfficial they're useless anyways if we can't validate them

[@CircuitrinosOfficial]

@@rumfordc it's a lose lose situation

[@Socktupus]

Nothing will detect them for long it’s just like how the biggest companies in the world can’t combat the heroic nerds making adblockers and posting pirated content. You better believe there will be nerds making AI’s that get past detection tools.

[@lactobacillusprime]

In a world where the difference between beliefs and facts are eroding, people disappear into their filter bubbles this is worrying.

[@stormthrush37]

Agreed. Life gets painful so we retreat into our bubbles that tell us what we want to hear. Only that often makes problems worse in the long run when that's the only coping mechanism we have.

[@DanyCervantes]

In a world…

[@guillermoelnino]

a little common sense is a great filter for truth. But pattern recognition is illegal these days.

[@pattybonsera]

A close friend of mine recently had her voice faked in an audio on Facebook. Her FB account was hacked, and the scammer was reaching out to every single one of her friends trying to get personal banking information. I have to say, it sounded exactly like her.

[@stormthrush37]

More than simply AI images is the problem with how used to so many little distortions we've gotten used to in our daily lives. Filters, editing, suspension of disbelief, lifestyle bloggers and influencers that glamorize their lives to make them look way better than they actually are; heck, our own cognitive biases that distort our memories of events. The painful truth is we often don't want to see reality, because life can be so terrible sometimes.

[@EagleProduction-uo7zw]

Hello, I am from Bangladesh and I need a friend from Germany. Can you my friend?

[@Charles-Darwin]

This is truth. I can see the effect on my SO. I've noticed less appreciation for a realistic life in exchange for these grandiose and delusional trappings exhibited by the 'social' one.

[@stormthrush37]

@@EagleProduction-uo7zw If you're talking to me, I don't live in Germany unfortunately.

[@stormthrush37]

@@Eet_Mia That's a central part of my point. Lies usually feed our egos, that's why we buy into them. It takes humility and courage on the other hand to set these aside to face the unvarnished truth. No one is saying it's wrong to disconnect from reality now and again. Life often is a harsh and cruel thing even in a developed country like America, maybe moreso in certain ways at times. The problem comes in when we've gotten so used to these lies that we struggle to recognize the truth at all any more.

[@rumfordc]

the reality is that distortion is part of reality

[@Rikki-Tikki-Tavi]

What about an open source collection of software that can be maintained by the crowd and made available to everyone?

[@cl2791]

The problem is some governments are the very ones who use AI or allow it to be used nefariously. We can only rely on each others' moral obligation for the betterment of the human race, sadly that is not happening and therefore there is no way to guard against total moral and ethical corruption in human propensity for deception.

[@jasonpekovitch7927]

Any detector would greatly improve the quality of the deep fake as it would now have a realtime feedback system to train on. I appreciate that you included well just keep them out of the hands of the “bad guys”. That is never a successful strategy.

[@testtest-xg8jk]

Unless the tech is hidden and not exposed as an api

[@vctaillon]

here's a great solution. push all the power buttons to off. Go for a walk in the woods, remember the woods?

[@divetank]

Thank you Sam. Been following 'Prepare Don't Panic' campaign for several years and show your videos to my university communications students.

[@yungsookevinhong7943]

one of the work around is the file must have to include the machine and chipset that process the file in block chain technology plus encryption to conceal some of the real important info for privacy and scurity.

[@matthewdozier977]

I missed the part where you told us how we achieve anything without ultimately putting all power to proclaim what is real in the hands of some group of people.

[@jaredjeffery8476]

The core technologies he's referencing are the fix. Look up "self sovereign identity." It's a new set of technologies designed specifically to decentralize identity back to those who the identity is about.

[@Azel247]

It is dangerous to believe everything you see. It is also dangerous to believe nothing at all. The challenge is finding that balance.

[@loner1295]

I got an ad for an AI program that “helps “study aquatic biology somehow. Good timing.

[@nobodyexceptme7794]

Same here, skynet is already here 😂

[@ericmedlock]

The exact same people I currently trust, no one

[@linachen8475]

didnt know robert downey jr in interested in deepfake

[@RISCGames]

Cat is already well out of the bag and it’s a little too late.

[@stevechrisman3185]

This leaves me with a sick feeling.

[@easkey123]

OMG.... What a world we are giving our children 🙏🙏🙏🙏🙏🙏🙏🙏🙏

[@mats66]

😔

[@aamit23]

Agree

[@dottnick]

This subject reminds me of certain episodes of Star Trek voyager when the holographic computer doctor made his holodeck drama. What the rights were and why? Here we are… lol

[@oomraden]

Taking a step back and reading between the lines always work for me, at least until now

[@mrs.sherry]

Can you guys ask the AI robots which appearance style they find more appealing, humanoid or like a comical toy or mechanical or like a pet animal? Do they even have a preference? Does it bother some of them that the wiring is showing or parts not complete?

[@Zaniel8]

Plot twist, humanity that created AI Already , HUMANS ALREADY, HAS A MASSIVE ISSUE WITH ACCEPTIN REALITY

[@astrodysseus]

8:10 of course we need proper provenance / references. we needed them before to structure data, we need them now even more to help in the detection of deep fake and fake data

[@Elevate63]

Plot Twist, this video is Ai generated.

[@mistressfreezepeach]

a TED talks text2video model definitely isn't that far away

[@MrCalverino]

😹😹😹😹

[@thanhphuoc1209]

​@@MrCalverino 🤨📸

[@planetmuskvlog3047]

😂 good one

[@nadademamaditas]

2nd plot twist, Furyology is an AI

[@dhruvdangi_]

It's getting "HARDER" ISN'T IT

[@EvyatarAI]

🎯 Key Takeaways for quick navigation: 00:03 🕵️‍♂️ *It's becoming increasingly challenging to differentiate between real and fake content, especially with the rise of generative AI and deep fakes.* 01:32 🔍 *WITNESS, a human rights group, leads efforts to combat manipulated reality, including a deepfakes rapid-response task force to debunk false claims.* 04:07 🌐 *Deepfakes pose a significant threat to shared, trustworthy information, vital for the functioning of democracies.* 05:41 🛠️ *Structural solutions are necessary to discern authentic from simulated content, including tools for credibility fortification and reliable detection technology.* 07:43 🔎 *Detection skills and tools must be accessible to journalists, community leaders, and election officials to effectively combat deepfakes.* 08:43 📝 *Content provenance and disclosure, along with cryptographically signed metadata, are essential for understanding the origins and authenticity of AI-generated media.* 10:52 🔍 *Transparency, accountability, and liability within the pipeline of responsibility for AI are crucial to prevent misuse and abuse of AI-generated content.* Made with HARPA AI

[@mk1st]

I just had to look up "Wankers of the World". Looks to me like the AI version of Private Eye.

[@williammillerjr9028]

It must be in the hands of our first line of defense...

[@Libertariun]

… the journalists, the community leaders, the election officials, who are our first…. No, no, I can’t go on This guy really thinks that journalists are our first line of defence in discovering the truth, the truth?

[@Leto85]

Maybe I think too simple or am missing the point, but wouldn't it help just to be more sceptic and don't take anything at face value? We can't trust the news nowadays, but we never could. That's just something we have to accept. You can follow the news if you like, but don't even think to assume that we are given information we are not supposed to know.

[@minimaxhall]

See news agencies as entertainment. See what they write as "the official" story. But not necessarily the truth.

[@erikreber3695]

Wait. Why shouldn't everyone be able to detect this bullcrap?

[@groboclone]

As he explained, any verification tool that is publicly available immediately becomes useless because those with an incentive to will figure out how to fool it

[@rumfordc]

@@groboclone his explanation was invalid. the incentive to fool the detector exists as soon as *_any_* number of people start using it, so by his own reasoning nobody should be allowed to use any detectors.

[@groboclone]

@@rumfordc Yes the incentive exists regardless, but learning how to fool the detector requires access to the detector so that you can learn exactly what it is looking for, bad actors could even use machine learning to achieve this. Obviously there may still be bad apples in a curated selection of people granted access, but the chance of a detector being quickly cracked at least goes below 100% in that case

[@rumfordc]

@@groboclone and learning to trust the detector requires access to the detector so that you can learn exactly what it is looking for. it doesn't matter if the system gets cracked at that point, because it was never useful in the first place, as its answers couldn't be validated.

[@trqster]

Its a revolution coming and its gonna be havoc no matter how optimistic this message sounds... As Elon Musk put it better, the era of (super) AI wars is uppon us. And there's no turning back...

[@trqster]

@@deanhurley311 He wasn't mentioning any products, he was giving his opinion about general A(G)I evolution and what it meant in a geo political pov - wars will be fought between nations, the west (good guys) and the others (bad guys') more and more by their own AI super powerful entities

[@bert3163]

The sheer volume of media, and amazing speed of consumption amid busyness, means tribalism and confirmation bias will rule the day.

[@brunoschaer3758]

This robot at first fooled me. It really does look like a real person and seems to speak like one too. It's amazing what technology is capable of these days!

[@AnnCatsanndra]

If your security schema is weak enough that having its systems under open public scrutiny can defeat it, it is probably not a very good security system at all. Security through obscurity is seldom viable.

[@beegood1215]

The end of seeing is believing.

[@SeaScoutDan]

Using an AI to test if something is AI generated. Sounds like this is a computer vs computer Turing test. That feels similar to a student using chat GPT to write a term paper. Teacher says not to use AI to write, then submits the paper to AI to grade and check for plagarism.

[@peachmango5347]

This isn't helpful. It assumes "frontline people" are ethical. NO WAY.

[@KT-po8mj]

thank you. Hope good everything!

[@tekannon7803]

What is good about the problem of deep fakes? I am using Anthony Robbins' question in one of his books about how to look at problems. The AI phenomenon is here to stay so what can we start doing today to insulate ourselves from getting conned? We must multiply our resources for everything we hear or see that is happening in the world around us. Yes; that's right: instead of listening to only your favorite news channel, get in the habit of tuning into other news channels.

[@NikoKun]

His proposed solutions are not enforceable or sustainable, and that will only get worse with time. We need to let go of outdated mentalities and justifications for how our systems work, it is our only option to adapt to AI.

[@ripkm-iwaly]

so the only option is no option but that option? that seems not only close minded but also an ultimatum, there is something you can do about it, give up on the internet but then again that might be too extreme for 99% of us

[@NikoKun]

​@@ripkm-iwaly Nobody's going to give up on the internet, and expecting them to do so as some kind of solution, is much more an "ultimatum", in my view. The option I'm suggesting is a freeing one, which opens up significant improvements to our lives. We simply need to let go of something which will no longer be morally justifiable, we need to stop valuing other humans, based on outdated and greedy mentalities. The requirement that people must compete for a "job" to survive, as the determining factor for how we distribute basic resources to people, must change. Everyone has the same right to exist as everyone else, the same right to the inherited resources of the planet. And AI exists thanks to years of data, collected from all of us, which trains it. It is a creation of society, so we must find a way to ensure everyone benefits.

[@ripkm-iwaly]

@@NikoKun that is very altruistic although wrong...there's been a dumbphone movement going on so people are already adapting to it and "nobody's going to give up on the internet" that's a mindset not empirical evidence, fact check it and people having the same rights...i would love that to be true, i truly would, trust me i would love for that world to exist but unfortunately that is simply not true, not only is it not true as it is impossible according to human nature, then again AI might be able to make that true since it is far smarter than humans but then again, AI is in the hands of humans, not the other way around

[@ripkm-iwaly]

@@NikoKun you know nothing jon snow...i mean, NikoKun, I would tell you more but certain truths are infohazards, f.y.i.: infohazards is information that just by knowing it could put you in danger, therefore i will refrain from saying it because although i disagree with your opinion i value your safety, farewell and keep that critical thinking going

[@ShutUpWesley]

"Who can you trust?" As many as I do now.🧐

[@onjofilms]

We're all going to die.

[@trustworthyguy8645]

Me! You can trust *me!* I am a very trustworthy guy!

[@georgesos]

I don't trust any public figure anyway,so AI or not ,i don't give a dam. Those who form their " reality" based on politicians or other public figures ,are already a lost cause.

[@izzy1161]

True

[@Octwavian]

I got zero hope. We won't prevent this, there is no real solution. At most, we can delay it.

[@HusKrWolf]

I used AI to take notes of this for me lol

[@manuelb1255]

I don't agree with his justification for elites. I would prefer a more democratic approach, I am more on the side of the open source projects that he mentions later.

[@justwanderin847]

I have thought about Artificial Intelligence and I think it would be constitutional and necessary to add one item to the US Public Law on Copyrights. Please just add a definition to the word AUTHORS to say, that Author is to be defined as Human. You must be human to copywrite something (even if created by a computer). That way any output from AI, (image, picture, song, voice, science, math) cannot be copyright , only a human can copyright the output, thus the copyright would apply to a person. Copyright law (Title 17, U.S.C.), does not define the word “author”, so just define it as Human. Just as the Constitution gives the USA NO authority to dictate to any country the type of weapons they can Have. So be it with AI. The constitution and common sense tells you that you can not govern computer programming on a world basis. Or constitutionally within the USA. We DO NOT need Government to regulate Computer Programming (AI). Big media and big government are just trying to scare people into giving up their liberty for some faux safety. AI is a computer program and has no need of an "AI Bill Of Rights". AI has No Rights. But the President of the United States already has the "Blue Print for An AI Bill of Rights".

[@nestorrfortuna1]

We will go back to talk to each other face to face, for important issues. This is simple to solve. Sign: A.I.

[@nivethanansivarajah8426]

Thought this guy was Ryan Renolds for a sec...

[@balonh1052]

I think we should have private “interpassword”

[@tastemysaucer]

I didn't trust anything before AI, so it's the same to me, really.

[@maodamorta4346]

This future sucks and we could've lived perfectly happy lives without this technology having ever existed. I can't even help but laugh at this fool's errand he's trying to hustle. "Of course! We'll just get another neural net to figure it out for us, because we have no clue how to deal with this problem with our own brains." The fact that these conversations are starting this late to the punch and that this is the solution offered means we already fumbled this ball and trying to stop it will be like trying to plug the Titanic with the wine corks from behind the lounge bar. What an ugly future. Life could've been so simple man...

[@gideonking3667]

Seems like a losing battle

[@lofichadii]

😐

[@derekholland3328]

if it gets to a point where you CAN'T tell if a video, pic, etc. is real. then how can you reliably tell if this life is real.

[@UnDaoDu]

AI will be able to discern deep fakes by simply simply by seeing their source.

[@TiredMomma]

Spelling and grammer mistakes is sometimes a dead giveaway too. You have a grammer mistake, and could easily be a bot for me to report to Google.

[@megaplay]

Primitive humans one of the most powerful tools in the universe and your worried about deep fakes. 👁 This world no longer belongs to the old.

[@PolarisClubfan]

When the AI fake the reality who can you trust? Nobody, go back to school 😂

[@PuckerFactor10]

The key to detecting lies and misinformation is to take great care of your internal compass. Your conscience. Keep it clean. Stop ignoring it when it condemns what you are doing. Instead change what you are doing. Stop searing it with the hot irons of denial. If your conscience bothers you there is a reason. Don’t let others tell you that it’s ok when you know it is not. You will always filter the things you hear, read, and see through your conscience. If it is clean and strong it is virtually impossible to be lied to successfully.

[@big_dad9465]

“Deepfakery” it really is tchiswasday innit.

[@Hardwareai]

100% organic meatbag generated comment here ♾️

[@mistressfreezepeach]

good bot, here 🍪

[@LeeCarlson]

Until AI discovers context it will not be capable of actually deceiving humans.

[@CircuitrinosOfficial]

The whole point of models like ChatGPT-4 is to understand context. It can already do that quite well.

[@PorkchopExpression]

There's literally millions of informational youtube videos with the script and voice entirely AI generated. I doubt most viewers even realize it. It takes me a minute to notice it sometimes and I do tons of AI work for my job.

[@LeeCarlson]

@@PorkchopExpression, and anytime that I get one of those KZbin channels I tell the algorithm not to recommend it to me again, because they drive me nuts. You see, I recognize that the script and voice are generated by an algorithm and figure that if the channel cannot afford to pay voice talent they have nothing that interests me in the first place.

[@hugedoofus]

You talk about serious issues, but your solution isn't realistically enforceable, or something that would be good to have.

[@kevinsayce2248]

AI generated answer (cannot detect sarcasm) 😉👍

[@EagleProduction-uo7zw]

There is lot of song in KZbin which is generated by AI with famous singer voice although the song was not sung by that singer. I'm really agree with Mr. Elon Musk. He said "AI is far more dangerous than nukes."

[@mencken8]

And to think, it all started with a can of Coke on a coffee table…..

[@benhud]

Why mitigate against that which accommodates corruption? How do we, from now on, apply AI, which will inevitably get better at information than all of us combined, to cause us to surpass our reliance, trust and dependence on information all together? What processes and outcomes can we collaborate with AI to accomplish, even now, to cause humanity to get better results faster and in more effective and safe ways that transcend our sole primary reliance upon this harmful, risky, corruption-producing thing we call information? Oh wait, we can't! Information is our God! And thinking is our way of worshiping it. *sigh*

[@soundhealingbygene]

My whole life is deep fake

[@mats66]

Democracies will die out because people will just tune out.

[@thanhphuoc1209]

Lol are u sure u can protect all of users in the internet 😂?

[@kathleenmiedema6894]

Please......when they say peace and safety

[@vickygllc]

👏🏻👏🏻👏🏻

[@Monterrey_Manila]

Forma now on you need your signature and info to back up your words and your content no identity verification is sent to internet trash bin

[@TheCaphits]

No, that's Ryan Reynolds.

[@sciencesaves]

Ted talks can now have 15 people in the audience? Is this a fake Ted Talk video? The irony

[@pietercoetzee9376]

😮ooooooooeh

[@gregw322]

Please read “The Hedonistic Manifesto” by British philosopher David Pearce. He proposes using AI, but also other advanced tech to eliminate all involuntary suffering in all sentient life. He convincingly makes the case that it should be the primary goal of mankind.

[@user-td4pf6rr2t]

I feel like if government gets involved would be violation to freedom of speech. Since the responses are guided - like how gpt doesn't like talking about sensitive data, if this was fashioned by government entity would literally be the definition of not free speech.

[@MoritaJunichiro]

Human and AI research: Reckless research

[@jayanthanj858]

PTR palanivel Thiyagarajan maati kondara?? 😢 Udhayanidhi thiruttu அம்பலம்

[@rogerc7960]

Lawyers are not the solution

[@minor12828]

You can't trust nobody anyways 🤷‍♂️

[@cbaschan]

we are doomed

[@mahedihassan7566]

One day A'I will destroying our Humanity.....!

[@alphabetsoup488]

Fortunately, I trust God, and B.S. gives me a bad feeling in my gut! Never rely on man for answers! If you want to screw up AI, use sarcasm in your speech!

[@Samuelir96]

harming women and girls but not men and boys, the fuq.

[@ViolinistJeff]

I´m not afraid of deepfake. It make take some time, but we will get used to the possibility of deepfake everywhere. It´s like when the modeling industry used extensive photoshopping to make women look too perfect. Some of us had to learn that that is an unrealistic standard of beauty and to never trust that what you see online is what it looks like in real life. I predict that major video websites like KZbin, Facebook, Vimeo, etc. will have extensive and very accurate deepfake detection techniques and algorithms. As there is a check mark beside the channel name TED to help prove to you that this is the official TED talks channel, there will be a check mark beside the video to tell you whether it has been deepfaked or not. But the detection methods of these websites will remain top secret to stop people from being able to get around them. Many aspects of Google and KZbin are already a secret.

[@strangereyes9594]

Without watching, obviously only those with a digital ID.... d'uh. Let's get one bois, ze bugs arent gonna eat themselves.

[@Leto85]

This works both ways though: now we can finally commit online atrocities on KZbin and just say it's AI generated, just like this comment, really.

[@Eaglemadhatter]

None of this is real

[@MostlyLoveOfMusic]

Ok but using A.I. to put any person into a pron situation is a brilliant idea, and don't listen to prudish counter-arguments

[@raggamuffin2682]

Jewish

[@fajam00m00]

@@raggamuffin2682Jealous

[@mistressfreezepeach]

Hannah Arendt also talked about the "Banality of Evil". It's impossible that there will be enough accountability and a "pipeline of responsiblity" for this to work, especially with women holding our institutions in a grip.

[@rockshankar]

I hate to listen to old people that comes out all of a sudden knowing a lot about AI, when young people cant afford rent to stay in a world created by them.

[@RohitKumar-sj5wg]

I don't know why these people are making human life so complex, people are losing jobs because of this, the ai creator should be jailed, 😠😠😠

[@jzjsf]

From the Transcript: "It's getting harder to identify deepfakes.. What this means is that governments need to ensure that within this pipeline of responsibility for AI, there is transparency, accountability and liability." 1. We knew the first part before we starting watching your video. 2. We knew the second part because of course this can only be combatted with transparency, accountability and liability. 3. YOU DIDN'T SAY EXACTLY WHAT SHOULD BE DONE!!! I hate most TED talks because they talk about some obvious problem and then, at the end, provide useless generalized solutions. I don't want useless generalized solutions. I WANT EXACT AND PARTICULARIZED SOLUTIONS. Thanks for wasting my time, again, TED.

[@PorkchopExpression]

Because there is no viable solution.

[@jzjsf]

@PorkchopExpression I believe there is a specific viable solution. We already criminalize fraud. Follow that process, and criminalize intentional deceit.

[@tobfrank_]

Ted is out of new ideas

[@yannickmantele5788]

WE NEED REGULATION