In May 2019, a video of Speaker Nancy Pelosi, appearing to be drunkenly slurring her speech, went viral. It was quickly debunked as a deepfake—a manipulated video designed to mislead viewers.

This was many internet users’ first interaction with a convincing deepfake. The ensuing furore highlighted the growing threat posed by deepfakes and AI-generated content. Deepfakes leverage advanced machine learning techniques to create hyper-realistic but entirely fabricated videos, images, and audio. As these technologies evolve, distinguishing between authentic and fake content becomes increasingly challenging. Since then, content authenticity has become a central part of the debate surrounding AI.

Let’s dive in.

Challenges Posed by Deepfakes and AI-Generated Content

The term deepfake was coined in 2017 after a Reddit user created a portmanteau of “deep learning” and “fake”.

While the term is relatively new, this kind of process has been possible since the late 90s. Momentum only really picked up when Ian Goodfellow created Generative Adversarial Networks (GANs) in 2014. GANs consist of two neural networks—the generator and the discriminator—that work together to produce increasingly realistic content. Since then, the technology has advanced rapidly, making it easier to produce convincing deepfakes with minimal technical expertise.

Vulnerabilities in the AI Generation Process

Misinformation online doesn’t just come from shocking deepfakes or polarising social media content. It can also sneak into the very AI generation process itself. The creation of AI-generated content involves a complex process that, while innovative, introduces several vulnerabilities at different stages. These weaknesses can be exploited to manipulate or distort the information that AI systems produce, leading to the dissemination of false or misleading content.

Let’s take a look at some of the vulnerable stages:

• Data Collection: The initial stage where data is gathered can introduce the first set of vulnerabilities. Biased or fabricated data can skew AI learning algorithms, resulting in outputs that may perpetuate inaccuracies or false narratives.

• Output Generation: Finally, the actual content produced by the AI can be altered. Manipulating the final output can be particularly harmful as it directly affects the content being consumed by the public.

Combating Content Inauthenticity with Blockchain Technology

Blockchain technology is one viable solution for tackling the internet content crisis because its decentralized nature offers an immutable audit trail for each piece of content, ensuring a verifiable and unalterable record of its origins and any modifications. Every piece of data that is recorded on a blockchain is timestamped, documenting the exact time of creation and any further changes that are made. Timestamps are created via a digital signature verified by hash functions, which create a sort of digital fingerprint for the original data. This information is publicly available and verifiable, allowing users to certify the authenticity and provenance of data rather than simply having to trust a centralized authority.

Blockchain is already being used to support content authenticity, from data gathering to distribution:

1. Transparent training and data collection: Blockchain can address AI biases at the dataset level by tracking data provenance, adding transparency to how models are trained. Blockchains also allow for sharing of data between parties, avoiding biases from centralized datasets. In this sense, blockchain is not a magic fix, but rather a useful lens to observe, track and correct biases.

2. Content authenticity: Blockchain can create provable attestations for content that document its history, taking in creation, editing and distribution, making it easier for viewers to know if they can trust what they are seeing. It’s even possible to note the camera, creator and encoding details, all of which can be checked independently.

3. Built-in Verification: These attestations can be integrated into the stack of the existing media ecosystem. Browsers, media platforms and verification systems can then use this to both tackle fake content and build the trustworthiness of their brand. This can be done through cryptographically secured identities and domain names, such as ENS handles.

Blockchain can help stem the rising tide of deepfakes online too. U.S. media company Fox partnered with Polygon to create Verify, a protocol to combat deepfakes and AI-generated stories. According to Fox, “publishers can register content in order to prove origination. Individual pieces of content are cryptographically signed onchain, allowing consumers to identify content from trusted sources using the Verify Tool.”

Joining the Coalition for Content Provenance and Authenticity

These stages underscore the importance of robust frameworks like the Coalition for Content Provenance and Authenticity (C2PA), which was established to create standards that ensure digital content remains genuine and reliable from creation to consumption.

The C2PA, initiated by industry leaders such as Adobe, Microsoft, and the BBC, focuses on embedding provenance metadata within the content. This metadata, secured through cryptographic means, offers a comprehensive record of the content’s origin, creation process, and any modifications, fostering transparency and accountability. In practice, this means that every piece of content could come with a C2PA ‘nutrition label’, in which it’s possible to check:

• The creator’s identity

• The tool used to make the content (Adobe Photoshop, DALL-E 3, camera)

• Time of creation

• Assets used and actions performed on them during creation

This so-called ‘label’ would travel with the content across all platforms and could be checked at any time to see if any alterations have been made.

Livepeer is committed to building the world’s open video infrastructure, so ensuring that video content is authentic and clearly labeled when generated by AI is an important part of the firm’s vision. Livepeer intends to follow the standards outlined by the C2PA to ensure the integrity of digital media, embedding crucial metadata that tracks content from its origin through all modifications. This approach not only combats misinformation by verifying content authenticity but also supports creators by protecting their work in a landscape increasingly influenced by AI technologies.

Establishing the authenticity of content online is fast becoming one of the most pressing issues of our time. The sheer scale of the challenge means that companies, governments and organizations will need to work together to make any kind of meaningful impact. The technology behind deepfakes will unquestionably improve with every passing day, but so will both the tech and resolve of those trying to stop them. As a company that helps creators with their video content, we are very intentional about tackling authenticity. The road ahead is long, but with C2PA and blockchain we’re already making a strong start.

Livepeer is a video infrastructure network for live and on-demand streaming. It has integrated AI Video Compute capabilities (Livepeer AI) by harnessing its massive GPU network and plans to become a key infrastructure player in video-focused AI computing.

Twitter | Discord | Website