Hoskinson proposes ‘generative AI-proof watermarking’ to thwart deep fakes

Input Output CEO Charles Hoskinson said the advent of AI deepfakes will force us to assume everything we see and hear is a scam.

AI deepfakes become indistinguishable from reality

Hoskinson at the recent AMAs Asked About using blockchain technology to authenticate visual content. In response, he acknowledged the need for a verification system.

He further said that within the next 12 to 24 months, rapid advances in AI technology will render deepfake videos and audio indistinguishable from reality. Along with that, there will be a gradual change from the conventional “seeing is believing” mentality that is now prevalent to thinking that everything is fake.

He went on to talk about the propaganda use of the technology, which involves nation-states creating and distributing deepfakes as a “means of polarization.”

Hoskinson proposed a way to combat this by taking advantage of the immutable nature of blockchains to store a verifiable “chain of evidence” from image capture to upload. Hoskinson called this “generative AI-resistant watermarking.”

“The only way out of this situation is to have verified information and verified content. must be prepared.”

Other approaches to combat deepfakes

gizmodo He pointed out that deepfake detection technology was in development long before ChatGPT was launched. However, due to its recent widespread use, finding the right detection solution for the task has become more urgent.

Hoskinson talked about blockchain-based content traceability models, but companies like Optic and FakeCatch are focused on identifying AI engagement at the audio and visual level. Similarly, Fictitious.AI takes a similar non-blockchain-based approach, but in terms of written content.

Arjun Narayan, Google’s former trust and safety lead, said detection systems have made some progress, but he doubts the technology is “catching up” with deepfakes. .

A post by Hoskinson proposing a “generative AI-resistant watermark” to thwart deepfakes first appeared on CryptoSlate.