Blockchain for AI

Authors: Rajarshi Gupta, Vijay Dialani

Published on: March 8, 2024

Abstract: Blockchain can bring many potentials to the AI ecosystem, and this blog introduces some specific opportunities and use cases in this field. Coinbase is a leading company dedicated to advancing blockchain-based digital assets and decentralized platforms, and we are also keen to collaborate with participants in the AI field to jointly realize these emerging opportunities.

Over the past year, we have witnessed an explosive growth in the capabilities and applications of AI. This includes significant advancements in text-to-image models, the maturity of large language models, and their applications in various business use cases such as search and recommendation, facilitating software development, and big data analysis. It is predicted that by 2032, the generative AI market alone will reach a market size of $1.3 trillion, with a compound annual growth rate (CAGR) of 42% over the next decade.

When we explore the intersection of blockchain and AI, there are two main aspects to consider. The first is the use of AI/ML models and methods to enhance blockchain platforms, decentralized applications, and the trading of digital on-chain assets. The second, and the main focus of this article, is leveraging the capabilities of blockchain to provide value to developers and users of AI/ML products and services.

Fundamentals of Blockchain Capabilities

There are many fundamental blockchain capabilities available for the development, deployment, and operation of AI models. These capabilities encompass cryptographic primitives, blockchain protocols, and smart contracts. They possess the following fundamental properties, which are highly relevant to AI use cases, and we will discuss them in the following sections:

• Data Security: Helps AI store data in a tamper-proof and immutable manner, achieving high availability through decentralized servers that are less susceptible to attacks, manipulation, censorship, and denial of service.
• Data Auditability, Traceability, and Verifiability: Records transactions and assets in an immutable and transparent manner, facilitating the tracking of data provenance, ownership, and origin, along with digitally signed agreements and timestamps. This also provides auditable and verifiable capabilities.
• Decentralized Decision-Making: Enables decisions to be made by multiple entities or directly between parties without an existing trust relationship among them or with a central entity.
• Autonomous and Transparent Code Execution: Allows programs to execute as smart contracts that are transparent to all relevant parties and can operate autonomously without relying on trusted and centralized intermediaries.
• Decentralized Identity: Provides secure digital identity mechanisms that allow users to interact with services without compromising their privacy.
• Micropayments: Offers a secure and lightweight payment method that reduces friction in payment processes.

1. How Blockchain Can Benefit AI

As shown in the figure above, the AI ecosystem has a diverse set of stakeholders who interact with each other regarding data, models, and computational infrastructure. Due to administrative and economic boundaries, there is a certain separation among these potential stakeholders, leading to trust and payment issues that need to be addressed, and blockchain can help solve both problems.

We can categorize the potential advantages of blockchain for AI and the corresponding products and services that can be developed into four main categories, as illustrated in the figure below. In the following sections, we will discuss these categories one by one.

2.1. Data and Model Integrity

Blockchain can be used to develop solutions that help users and developers ensure that data and models are not modified without their knowledge. For example, API-based services can allow data owners and AI developers to record timestamped hash values of datasets and models to ensure their integrity, and document the entire process of model development and the datasets used, allowing for third-party audits or regulatory oversight. This system could even be directly integrated into ML development tools, such as Pytorch. This can help improve the integrity and trustworthiness of models by making the development process more transparent and secure. It may also be possible to record proof of relevant "abandoned" specific data on the blockchain to demonstrate to regulators that certain provider data has been removed from a given model. Recording hash values of on-chain data and model outputs also helps combat deepfakes. For example, applications may be able to ensure the authenticity of the data used by checking digital signatures associated with on-chain data sources, or a decentralized version of "Snopes.com" could be designed and implemented on the blockchain to flag deepfakes.

2.2. Data and Model Usage and Access Management

Non-fungible tokens (NFTs) can prove ownership of any given digital content or data. Depending on the use case, the content involved can be model inputs, such as prompts for generative AI tools, data used to train models, model parameters, or model outputs. NFTs would allow users or developers to maintain their ownership and further transfer ownership of the corresponding digital assets to others. A blockchain-based data and model access control mechanism could also be envisioned, such as smart contracts that allow/restrict access based on a given list of user addresses. Alternatively, it could integrate with decentralized identity solutions (possibly using advanced cryptographic techniques like zero-knowledge proofs) to allow access based on certain proven attributes (e.g., allowing access based on proof of user age or only from certain geographical locations) while protecting user privacy.

2.3. Incentives and Payments for Data, Models, and Computational Resources

Blockchain can use stablecoins for low-fee micropayments for using generative AI models. Smart contracts can allow for decentralized revenue sharing among multiple co-owners of a model. This co-ownership model, in essence, is a "decentralized Hugging Face," which may allow small to medium-sized model developers to band together to compete with larger companies in the field. It can also be used to incentivize data providers, data annotators, model developers, or human feedback providers worldwide to join a new decentralized project to develop new generative AI models or solutions, with appropriate mechanisms to track contributions so that incentives can be fairly distributed. Blockchain can also be used to create a decentralized data/model/computational marketplace that enables computational providers, training data providers, model developers, and users to easily search and match with each other, providing incentives, payments, and contracts and agreements. A blockchain-based decentralized review system implemented using smart contracts can incorporate both automated and human reviewer-based systems to incentivize high-throughput, thorough, and high-quality reviews of data and models.

2.4. On-Chain Deployment of AI

This category involves running certain AI models directly on the blockchain for greater transparency and trust. AI models may directly provide some inference or generation use cases to end users, assuring them that only the models they intend to provide input to can receive the said input and generate the output they see, without manipulation, forgery, or censorship. Alternatively, AI models could be deployed to help smart contracts adjust and optimize their parameters in response to user transactions. AI models could also be a smart contract that makes buy/sell/trade decisions on digital assets based on historical and current transaction data on-chain to profit. These models can be deployed as smart contracts on Layer 1 chains or through Layer 2 systems (such as zk-rollups). These models may be privately owned or decentralized in the form of a DAO, allowing multiple individuals and entities to have "shares" in a given chain model. In the long term, for such applications, given the high demands of AI applications on data and computation, there may be interest in exploring and developing an entirely new platform to support AI workflows.

3. How We Can Help

Coinbase's mission is to empower economic freedom for over a billion people. As crypto usage grows, we focus on building the most trusted and compliant products and services and supporting other builders. It is clear that AI blockchain aligns with this strategy, which is to enable individuals and organizations becoming part of the emerging generative AI ecosystem (currently almost entirely based on centralized and opaque Web2 frameworks) to benefit from blockchain and chain-based crypto solutions, in a broad sense.

We believe Coinbase is particularly well-positioned to be a major contributor in this space because a) it has significant brand recognition and influence among both retail customers and institutional users of blockchain assets and services; b) it has a strong track record of helping bridge the gap between the emerging Web3 world and existing Web2 systems; c) it has a powerful ML development team with a deep understanding of the growing generative AI ecosystem and the needs of developers and other stakeholders.

Coinbase is keen to explore partnerships and integrations with like-minded companies, such as those with expertise in AI solutions and ecosystem services, and are committed to turning ideas into reality.

Acknowledgments

We would like to thank Dr. Bhaskar Krishnamachari from the University of Southern California for his contributions to this article. Dr. Krishnamachari is a paid advisor to Coinbase and assisted in writing this article in that capacity.

Related Content

• Coinbase Institute White Paper: Blockchain and Artificial Intelligence (AI): Complementary Technologies That Can Improve Each Other, Fall 2023.
• David Duong, "The Intersection of AI and Crypto: What Common Issues in Generative AI Can Blockchain Technology Solve?", Coinbase Research, May 2023.
• Use Cases of AI in Blockchain, Chainlink Blog, May 2023.
• Steve Vassallo, "AI x Blockchain: Reaching New Heights", Forbes Digital Assets, June 2023.
• Salah et al., "AI Blockchain: A Review and Challenges", IEEE Access, 2019.
• Tian et al., "Blockchain for AI: A Disruptive Integration", IEEE CSCWD 2022.
• Karger et al., "AI Data Blockchain: Current Status and Open Research", ICIS, 2021.

Author Biographies

Here is a brief introduction to Rajarshi Gupta and Vijay Dialani.

Rajarshi Gupta is the Head of Machine Learning at Coinbase, providing intelligent automation and protection for crypto users worldwide. Prior to this, Rajarshi was the General Manager of ML Services at AWS. He also worked for many years in Qualcomm's research division, where he created "Smart Protect," the first product to implement machine learning for security on-device, delivering over a billion Snapdragon chipsets. Rajarshi holds a Ph.D. in EECS from the University of California, Berkeley, and has very unique and rich expertise at the intersection of AI and blockchain. Rajarshi is a prolific inventor with over 225 issued US patents.

Vijay Dialani leads the Machine Learning Risk and Machine Learning Platform teams at Coinbase. He has led teams of machine learning researchers and engineers at Twitter, Apple, Google, and Microsoft. His research has been published in WWW, CIKM, and ICDE. He has published over 20 papers, cited more than 550 times, and filed over 17 patents. During his academic career, he created graduate courses in data science, machine learning, and cloud computing. He has mentored several graduate students and served on the graduate committees of several master's and doctoral students.