Interpretation of IMO: AI models can also be tokenized

*Forward the Original Title：解读 IMO：AI 模型也能被代币化发行，币圈抱紧 AI 大腿的新姿势

The crypto market is never short of novel concepts. However, many of these “new” ideas are just slight tweaks of existing strategies; it’s precisely these incremental innovations that often spark the next wave of enthusiasm and speculation. One of the most illustrative examples of this phenomenon is the evolution of asset issuance methods. From the ICO boom in 2017 to the subsequent emergence of IEOs, and the current popularity of IDOs and LBPs (Liquidity Bootstrapping Pools), each shift in the asset issuance paradigm has ignited interest in new projects and provided fresh opportunities for profits among the Degen crowd. The surface may change, but the underlying principles remain constant.

As we step into 2024, with AI becoming a “new pillar” in crypto narratives, the idea of asset issuance centered around AI has opened up fresh avenues for conceptual innovation. A case in point is the recently introduced concept of “IMO,” which stands for Initial Model Offering. On March 2nd, an AI project named Ora Protocol first floated the concept of IMO on its social media, capturing significant attention.

At its core, the notion behind IMO is straightforward: if everything can be tokenized, AI models are no exception and can be tokenized for issuance as assets. However, putting IMO’s principles into practice might not be as simple as it sounds.

Quickly understand the tokenized issuance of AI models

For ICOs and their variants, the essence lies in creating a token with specified quantity, release conditions, functionalities, and various other stipulations, eventually establishing a market value. The “Token” here doesn’t necessarily have a real-world counterpart and can be created out of thin air, colloquially known as “minting a coin.” IMO, however, diverges from this path.

The core essence of IMO lies in the monetization of AI models in the real world. Many open-source artificial intelligence models encounter challenges in monetizing their contributions, leading to a lack of motivation among contributors and organizations due to the absence of financial gains. This is why the AI industry today is predominantly led by proprietary, profit-oriented companies. For open-source AI models to thrive, the key is to gather more funding and build openly.

Therefore, the purpose of IMO is to offer a new method of asset issuance, assisting open-source AI models in raising more funds to support their development. Drawing parallels with previous IXOs, if you’re bullish on a certain token asset and choose to invest in it, you might see returns as the market value of the token improves and share in the income generated by the protocol associated with that token.

Now, in the context of IMO, if you believe in a certain AI model, you have the option to invest in the corresponding token. The provider of the AI model then receives funding for development and growth; and if the model generates economic benefits through its practical application in the future, you may also share in those benefits.

To represent AI models in the form of tokens and share the profits, there are inevitably several key issues that need to be addressed:

• How can we ensure that a certain AI model is real and corresponds to the token you hold?
• How can we guarantee that token holders can truly share in the profits generated from the use of AI models?

Ora Protocol employs two different ERC protocol standards, ERC-7641 and ERC-7007, along with oracle and ZK (Zero-Knowledge) technology to solve the aforementioned issues.

• How do we verify that an AI model is legitimate and not just an empty concept used to raise funds through token issuance?

Firstly, it’s important to understand that Ora Protocol originated as an AI oracle, with its core product being the Onchain AI Oracle (OAO).

The function of this oracle is to validate and execute AI models on the blockchain, ensuring that the deployment and operation of AI models are entirely conducted on-chain. This guarantees the transparency and verifiability of their execution process.

However, since AI models often constitute core competitive advantages, fully exposing them would compromise their commercial edge. Thus, Ora Protocol incorporates an additional technology – opML (Optimistic Machine Learning). In layman’s terms, opML could utilize zero-knowledge proofs or other cryptographic methods to validate the correctness of the model’s outcomes without disclosing the specifics of the model itself. This approach ensures the model’s authenticity and effectiveness while also safeguarding its privacy and exclusivity.

The specific implementation of opML is supported by the publicly available papers referenced above. While we may not be able to assess the technical merits and drawbacks in detail, it’s essential to grasp the impact of this technology. With the AI oracle and zero-knowledge proof, we’ve addressed the challenge of proving the genuine existence of an AI model.

The next issue concerns how to ensure that you own the token corresponding to this AI model and can share in its profits. Tokenizing an AI model is crucial for IMO. Ora Protocol introduces a token standard named ERC-7641, which is compatible with ERC-20.

If an AI model developer believes their model is valuable and wants to launch an Initial Model Offering (IMO) in the crypto market, their approach might likely be as follows:

First, they would associate the AI model with an ERC-7641 asset and stipulate the total number of tokens in the smart contract of that asset;

Second, investors in the crypto market would purchase these tokens, and based on the quantity purchased, would obtain a corresponding proportion of ownership in the AI model (equivalent to being shareholders);

Third, once the AI model is operational on the blockchain, any revenue generated by the AI model or its content (for example, usage fees when the model is accessed, or royalties from the sale of AI-generated NFTs), the ERC-7641 protocol allows for predefined rules for the distribution of profits within the contract. This enables token holders to automatically receive their share of the profits, proportional to the amount of tokens they hold.

Through this mechanism, ERC-7641 tokens become a bridge between AI models and their generated economic value and the token holders, allowing contributors and investors of open-source AI models to share in the models’ long-term value. Hence, ERC-7641 tokens are also referred to as Intrinsic Revenue Sharing Tokens, which can be interpreted as a token standard designed specifically for sharing the profits generated by AI models. This makes the overall logic of IMO quite clear: AI model developers need to raise funds and bind their model to a token for an IMO; buyers purchase the tokens and, in accordance with the rules of the token’s smart contract, enjoy a share in the profits from the subsequent use and creative works of the AI model.

However, this brings us to a critical flaw:

How do you know that AI works created later on the blockchain (such as NFTs, images, videos, etc.) truly originate from the AI model that underwent the IMO and are not fabricated?

The solution proposed by Ora Protocol is to mark these AI-generated works and implement this through ERC-7007. Stripping away the technical details, you can understand ERC-7007 as a token standard designed specifically for AI-generated content. It ensures the authenticity of the content and the traceability of its source.

This standard works by recording the metadata of AI-generated content on the blockchain (such as the AI model used to generate the content, time of creation, conditions, etc.) and leveraging smart contracts to automatically carry out these verification logics. Developers can use zkML or opML to verify whether the AIGC data for a specific NFT genuinely comes from a certain machine learning model and specific inputs. This approach enhances the transparency of the authenticity of AIGC content and, thanks to the immutable nature of the blockchain, ensures that once recorded, the information cannot be altered or forged. Therefore, within the ORA protocol, ERC-7007 is also referred to as the “Verifiable AI-Generated Content Token.”

This standard is now open source and available for review; click here. With this, we have fully understood the logic of IMO:

• Bind AI models to tokens with revenue-sharing features to conduct an IMO.
• Investors enjoy a share of the profits from the future use of the AI model and derivative creative works based on their held token shares.
• Utilize a token protocol that can verify the ownership of content creation to determine whether a work is genuinely produced by the model and share the profits accordingly

It’s still an asset game, and it’s not perfect.

From ICOs to IMOs, the tokenization and issuance of AI models mean that this year’s crypto craze will inevitably be tightly linked with AI. However, the IMO gameplay introduced by Ora Protocol is not without its imperfections.

• Off-chain Usage Issue: Even if IMO can achieve on-chain tokenization and profit sharing for AI models, it still struggles to address profit sharing for off-chain use of the models. When AI models are applied in non-blockchain applications, tracking and distributing these uses’ profits to token holders present a complex challenge.
• Uncertainty of Market Demand: While AI-generated content on the blockchain (such as NFTs) has opened new possibilities for the creative industry, the market demand for these works remains highly uncertain. The market value and liquidity of AIGC works, as well as how much people are willing to pay for these works, remain unknown, making stable AI model revenue sharing an elusive goal.
• Practical Effectiveness of Profit Sharing: Theoretically, profit sharing through ERC-7641 tokens sounds like an appealing idea. However, the effectiveness and feasibility of this mechanism still need to be tested by the market. Especially given the high volatility of blockchain projects and tokens, the actual profits that token holders can receive may vary significantly.

In the crypto world, everyone can innovate with asset issuance, but few can provide a preset definite answer regarding the asset’s utility or user base. Nevertheless, the new model of asset issuance through IMO does provide an innovative framework, allowing open-source AI models to secure financial backing and achieve value sharing through tokenization.

This framework itself is a narrative that closely follows hot topics and carries positive value. In a game where no asset is perfect, embracing the fervor around AI often leads to greater chances of success.

Disclaimer：

1. This article is reprinted from [ Deep Tide TechFlow]. Forward the Original Title‘解读 IMO：AI 模型也能被代币化发行，币圈抱紧 AI 大腿的新姿势’. All copyrights belong to the original author [*Deep Tide TechFlow]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. Translations of the article into other languages are done by the Gate Learn team. Unless mentioned, copying, distributing, or plagiarizing the translated articles is prohibited.