Challenges of merging AI and Web3 infrastructure
While there are significant opportunities in both marketplaces, we have identified major challenges that prevent the development of standout applications. We will analyze these difficulties from both market and technical perspectives to better inform potential solutions for integrating web3 with the AI marketplace.
Blockchain resources are inherently costly: When it comes to the blockchain consensus infrastructure, resources are, by design, typically expensive. The FLP impossibility theorem states that in an asynchronous distributed system, where at least one process can fail, it's impossible to design a consensus algorithm that simultaneously guarantees both safety and liveness. This is a primary reason most blockchain systems adopt synchronous or partially synchronous consensus mechanisms such as bitcoin or ethereum. However, such systems often have substantial storage and bandwidth costs, especially since they store n replicas of the global states. It's therefore essential for the protocol to maintain only the necessary states, minimizing storage requirements. Given the rapid development of the AI marketplace, embedding the entire system into a layer-1 (L1) blockchain solution might not be the most efficient strategy. Such systems generally uphold a consistent block production rate, thereby ensuring a consistent transaction throughput capacity. However, in the case of decentralized AI marketplace, the workload can vary dynamically based on market supply and demand. There might be instances where the system witnesses inactivity due to an absence of incoming training tasks, resulting in most nodes becoming stale without a continuous reward stream. In this setup, the primary goal is to orchestrate AI market activities in the network, with transaction validation serving as a secondary role. A well-constructed framework should:
address these aspects by dynamically adjusting system workload based on the influx of jobs and tasks
enable seamless system upgrades over time
ensure the ease of use and security for users’ assets
Unfortunately, such a system is currently lacking in the industry.
Payment frictions in AI service subscriptions: Even with the assumption that cryptocurrencies offer easier access and operation, the business revenue model for various AI services remains a challenge. While customers are willing to pay for specific tasks, they resist being charged repeatedly when switching between services—a common occurrence in traditional AI businesses. For instance, if you purchase a ChatGPT premium for access to GPT-4 and additional features, you'd still have to pay separately for a Midjourney premium should you need its services. Consequently, customers wanting to use a broad array of AI services could face hundreds of dollars in monthly subscription fees. Even within the same company or network, users don't want to be charged each time they order tasks, as seen with the GPU tasks pricing model in the render network. Exploring how web3 solutions can enhance the user experience regarding subscription practices is of significant interest.
Integrating multiple parties: In the traditional AI business model, there is a direct value exchange between two main parties: the customers and the service providers. Similarly, in most blockchain models, there are only two primary participants: the crypto users who send the transactions and the crypto miners who validate those transactions. Both traditional AI products and web3 communities involve only two major parties. While web3 infrastructure has the potential to broaden the accessibility of AI and offer better market rates, its integration introduces additional participants into the network, thereby increasing complexity. In general, there are at least three parties involved: the customer, the miner providing computing power and storage, and the product designers who contribute the foundational building blocks for various AI services. Developing a system framework and reward models that benefit all three major parties poses significant challenges.
Securities: Web3 emphasizes decentralization. However, distributed systems are inherently unstable and insecure. In designing the system/framework we describe, we must account for a significant number of nodes being faulty or malicious up to a certain percentage. All blockchain systems employ mechanisms to prevent attackers from initiating various types of attacks. Attacking the system should be more costly financially for the attacker than the total potential reward they might gain from the attack. Different blockchain systems implement their own consensus mechanisms to prevent attackers from forging and tampering with data and states, with Proof of Work (PoW) being the most widely adopted.
The consensus mechanism for integrating web3 and AI requires a novel design, as proving service provision can be quite tricky. This mechanism must account for various participant roles. It needs to ensure that service providers are executing their tasks both honestly and diligently. If service providers conduct denial of service or provide low quality service to too many customers, the system should either forfeit some of their rewards or, at a minimum, impact their reputation. This will alert future customers to be wary of these specific providers. Moreover, if a reward forfeiture or reputation system is part of the consensus mechanism, there must also be a safeguard against customers providing unjust or malicious reviews. Without a robust protocol, genuine service providers could become targets of sybil attacks. Lastly, nodes responsible for maintaining global state records must be given sufficient cryptoeconomic incentives to act both honestly and diligently, given their crucial role in ensuring system security.
To the best of our knowledge, such protocols addressing all the challenges mentioned above are currently lacking in the research field, and we don't see many implementations in the industry field, other than fetch.AI and singularityNET which partially addressed the challenges. While the potential market size and areas of opportunity can be tremendous, the following challenges must be addressed to succeed in the large-scale commercialization of the AI marketplace integrated with web3 infrastructure:
Protocol capacity and scalability: The system/platform should be capable of coordinating clients, miners, and AI product development, and it should empower self-governance to initiate, process, and finalize services. The computational power needed to maintain the system's global states should be possible to analyze theoretically. Additionally, the protocol should consider certain commercial factors, integrating "free" features that are specific to the blockchain industry, such as the inflation model, to enhance its appeal to potential customers.
Protocol securities: Given the nature of the AI marketplace, it's impractical for a central ledger to check on every transaction, such as AI services, to ensure they're executed honestly. AI services require substantial computation and incorporate randomness, making it challenging for other nodes to determine if a single node is functioning accurately. A protocol safeguarded by cryptoeconomics is preferable. Systems must be intricately constructed to prevent the potential rewards from being so attractive that the system's design itself becomes a target for malicious activities. When attackers recognize that their attacks will be easily corrected by the system, they have tiny incentive to proceed.
Last updated