Questions around copyright are going to have a big impact on the future of generative AI, but rather than letting the courts decide what does and does not count as infringement, it would be better for all involved if the parties could agree.
With companies like OpenAI now racing to sign up content licensing deals with multiple publishers, it’s clear that AI providers are willing to pay for the training data they need. But the industry is currently lacking an effective tool that enables creators to track when their content is being used by AI models, which makes it tricky to determine what, if any compensation should be paid to them.
One possible solution that might aid in this is blockchain, the decentralized ledger that underpins cryptocurrencies. It’s a novel technology that could form the basis of a fairer system that keeps track of who owns the copyrighted material used to train LLMs, and compensates the creators of that material.
Projects such as droppLink have already made progress in this area, creating a tokenized system that’s able to tokenize the actions of AI models. In doing this, it can maintain the rights of content owners, and even compensate them for the use of their copyrighted materials.
It’s not magic
Generative AI often seems like magic. Image generators such as DALL-E 2, Microsoft Designer and Stable Diffusion can produce stunning visuals in almost any style, such as a watercolor picture or an old, crumpled black and white photograph. They create these visuals in seconds, and the quality supersedes what many humans could ever hope to achieve. Text generators such as ChatGPT are just as impressive, with their ability to craft essays, write poetry and more in just a few seconds.
Users could be mistaken for thinking these tools are just creating new content out of thin air, like humans do, but it’s not as simple as that. In order for generative AI models to create new content, they first have to learn how to do it, and that means processing vast amounts of information. That information often comes from the web, and much of that content is copyrighted
This is where the questions arise, and many of these questions are yet to be resolved. For instance, do copyright protection and patent infringement laws even apply to AI-generated content? And who actually owns the content created by generative AI platforms? Is it the AI company, the user who prompted the system to create that content, or is it the creator of the original content the models were trained on?
If generative AI is to live up to its potential, these questions need to be answered.
Why blockchain is the answer
Blockchains are decentralized digital ledgers powered by a series of public nodes, and they enable the transparent sharing of information, while preventing that data from being manipulated in any way.
Although it was designed for crypto, it has long become clear that blockchain can also be applied to many other use cases where transparency is desired. By tokenizing physical assets such as real estate deeds, artworks and stocks and shares, we can create more efficient, transparent markets without any intermediaries, with greater trust and efficiency.
As such, blockchain could be the perfect vehicle for increasing the transparency of generative AI, by creating an open, immutable record that verifies where LLMs derive their responses from.
For creators, tokenizing their content on-chain provides a number of benefits. One of the most important ones is that, in the case of AI, they might tokenize their content and give LLMs permission to use it. Because this content is hosted on a transparent blockchain, content owners have a way to track and verify who is using their content. This could result in the emergence of a symbiotic relationship between generative AI and blockchain. Decentralized networks will provide AI models with a clear framework that verifies where they are getting their facts from. With blockchain, a content creator can discover exactly when their content is being used by an AI model to generate a response.
Such a system could then be used to reward content creators. Access to the most powerful generative AI models tends to come at a price, which means those models generate revenue. Blockchain would enable some of that revenue to be shared with those who enable LLMs to use their content.
An analysis of the decision-making patterns of an LLM can help to identify when training material that’s tokenized as an NFT is used to generate revenue. Using smart contracts, it would be possible to allocate a portion of that revenue to the creator of that data, as a kind of royalty payment.
Tokenizing the AI lifecycle
Such a system already exists in the shape of droppGroup’s newly announced droppLink, which is an intelligent service aimed at generative AI developers looking to create more ethical LLMs. The system gives content creators ownership and control of their data. It democratizes data contributions, abstracting away the computational complexities and facilitating micropayments to ensure that rights’ holders are rewarded fairly.
droppGroup has emerged as a champion of ethical AI, building a system based on traceable data practices and compensation for IP usage that directly addresses the concerns many organizations have about the risks of using AI.
There are three primary tech components to droppLink, including the Proof-of-Gen consensus mechanism that validates the authenticity of AI-created content, as well as the training data used in any LLM. It’s uniquely able to track where, when and how this content is used to inform LLMs’ responses to prompts, so creators and IP owners can be compensated for the use of their materials. Stakeholders, known as Computational Resource Patrons, act as validators, participating in this process in return for rewards.
droppLink’s Data Genesis is the underlying protocol that automates the tokenization of AI training data. It establishes an immutable record of the data’s origin and usage.
Finally, droppCoin is the native token of the droppLink ecosystem, and is used to pay rewards to content creators and IP owners whenver their contributions are used by AI systems. Network validators also earn doppCoin as a reward for facilitating the microtransactions that enable these rewards to be paid out.
It’s a promising system that’s uniquely able to address institutional concerns around AI risks, IP protection and data provenance, and it could play a key role in helping the generative AI industry to get around one of the biggest obstacles preventing its wider adoption.
A perfect marriage?
droppLink is just one example of how blockchain can potentially answer questions around copyright ownership and infringement in the generative AI industry, but there are other ideas on the table. For instance, datalatte is an open-source, blockchain based AI chatbot that’s trained on tokenized data. The owners of its training data have full authority over who can use and access that data, and most opt to make it freely available in return for compensatory payments every time it’s used by the chatbot.
As generative AI continues its seemingly unstoppable rise, blockchain is emerging as one of the most promising solutions to resolve the copyright infringement issues that are currently holding it back.
Blockchain is an ideal technology for copyright owners to protect their materials and retain control over it, or else monetize that content if they choose to do so. For generative AI developers, blockchain can provide them with access to an ethical source of training data to empower and improve their models. The marriage of generative AI and blockchain can be both enduring and rewarding, solving the concerns of stakeholders and paving the way for both technologies to mature.
