What is PIN AI, which has received tens of millions of dollars in funding and has been selected for the a16z and Stanford Crypto Accelerator?

Original Title: Introducing PIN AI: The Open Platform for Personal AI

Original Authors: Davide Crapis, Bill Sun, Ben Wu, Co-founders of PIN AI

Compiled by: Scof, ChainCatcher

AI infrastructure company PIN AI today announced the completion of a $10 million pre-seed funding round, with participation from notable VCs and angel investors including a16z CSX, Hack VC, and Blockchain Builders Fund (Stanford Blockchain Accelerator). PIN AI is developing the world's first open-source Personal Intelligence Network (PIN), enabling AI developers to provide very everyday and practical AI services such as shopping, organizing travel, and planning finances.

This article is a systematic introduction to the project co-authored by the three co-founders of PIN AI, compiled by ChainCatcher.

In the past year, with the widespread adoption of ChatGPT and similar AI tools showcasing impressive capabilities, the blockchain industry has increasingly focused on how to combine cryptographic primitives with AI. The Crypto × AI field has emerged. On the infrastructure side, many projects have focused on providing services such as decentralized computing, data availability/storage, and verifiable reasoning. On the application side, experiments like AI prediction markets, trading agents, and DAO bots have emerged. The vast majority of applications focus on helping users understand and interact with web3 protocols. While this is an important and growing use case, it still has certain limitations.

Meanwhile, the AI industry is thriving. Living in the Bay Area, we have witnessed the excitement and vitality brought by the construction of the AI field. Many friends, including excellent engineers and scientists, have left their full-time jobs to start AI projects. On the infrastructure side, there are already some successful projects providing new and necessary services, such as vector databases, evaluation services, and model routing services. However, on the application side, despite much hype, many well-funded projects still struggle to find market fit and achieve widespread adoption. The core issue is that AI applications require contextual data to provide useful reasoning services, and without this data, it is nearly impossible to offer services that are more useful than the latest version of ChatGPT.

Furthermore, large AI labs rapidly absorb high-quality innovations from open-source builders, effectively stifling or absorbing them at the inception of new projects, while large tech companies attempt to monopolize access to AI services.

We are building PIN AI from a different path: creating an open AI network that accesses vast contextual data, enabling AI builders to create a variety of useful AI applications. PIN is rooted in open-source AI and Ethereum, with three foundational layers: personal data (focusing on privacy and data ownership), personal AI (a practical and trustworthy companion AI on personal devices), and external AI (an open AI service marketplace).

A Practical AI Application Platform

Our goal is to enable AI developers to provide practical AI services, such as purchasing groceries or other retail items, organizing the next trip, and planning financial actions. Today, these services are not yet achievable, because external AIs like ChatGPT lack the necessary user context, history, and preferences. In contrast, our personal AI possesses the full context of its human owner, acting as an assistant and mediator, calling upon more powerful external AIs in the cloud to fulfill user intentions while always protecting user privacy. Our network can realize the aforementioned services and more. In the future, personal AIs will also be able to collaborate to achieve their owners' common goals.

To achieve this, personal AIs need to have the ability to access a wide range of owner data and the capability to trust and pay for services when outsourcing to external AIs. The interaction between personal data and external AIs determines the fundamental principles of our protocol design. Personal AIs reside on user devices, but due to system limitations and the increasing volume of personal data we generate, rich user context cannot fully reside on that device. Personal AIs need to store personal data on secure storage devices, encrypted and made available to external AIs on demand.

Secure storage solutions need to ensure: (1) the privacy of personal data, and (2) user data ownership and control. The ideal solution is decentralized storage with data accessibility/availability mechanisms. Web2 cloud solutions are at a disadvantage on various dimensions: privacy protection depends on the provider's internal policies, which are not robust, the provider owns user data, and cannot guarantee high-frequency access to data for AI users, as there are no inherent mechanisms to prevent witch attacks and spam. If paired with confidential computing (TEEs) and client-side encryption, a certain degree of service centralization can be tolerated, which is a solution we are currently exploring.

On the other hand, the interaction between personal AIs and external AIs may involve the exchange of personal data, which also needs to have similar attributes. For example, unless external AI computations are private and we can ensure that external AIs cannot "remember" private data, private user data cannot be disclosed to external AIs. The best way to ensure privacy and fairness when outsourcing services to external AIs is through auditable exchanges, such as exchanges implemented on public blockchains, rather than on the servers of private service providers like Apple.

The primary goal of the PIN protocol is to achieve an open ecosystem for personal AI applications. We will focus on creating an infrastructure layer to acquire user context data and match user intentions with specialized external AIs, while always protecting user privacy. This also provides the best resource allocation efficiency, as the historical performance of models can be monitored and evaluated, minimizing AI MEV (Maximum Extractable Value).

PIN Protocol

The PIN protocol is the pillar of the open-source ecosystem built around PIN AI. It provides minimized trust activity tracking and value exchange, access to valuable personal data, and an open innovation platform for new AI services. We need to guide the three key components of the PIN economy as follows.

1. Data Connectors and On-chain Registry: Track and verify user data sources connected to the network. Utilize state-of-the-art zero-knowledge cryptography (such as zkTLS) to verify data sources and employ machine learning techniques to update the true user data contributions in the registry at the end of each cycle.
2. Private Storage and Computing Layer: Provide secure and private data storage that exceeds the capacity of user devices. Use permanent storage solutions and data access mechanisms to store vast amounts of user data (including photos, videos, etc.) and make the most relevant data available for personal AI at any time.
3. Agent Links and Intent Marketplace: Open the realization of user intentions to AI agent applications. Build a permissionless agent registry and exchange mechanism, where the registry tracks performance metrics, and the exchange mechanism specifies how personal AI queries match agents based on each agent's cost, performance, and quality. To guide the development of this component, we will assist developers in converting current Web2 applications (like Amazon, DoorDash, Uber, etc.) into agent services, which we call Agent Links.

PIN AI

In terms of AI, PIN AI introduces a sophisticated architecture designed to balance privacy, performance, and personalization through hybrid models. These models combine processing on devices with cloud-based computing, ensuring that sensitive operations remain under user control while leveraging powerful cloud models to accomplish complex tasks. This hybrid approach allows AI to continuously learn from the Personal Index. The Personal Index is a dynamic knowledge graph that organizes user data from devices and the cloud, providing contextually relevant personalized responses.

The key innovation of PIN AI lies in its focus on privacy and context management. The platform integrates BERT-based models specifically designed to detect and anonymize personally identifiable information (PII) at various stages of data processing, ensuring confidentiality and compliance with data protection regulations. Additionally, RAG (Retrieval-Augmented Generation) and GraphRAG models enhance the AI's ability to retrieve relevant information and perform complex reasoning by leveraging structured knowledge graphs, ensuring that responses are both context-rich and personalized to meet user needs.

In the future, PIN AI will implement autonomous workflows for fine-tuning. Personal AI models will continuously fine-tune, with device models adapting in real-time to user preferences, while cloud models aggregate anonymized data to enhance broader capabilities. By leveraging unique personal data and user patterns, PIN AI will push the frontier of AI personalization, providing each user with a truly personal AI that becomes increasingly intelligent and personalized over time.

Guiding the PIN Economy

At the core of PIN is a bilateral market where users and their personal AIs can obtain services from external AIs. As the number and value of user context data opened to these services increase, the services that can be provided also increase. Data connectors and on-chain registries are part of our Proof of Engagement (PoE) protocol, which aims to incentivize users to connect data and perform valuable transactions on the platform. This protocol has two main components:

• Data Connection Incentives: Users can receive allocations simply by connecting their personal data to the network. Privacy is ensured through user-level encryption. Resistance to witch attacks is achieved through the following combination: cryptographically authenticated data sources (zkTLS), machine learning techniques for assessing human digital footprints, and digital identities (e.g., WorldID).
• Proof of Valuable Transactions: When users complete provable valuable transactions through agent links, they can receive allocations from the protocol. Transactions need to transfer economic value (in cryptocurrency or fiat currency) and must be provable on-chain (e.g., proof of fiat currency payment via zkTLS).

Ecosystem Participants, Incentive Mechanisms, and Economic Flow

There are three main participants in the PIN ecosystem: end users, data connectors, and agent services.

• Users connect their personal data to the PIN network through data connectors, thereby receiving incentives while maintaining ownership and privacy of their data. This allows the PIN network to accumulate rich user data context, which agent services can leverage to better meet user needs.
• Data Connectors will initially also receive incentives as part of the PIN network's infrastructure. These connectors can be operated by third parties and ensure their security through staking and penalty mechanisms, with operators and stakers receiving rewards.
• The protocol will support the creation of new agent services and make their deployment easy through agent links. These agents will be able to utilize user context data to better serve user intentions and provide useful services. Operators of agent services will also be secured and incentivized through cryptoeconomic means.
• In the future, users will be willing to pay fees to access valuable agent services, while agent services will pay fees to the protocol for accessing personal data and user intentions, with these fees dynamically determined by market mechanisms based on supply and demand.

Image: Flow of Transaction Value in PIN

Conclusion

The vision of PIN AI is grand and requires the collective effort of a vibrant community of users and developers to bring it to fruition. With an excellent core team and secured funding, we are in a good position to embark on this transformative journey. As we prepare for the upcoming AI applications and the testnet of the PIN AI protocol, we are very excited to invite the community to join us in shaping the future of personalized, privacy-focused AI. Together, we can realize the full potential of this groundbreaking platform, delivering personal AIs that are fully owned and empowered by users.