Overview
Last updated
Last updated
AOS(AI for Science OS) is a AI inference verification and sampling network for Hetu protocol on Eigenlayer. AOS can bring enhanced security to AI networks, while supporting more efficient and clean verification services for AI networks.
The AOS network has three roles: staker, operator, and dispatcher
A staker is a user who has some funds in the form of LSTs or native ETH. When a staker interacts with AOS on EigenLayer, it does one of 4 things:
deposit funds
withdraw funds
delegate funds to an operator
undelegate funds from an operator
An AOS-AVS is a combination of onchain and offchain code.It accepts operators who are interested in performing the AOS service and rewards them for this service AOS-AVS Registration with EigenLayer. it does one of 4 things:
distribution task
record task
act as full node of Aos Netwrok
record incentive
An operator is a user who:
has some type of hardware and
wants to use this hardware to make money
By registering with AOS on Dispatcher, the operator could service the AI inference verification task.The staker can delegate funds to an operator by Delegation Manager contract.
Services who accept operators require that the operator have some type and amount of bond which can be taken away from the operator in case the operator negatively affects the health of the service.
AOS is an AI inference verification and sampling network for the Hetu protocol on Eigenlayer. Within AOS, we provide three distinct verification capabilities: verification for OPML (optimistical Machine Learning), verification for ZKML (Zero-Knowledge Machine Learning), and a general verification capability for model inference.
The image illustrates the process flow of an inference task within the AOS network. The AOS network architecture possesses a modular processing capability, enabling seamless integration of various AI verification and inference capabilities. It supports multiple large models, including Llama2-7b and Llama2-8b, as well as custom large models defined by individual users. Consequently, the AOS network exhibits a significant advantage in terms of scalability and extensibility.He modular design of the AOS network allows for efficient and flexible deployment of AI verification and inference capabilities. This modularity enables the network to adapt to evolving AI technologies and accommodate diverse use cases, ensuring long-term sustainability and relevance.
Furthermore, the AOS network incorporates robust security measures, leveraging advanced cryptographic techniques like zero-knowledge proofs (ZKPs) and secure multi-party computation (MPC). These techniques ensure the confidentiality of sensitive data and models, while enabling verifiable and trustworthy AI computations.
The AOS network is designed to be highly performant, utilizing optimized algorithms and parallel processing techniques to handle large-scale AI workloads efficiently. This performance advantage is crucial for real-world applications that require timely and accurate AI inference results.
1. AI-Powered Fraud Detection in Crypto Transactions
Use Case Description: A blockchain platform employs multiple AI models to detect fraudulent activities in real-time. These models continuously learn from new data and use AI inferences to flag suspicious transactions. An incentive system rewards users who contribute computing power to train and verify the AI models, enhancing the security of the blockchain network.
2. Decentralized AI-Verified Content Moderation
Use Case Description: To maintain community standards, a decentralized platform uses AI to moderate content. Multiple AI inferences are made to evaluate user-generated content against community guidelines. Users can challenge AI decisions and, upon successful verification of misclassification, earn rewards. This system ensures content remains community-driven while providing incentives for accurate AI moderation.
3. Blockchain-based AI Model Licensing Marketplace
Use Case Description: AI developers publish their models on a blockchain-based marketplace where users can purchase access. To ensure the integrity of the models, an AI verification process confirms their accuracy and performance. Users who verify and validate the models through a Proof of Clock Sampling process receive tokens as incentives, creating a community-driven verification ecosystem.
4. AI-Enhanced Dynamic Pricing for Digital Goods
Use Case Description: A digital marketplace uses AI to dynamically price NFTs and other digital goods. Multiple AI models analyze factors like rarity, demand, and market trends to set prices. Users who participate in the verification of these AI inferences through a decentralized process earn rewards, aligning their interests with the accuracy of the pricing algorithm.
5. Incentivized AI-Verified Prediction Markets
Use Case Description: A prediction market platform leverages AI to forecast outcomes of real-world events. Users can create markets and use AI models to predict results. The platform employs multiple AI inferences to validate the predictions and uses a Proof of Clock Sampling mechanism to select which predictions to verify. Users who contribute to the verification process by challenging and validating predictions are incentivized, ensuring the accuracy and integrity of the platform's forecasts.
For more technical details you can go to "INTRODUCE TO AOS" page to look