The SOLUTION to FAKE AI pictures & video

Tony Northrup describes the two-sided problem of fake AI images & video. The first side is obvious: AI pictures and video are constantly used to mislead and scam people. The second side is that real images and video are being falsely declared as AI so they can be dismissed.

Today, social media is trying to fight this problem by indicating which pictures and video are AI-generated, but they're doing a TERRIBLE job at it. Most fake AI images can't be detected, and all efforts at labelling them are ineffective against bad actors and scammers.

The better solution is to label REAL images, but that's technically complex. Tony Northrup lays out a workable solution using public key cryptography and secure image processing hardware that can eventually give us green verification checkmarks on images that ARE actually real. People will naturally begin to question any image that isn't verified.

[noriller]

The biggest reason this might not work is: people just don't care.
You can show them irrefutable proof, but if it's not what they already believe in... they just won't care.

[RioHelmi]

As a photographer with a 44 year career behind me (hey i am still active), two things you said: "people want to connect to real people" and "everybody is sick of perfect pictures" (more or less what you said). I have been shooting digital since 95. Do I want to see TIM implemented? Absolutely. I can pick an AI generated picture most of the time, but it's getting slicker. AI is one of the most destructive forces for human creativity.😢

[adamginsburg]

The solution is don’t use Abode software to edit your files, because they steal every image you edit in their programs and upload it to their cloud server and feed it into the algorithm

[NamIess]

I am in a beginner photography group on Facebook and the over use of the photoshop AI stuff is crazy. Everyone wants "perfect" pictures and they happily do body modifications as well 🤮 This should be called compositography or something like that...

[RetrieverTrainingAlone]

Is denoise AI? Is cropping AI? Is dodge/burn AI? A slippery slope...what is "significantly edited" image?

[ronpierson]

If an idea like this were to be successfully implemented I feel like it would likely be short lived. We are on the precipice of a technological change that we can't even comprehend at the moment. Not only with the exponential growth and realistic capabilities of AI, but also with that of Quantum computing. In the same way we can rent time on various powerful AI computers at this very moment, we will also be able to rent time on Quantum computers. And with that comes massive processing power that has the potential capability of defeating most forms of current cryptography; rendering passwords, cryptographic keys, and security certificates virtually useless. And this is not something that's far off into the future. A search for something like "Qubits for Hire" will show that this is beginning right now. It's difficult to imagine what our very near future is going to look like as the capabilities of advanced AI meet those of Quantum computing.

[ghostreaper]

Terrible idea. NSA will be so happy if this implemented. Will be so easy to identify the source of a photo. This can be misused in so many ways. Privacy will be completely gone.

[GemmaHentsch]

the other issue would be compression… authenticated compression/resizing would need to be available for companies like WhatsApp/instagram et al

[genejoanen]

I'm so glad you're discussing this. This has been hard for me to keep up with we are generating all sorts of photographs for our business vote for technical purposes and for production and this has started to become a problem. I agree with other viewers about adobe. I have switched to other alternative editing software for this purpose alone. I hope we find a safe way to put our materials out that are authentic.

[aviationmedia]

Tony, this leads to big brother systems and governments will abuse it.

[MikeNovelli]

Any digital photo is fake.... You're literally taking 1's and 0's and converting them... And if you use any editing app/program or even an OS you are using something that is functioning off algorithmic math, which is what the so-called "AI" generative systems are also using.
It's hilarious when the same people who were using Lightroom and digitally stamping and generative filling, artificially cropping and blurring, color adjusting, watermarking not to mention digitally serving FILES that are translated and transformed into replications of graphics suddenly want to criticize tools others use.
The ship has sailed. Long long ago. You're just upset now because it has reached your artificial threshold for 'fake'.

[lojeda]

We're forgetting about privacy, by adding more information to a photo we create more metadata to be linked to us

[uncledavesplace]

The problem I see is grandfathering in existing equipment. You going to just label everything from a D850 as something. If yur plan take 5 years, the H with the D850. What do I do with that Z9 purchased in 2024? Yeah, you have to start somewhere. Just not sure how practical this is. I would much rather see some kind of screen locking mechanism for the camera.

[YiKidane]

kahma AI fixes this (AI Headshots). SOLUTION to FAKE AI images.

[Wissperwind]

If you buy a 4k TV, display a AI generated photo, take a photo of the screen with your TIM camera, size it down to social media size an ... BAM, you have a valid AI generated image. Seems to be a new business idea...

[bab]

FB feed is loaded with AI generated "nature" photos under the name "National Geographic something or other" to name just one. And I see friends sharing and liking what they think are real photos or videos of birds or other exotic (and impossible!) creatures in the past several months.

[riyaansheikh]

Plot twist: This video was AI generated

[lyfandeth]

Well, as Dr. Land said 60 odd years ago, if it isn't a Polaroid, you just can't trust it.

[contentm]

This will never get off the ground bud. Listing you talk through this video and your case studies, this would be information overload for the majority of the public and you missed a key talking point. The green check marks are verification of real non AI images. But what you missed is if people don't sign up and use TIM then their photos and videos that are real will be accused of being fake which goes against what you are fighting for. This next level of control would be that no images can be posted on Meta unless it's verified as a real image. The point is AI content needs to be checked and verified as AI. That's it. That's the regulation all AI needs to confirm to and follow. AI regulation.

[STOVL]

I like the analogy of locks on a door. A solution doesn’t have to be perfect, it just has to make spoofing more resource-intensive for a bad actor.

I’m really glad that you’re talking about this!