Casper Labs and IBM Consulting announced they will work to help clients leverage blockchain to gain greater transparency and auditability in their AI systems. Together, Casper Labs and IBM Consulting plan to develop a new Casper Labs solution, designed with blockchain and built leveraging IBM watsonx.governance, that establishes an additional analytics and policy enforcement layer for governing AI training data across organizations.

The process of training, developing and deploying generative AI models happens across multiple organizations, from the original model creator to the end user organization. As different organizations integrate new data sets or modify the models, their outputs change accordingly, and many organizations need to be able to track and audit those changes as well as accurately diagnose and remediate issues. Blockchain can help organizations share their trusted context information via metadata in the ledger documenting that the models have changed while mitigating the risk of intellectual property crossover or unnecessary data sharing across organizational lines.

Casper Labs' solution is planned to be built on Casper, a tamper-resistant and highly serialized ledger, and leverage IBM watsonx.governance and watsonx.ai to monitor and measure highly serialized input and output data for training generative AI systems across organizations. Thanks to the Casper Blockchain's hybrid nature and permissioning system, organizations can expect to be able to better protect sensitive data stored in the solution from being accessible to external actors; they have control over who can access what data. The solution will also be built to support version control using the serialization capabilities of blockchain, so organizations can efficiently revert to previous iterations of an AI system if performance issues or biased outputs occur.

IBM Consulting's AI governance and technology experts will assist Casper Labs in building the solution, which Casper Labs expects to be available for clients in beta in first quarter 2024 and later available more broadly in their channels and in the IBM Cloud Marketplace.

"An AI system's efficacy is ultimately as good as an organization's ability to govern it," said Shyam Nagarajan, Global Partner, Blockchain and Responsible AI Leader at IBM Consulting. "Companies need solutions that foster trust, enhance explainability, and mitigate risk. We're proud to bring IBM Consulting and technology to support Casper Labs in creating a new solution offering an important layer to drive transparency and risk mitigation for companies deploying AI at scale."

The new solution is planned to help companies across industries, including financial services, healthcare, and retail, deploy AI responsibly at scale across their ecosystem of technology and services providers. Among other features, the solution aims to offer:

• Compliance Dashboard: A centralized dashboard for monitoring and managing AI systems as they're applied across organizations to support their compliance processes with an organization's ethical guidelines.
• Quality Control Toolkit: Tools for monitoring the quality and performance of AI systems, along with an interface to enhance the transparency and explainability of AI outputs.
• Version Control: The ability to correct for performance or other issues by "rolling back" to previous iterations of a given AI system that didn't display issues.
• Audit and Reporting System: A system for auditing AI processes and generating detailed reports based on context metadata captured by Casper Labs' ledger.

"While generative AI has justifiably excited organizations for its transformative potential, its practical applications have been severely limited by an inability to monitor and react to the data feeding AI systems," said Mrinal Manohar, CEO at Casper Labs. "With IBM's help, we're committed to delivering a better way to not only understand why AI systems behave the way that they do but also a clearer path to remediate behavior if hallucinations or performance issues occur. AI's long-term potential will be dictated by how effectively and efficiently organizations can understand, govern and react to increasingly massive AI training data sets."