Celebrity Deepfakes vs. Deepfakes and Effective Celebrities vs. Real Media

Sorry, until last week I had never heard of my new favorite British pop star FKA Twigs, but I was fascinated by it her testimony She told the U.S. Senate Judiciary Subcommittee that she developed a deepfake version of herself (with personality training and can speak French, Korean, and Japanese) to stay away from bots and interact with journalists and her many fans. reported. About her music. This seems like a very visionary and innovative use of technology. She says this will allow her to “spend more time creating art” in an era where there is a lot of press interaction and promotional activity required, but fans and journalists are also concerned about the authenticity of their work. Makes you wonder how you know you're looking at a fake. Not FKA Twigs, but fake FKA Twigs?

Let's make it until we can fake it

The whole real/fake thing can be very confusing.American pop singer Katy Perry's mother Was fooled The star is pictured wearing a swinging floral dress while attending the Met Gala, an annual fundraising gala to benefit the Costume Institute of the Metropolitan Museum of Art in Manhattan. Perry shared a screenshot of a text message from her mother on Instagram. “What a gorgeous dress, it's like a Rose Parade, you're your very own float lol” and her mother said “lol mom, the AI ​​caught you too. Be careful. !

Many of my adoring fans, friends, and family members will soon be facing similar problems. Like FKA Twigs, I have my own deepfake version up and running so I can focus more on my writing, my fantasy soccer team, and my AI-powered sock drawer reorganization. Yes, you can (I'm kidding, of course, there's no app for this yet, but I think it's only a matter of time).Acolytes can now visit Dave GPT Without interfering with communion with the muse. But, for example, a journalist asking about implementation options for a central bank digital currency is not sure if DaveGPT is the real DaveGPT and not a bot run by agents of a foreign power dedicated to destroying our democracy. How can we know?

This is a serious problem. Renowned venture capitalist Marc Andreessen recently spoke on his YouTube channel at Andreessen Horowitz. Said “Deepfake detection doesn’t work because the AI ​​is already too good, so the solution is to certify the content as real.” He’s absolutely right. previously suggested This means you should start configuring your browser and device default settings to not display text, images, and videos that don't have a valid digital signature. So if you're watching a video of Joe Biden playing ping pong with Vladimir Putin, at least you'll know who made it.

You might wonder how Instagram, Facebook, and WhatsApp knew that the image was created by Dave's Photoshop, the BBC, or FKA Twigs. Well, it's simple. Photoshop could be updated to write digital signatures to image metadata by default. So when you save an image, Photoshop calculates a hash of the image and metadata and encrypts it using your private key. Instagram can recompute the hash, follow the link to my public key, and use it to decrypt the metadata. If the hashes match, it's my image and it hasn't been manipulated.

FKA Twigs' public key is publicly known, so it's easy for anyone to check the signature and verify that the contents are from her and have not been altered.

(There is an opportunity here for someone, such as a bank, to provide a digital wallet where the public can securely access the keys and certificates they need without knowing anything about them.)

From now on, if someone takes a clip from an FKA Twigs video and edits it to get an extract, the extract will no longer include her digital signature. Let's say the BBC is doing the editing. Then, when the BBC saves the edited clip, a new digital signature will be calculated for it and signed using the BBC private key. Again, the BBC's public key has been made public and anyone can now verify that this clip is not from her FKA Twigs and is in fact from her BBC .

If only it looked a little futuristic.note that TikTok just announced We announced that we will begin labeling AI-generated images and videos using digital watermarks known as Content Credentials. The Content Credentials technology was led by the Coalition for Content Provenance and Authenticity, a group co-founded by Adobe.
adobe
Microsoft
microsoft
others. It has already been adopted by others such as OpenAI, the creator of ChatGPT. YouTube and Meta have also said they plan to use it. When you generate an image using OpenAI's Dall-E tool, OpenAI attaches a watermark to the resulting image and adds data to the file (using hashes and digital signatures, as described above). This allows you to later indicate if it has been tampered with. When a marked image is uploaded to his TikTok, it is automatically labeled as AI-generated.

Labels alone are not enough

But is it enough to (voluntarily) label AI content? The internet is drowning in “botshits.” As author Corey Doctorow points out, this bot can be created in: ”Scale and speed that stir your imaginationHe emphasizes Amazon's services.
Amazon
The decision to limit the number of self-published “books” an author can submit to just three per day! Given the tidal wave of AI-generated nonsense being uploaded, such as King Charles' cancer Also includes “books” about. We need to do more than encourage tool makers to label their AI content.

Labeling AI content is not enough. On the contrary, you should assume that: all Even if you don't know (or don't care) who that person is, it's crap unless you can provide cryptographic evidence that it was created by a specific person or organization. Marc Andreessen was right to point to digital signatures as the way forward, which means, as I like to repeat, IS-A-PERSON authentication is more valuable than ever. To do.