At Polkadot Decoded 2024, Joshua Waller, the Director of Technical Evangelism at Phala Network, delivered an insightful presentation on Phala Network’s role as the execution layer of Web3 AI. The talk detailed Phala’s journey from its inception to becoming a pivotal player in the AI and blockchain ecosystem, showcasing its innovative AI co-processor capabilities.
Historical Context and Evolution of Phala Network
Waller began by tracing the history of Phala Network, highlighting its consistent delivery of products since 2017. The network’s white paper was released in 2019, followed by the mainnet launch in late 2021 and the parachain launch on Kusama and Polkadot in 2022. Phala successfully bootstrapped its protocol with over 30,000 attested Trusted Execution Environment (TEE) nodes, now surpassing 36,000 nodes.
Phala initially focused on “fat contracts,” which enhanced on-chain smart contracts with secure off-chain computation, enabling cross-chain calls and extending to external APIs. This foundation laid the groundwork for Phala’s current focus on AI co-processors, leveraging the network’s middleware to build intelligent applications on blockchain.
Phala AI Co-Processor: High-Level Architecture and Applications
The core of Waller’s presentation revolved around Phala’s AI co-processor and its architecture. The AI co-processor enables the deployment of AI agent contracts, which can function as sophisticated HTTP servers without requiring an account or cryptocurrency. These agents can handle get and post requests, connect to various data sources, and integrate with AI models to perform complex tasks.
Phala’s AI agent contract system consists of two components: the agent DAO and the agent script. The agent DAO regulates the scripts, enabling monetization through bonding curves and providing a framework for developers to build and deploy robust AI-driven applications.
Demonstrating Phala’s AI Agent Contracts
Waller showcased several practical examples of AI agent contracts, including:
- Social Chain AI: A social engineering white-hat agent that used AI to generate convincing phishing messages.
- Migorithms: An application that allows users to build and choose their own web3 social algorithms.
- DevDo: An IDE extension that integrates AI to assist developers with code generation and rewards them with cryptocurrency.
- Jarvis: A multi-agent system for intent-based trading, using Phala’s co-processor to connect with various AI models and provide market predictions.
- Warp Ads: An application bringing Facebook-style ads to web3 social platforms, leveraging AI to generate ad labels and verify transactions on-chain.
These examples illustrated the versatility and potential of Phala’s AI co-processor, highlighting its ability to facilitate innovative solutions across different use cases.
Phala Network Addressing Centralized AI and Future Prospects
Waller concluded by addressing the challenges of centralized AI and proposing a decentralized model that ensures equal access and confidentiality. He emphasized the importance of creating a verifiable and confidential supply chain for AI models, a concept Phala has already demonstrated with its GPU TEE for confidential computing.
Phala’s vision includes developing a global, decentralized AI model where users and companies can contribute data securely, leading to the creation of superior models. This approach aims to democratize AI, moving away from the current dominance of centralized entities.
Phala Documentation and Developer Support
For developers interested in building on Phala Network, Waller directed them to the comprehensive documentation available at docs.phala.network. He encouraged developers to reach out for support and provide feedback to enhance the product and developer experience.
Joshua Waller’s presentation at Polkadot Decoded 2024 underscored Phala Network’s commitment to innovation and its pivotal role in advancing the integration of AI with blockchain technology. Through its AI co-processor and agent contracts, Phala is poised to drive the next wave of intelligent, decentralized applications.
