Verifiable AI describes the practice of making an AI model auditable through cryptographic verification of the training data and business logic on a public blockchain. Verifiable AI is part of the discipline of Explainable AI (XAI) that serves to increase transparency and trust by uncovering potential biases within AI training data and AI model logic. Verifiable AI is crucial for an organization in building trust and confidence when putting AI models into production and also helps organizations adopt a responsible approach to AI development.
Blockchain technology enables organizations to cryptographically verify the training data and business logic used by an AI model, allowing 3rd parties to audit to establish trust. Blockchain technology also enables organizations to fight AI deep fakes and theft of intellectual property through data provenance.
Data Provenance One of the key benefits of blockchain technology is its ability to provide unparalleled data provenance. Storing data on a blockchain can ensure data integrity, increasing trust and transparency. AI models that leverage large datasets can increase transparency and trust with data provenance solutions enabled by blockchain.
Zero-Knowledge Attestations Polygon's significant investment and breakthroughs in zk technology can allow users and applications to make privacy safe attestations to verify data and contribute to datasets used by AI models. This can improve the quality of datasets and potentially reward users who make relevant contributions, while preserving privacy.
Authenticity Verification The capabilities of deep learning models, such as DALL-E, Stable Diffusion, and Midjourney, have highlighted the profound potential of generating images and different forms of media purely based on natural language text prompts. Blockchain technology can help validate the authenticity of images, video files, text documents, or other types of media by being able to cryptographically verify where a piece of content originates from and whether it has been tampered with or altered in any way. In addition to combating IP theft and deep fakes, blockchain technology enables a primary and secondary marketplace for digital media, creating economic opportunity for artists.
Transparency A challenge presented by current deep learning models is the lack of transparency in their decision-making processes. Due to the immense complexity of these models, which sometimes involve hundreds of billions of parameters, even experts can struggle to explain why a particular model generates a specific output when prompted with a specific input. By facilitating a transparent record of data, blockchains can enable AI models to provide a clear framework for their operations. This allows for the analysis of audit trails on the decision-making patterns of algorithms, and the use of an immutable data ledger to reveal what data the models are relying on, ultimately helping to contribute to greater integrity of the recommendations generated by AI models.