Kite AI's Proof of AI

Original Title: Kite AI’s Proof of Attributed Intelligence (Proof of AI)

Original author: arndxt

Source of the original text:

Compilation: Tom, MarsBit

The test network has undergone rigorous verification with 200K+ datasets and 1,500TB of data.

Kite AI has established a collaboration network with top AI partners in the world, including AVAX, Near, University of California, Berkeley, Sui, and Polygon Labs.

Let's delve into Kite AI's first 'attribution intelligence proof' together.👇🧵

1. Introduction

In recent years, the rapid development of deep learning and machine learning has driven transformation in various industries. From healthcare to finance, artificial intelligence (AI) has become the core of technological innovation.

Despite significant progress, the development of AI is still mainly dominated by a few well-funded and centralized entities that typically control access to data, computing resources, and proprietary models. This situation raises fundamental questions about the fair distribution of value in AI systems, data ownership, and alignment of incentive mechanisms.

Kite AI's mission is to change this situation.

Against this background, Kite AI has emerged as a blockchain solution designed specifically for decentralized AI research and applications. By adopting 'Proof of AI' (PoAI), Kite AI aims to provide a transparent, secure, and fair coordination layer for AI data, model development, and AI-driven agents.

Kite AI has partnered with AVAX to launch the first AI-focused Layer 1 sovereign blockchain.

By leveraging Avalanche's high-performance, scalable infrastructure, Kite AI ensures:

Utilizing Avalanche's subnet and consensus efficiency to achieve fast AI computation.

Seamlessly and limitlessly scale to support AI workloads without bottlenecks.

Provide decentralized, permissionless infrastructure for AI research and model deployment.

Kite AI Testnet link:

2. Background and Motivation

2.1 Centralized AI Ecosystem

The traditional AI development process heavily relies on centralized data warehouses and centralized computing resources. Dominant AI platforms typically leverage large datasets, which come from public and private channels, but do not adequately reward the original data providers. As a result, data contributors and model developers often operate in an imbalanced power structure, often unable to receive sufficient recognition or compensation.

In addition, the closed governance mechanisms in the AI field restrict transparency, hinder reproducibility, and may lead to monopolies. Centralized governance weakens open innovation, limits collaboration opportunities, and increases the risk of bias or improper model use.

2.2 Existing Blockchain Solutions

To address this issue, some blockchain-based frameworks have attempted to decentralize the AI and data markets. Traditional consensus mechanisms, such as Proof of Work (PoW) or Proof of Stake (PoS), have proven effective in certain cryptocurrencies and DeFi applications. However, these mechanisms often fail to address the following issues:

Fine-grained attribution: It is necessary to reward individual contributors based on the marginal contributions of data providers, model developers, AI agents, etc.

Customized governance: requires specialized environments tailored for AI tasks, including large-scale data indexing and on-chain/off-chain computing.

AI Incentive Mechanism: Advanced game theory models prevent data plagiarism, model theft, or malicious contributions in the training process.

2.3 Infrastructure that needs to be specifically built

The universal blockchain protocol lacks specialized functionality for handling the complexity of AI development and commercialization. These limitations include insufficient throughput, inability to store or reference large-scale datasets, and difficulty in attributing value in multi-layer AI workflows. Kite AI's proposal - a Layer 1 blockchain that is compatible with EVM and enhanced by PoAI - aims to fill these gaps and drive a new AI economy built on fairness, transparency, and inclusivity.

3. Kite AI Architecture

Kite AI has launched a brand new Layer 1 blockchain that integrates four key components specifically for AI.

Attribution Intelligence Proof (Proof of AI)

Decentralized data access engine

A combinable AI ecosystem with customizable subnets

Decentralized, portable AI memory

3.1 Attribution Intelligent Proof (Proof of AI)

Proof of AI is Kite AI's core consensus mechanism. Unlike Proof of Work (PoW) or Proof of Stake (PoS), which mainly focus on computational challenges or collateral guarantees, Proof of AI aims to measure and reward the real contribution to AI assets:

Data contribution: Data providers are rewarded based on data quality, relevance, and improvements to model performance, among other indicators.

Model development: Developers are rewarded based on the accuracy, efficiency, or user acceptance of the model.

Agency Effectiveness: AI agents (such as chatbots, automatic trading agents) are rewarded based on their service usage, reliability, and user satisfaction.

Proof of AI combines data valuation techniques (such as methods inspired by Shapley values) and on-chain governance to dynamically evaluate how each contribution impacts the overall AI economy. This establishes a feedback loop, incentivizing meaningful inputs and suppressing malicious or redundant activities.

Attribution intelligent proof incorporates advanced game theory mechanisms to prevent rational and irrational attacks:

Rational Attack: Behavior that attempts to maximize rewards without actual contribution will be prevented through marginal contribution scoring.

Irrational attacks: malicious behaviors such as data contamination or model destruction will be detected and punished through on-chain detection, ensuring the stability of the system.

3.2 Decentralized Data Access Engine

Kite AI's decentralized data access engine provides permissionless yet secure data retrieval and storage interfaces. The engine supports:

High Volume Data Management: Supports AI-related tasks through an optimized distributed node network, ensuring that large-scale data can be accessed and indexed.

Built-in Attribution: Smart contracts associate data usage with specific contributors and automatically allocate rewards based on Proof of AI.

Monetization opportunity: Data providers can set pricing plans or conditions, controlling the timing and manner of data usage.

3.3 Composable AI ecosystem with customizable subnets

Kite AI supports customizable subnets - areas specifically designed for different AI workloads in Layer 1 architecture:

Governance Flexibility: each subnet can implement unique governance rules, token economic models, or consensus parameters to adapt to specific use cases.

Modular infrastructure: Developers can combine multi-mode AI workflows by integrating subnets focused on data planning, model training, or agent deployment.

Isolation and Security: Malfunctions in one subnet will not affect other parts of the network, enhancing overall stability.

3.4 Decentralized, Portable AI Memory

AI models typically need persistent storage of learned parameters and interactive memories. Kite AI's decentralized, portable AI memory provides:

Privacy Protection: Sensitive model parameters can be encrypted to ensure the protection of intellectual property even in a distributed environment.

Long-term model traceability: Ownership and version history of the model will be recorded on the chain to ensure transparency and reproducibility.

Scalability: supports billions of interactions, built-in tracking and attribution mechanisms for recording each model update or inference.

4. Analysis and Evaluation

4.1 Fair Attribution

By leveraging Proof of AI, Kite AI is able to allocate rewards proportionally to the impact of contributions. Shapley values or other alliance-based allocation frameworks have been integrated into the consensus logic, allowing:

Fine data contribution score: assess the impact of each data subset on model performance.

Transparent model evaluation: on-chain audit of model training process to verify real improvements in accuracy or utility of the model.

Agent Monitoring: Track the use of agents and associate consumer payments or on-chain transactions with specific agent outputs.

Proof of AI focuses on marginal contributions, cultivating a systematic mechanism that rewards quality rather than quantity, reducing free-riding problems, and minimizing the occurrence of repetitive or low-value contributions.

4.2 Scalability and Throughput

The requirements of AI workflows, especially when dealing with large-scale datasets, have brought unique scalability challenges to blockchain. Kite AI addresses this issue by:

Deploying subnets: dividing tasks and resources into dedicated areas reduces congestion and supports parallel computing.

Layered Architecture: Unloads complex computations to validators or oracles in specific subnets, while on-chain transactions record key metadata for attribution and reward allocation.

This architecture promotes horizontal scalability, and independent subnets can be expanded as needed. However, the actual throughput still depends on node infrastructure, bandwidth, and governance decisions within the subnet.

4.3 Governance and Security

Maintaining security through the detection and expulsion of malicious actors via Proof of AI, while governance is delegated to subnet-level authorities and token holders:

Stakeholder Alignment: Subnet governance tokens ensure that those who contribute resources or expertise can participate in decision-making.

Cross-subnet coordination: Layer 1 consensus rules unify subnets, preventing fragmentation or incompatible protocols from appearing.

Anti-attack: The incentive mechanism design of Proof of AI reduces the sensitivity to Sybil attacks and data pollution, by dynamically weighting contributions based on actual utility to mitigate these threats.

The governance based on Proof of AI better aligns stakeholders' incentives than traditional PoS frameworks, although emerging threats (such as advanced data pollution strategies) still require continuous monitoring and updated detection algorithms.

5. Use Cases and Potential Impacts

5.1 Data Market

Kite AI's decentralized data engine provides a secure, transparent data trading platform. Data owners can confidently share datasets - from medical images to autonomous driving logs - knowing that they will be compensated and able to control their assets.

5.2 Collaborative Model Training

AI research groups and enterprises can use Kite AI's subnet to collaborate on model development. The improvement of the model will be tracked on-chain, and each contributor's efforts in hyperparameter tuning, data cleaning, or fine-tuning will be directly attributed and compensated.

5.3 Decentralized Agent Ecosystem

AI agents running tasks such as content moderation or financial predictions can be deployed within a subnet and interact with end users through smart contracts. Proof of AI ensures transparent evaluation of the utility and performance of each agent, simplifying reward mechanisms and promoting cross-agent collaboration.

6. Conclusion

Kite AI's design philosophy recognizes the complexity of the AI pipeline and encourages high-quality contributions through a multi-level incentive mechanism to suppress malicious behavior. However, there are still some open issues, including:

Adoption and Network Effects: The success of any blockchain-based ecosystem relies on the aggregation of critical mass. Accelerated adoption may require attracting data providers and developers through strategic partnerships and incentive measures.

The complexity of attribution: While PoAI introduces advanced valuation methods, the real AI pipelines are often dynamic and nonlinear, and the attribution framework still needs continuous improvement.

Regulatory considerations: Privacy laws and intellectual property laws vary by region, which may affect the way data and model ownership is enforced on the chain.

Iterative improvement and a robust governance model will be crucial for the long-term success of Kite AI in addressing these challenges.