The rapid development of technology, and AI, in particular, brings with it the fear of the unknown. With each advancement that pushes us forward, it feels increasingly hard not to look over our shoulder at what might be coming.
Last night was one of those “look over the shoulder moments” for me with the release of OpenAI’s Sora model and the Google Gemini 1.5 update.
If you have been following the AI space — you will realize the capabilities these new models have to render images and video from text has all but solidified the dawn of an era where AI can generate any image or video. That means one day soon, anything you see in a digital format could be AI-generated.
Serious implications for the digital world. Why?
In general, when we see an image or a video, we infer that a camera recorded an image (or series of images in the case of video) of someone or something, somewhere. Of course, that doesn’t mean that these images or videos haven’t been altered but it’s generally believed that if an image or video was excessively altered from its original state, this could be verified or proven. Very soon, this may no longer be the case. We have known this was a possibility with AI for some time but the leap in quality we have seen in the past few days has come faster than expected.
Are we Ready for the Consequences?
Think for a minute how much of the content on the internet is either an image or video. Now think about not knowing if ANYTHING you see online is real. From the beach side candlelight dinner in your co-worker’s Instagram feed to the rows of dead and injured in an attack reported on by a large news agency. One might make you jealous — the other could elicit the response of a nation. What happens to social media when the power to manipulate and fabricate media is an app on your smartphone?
Does it Matter?
Fake news is nothing new. However, most people do not view all news as fake news. With these advances in AI — theoretically, all news COULD be fake news. And that matters.
You may have noticed the shift in the world of information and technology from “Trust, but verify” to “Don’t trust, verify.” But now it feels like we are being forced out of the “Don’t trust, verify” era to ….? Whatever happens when you can no longer verify.
Don’t Trust — Adjust
Many would agree that we have needed to adjust our broken “media” system for some time.. but despite knowing that, we really haven’t gotten around to it. A silver lining is that these newly released AI models might be the motivating force we need to “get around to it.” People have tried (myself included) to look for ways to facilitate the verification or digital “proof” of media. The good news is that there are tools that can establish elements of proof when it comes to our digital files. For example, blockchain tools can immutably verify when an image was uploaded and by whom. Providing you with an image’s timestamped origin, but it can’t tell you who or “what” created the image.
But — what if we combined an immutable storage protocol like Arweave with a physical medium like film? With Arweave, we can permanently preserve a scanned and uploaded image, timestamp when it was uploaded, and see a record of its transaction details and accompanying metadata.
For example — If we were the first to upload an image to Arweave from a scanned physical film slide or negative, we would have a record of the digital files origin. We would also have physical proof in the form of the film negative, proving that the image was recorded by a real person using a camera loaded with film. Would that suffice as a new kind of verifiable proof?
The argument could be made that an AI-generated image could be displayed or projected and captured with a film camera, replicating and nullifying the value of the film process. However, orchestrating that, replicating the camera, lens, and film attributes while substantiating the time and location where the photo took place would be exceedingly difficult.
The same process could be used to record a piece of music using all analog tapes prior to digitizing and uploading the files. This process would identify who uploaded this material, with the physical record of the recording acting as a further verifiable proof. The same process could be used to substantiate video or any other form of media with a physical non-digital counterpart.
The Irony of it All
Many have pointed out the irony of AI’s impact on the tech industry that helped create it. Could AI’s rapid development similarly impact its own advancement — pushing technology to revisit seemingly outdated physical mediums in the pursuit of trust?
Whatever the future holds, these recent developments in AI are a wake-up call. Will they usher in an era that leaves us scrambling to find ways to verify our digital media? Or might they provide the motivation needed to take action and find new ways to build trust?
To quote a recent X post from Sam Williams:
I would like to add:
The future future: on film and onchain or it didn’t happen ;)
