Presentation Material
AI Generated Summary
The speaker is discussing four pillars for controlling AI platforms, and ensuring their security and transparency. These pillars are:
-
Pre-establishing identity and workflow: This involves identifying people and systems involved in a project, , establishing certificates using Certificate Authorities, , and setting up workflows to ensure accountability.
-
Distributing tamper-evident verification: This pillar uses blockchain technology to create a distributed system that records all changes made to data, models, or projects. Each new block added creates a proper record, hash, and trial of what is happening.
-
Governance: This involves establishing a chain of authority to approve processes, from training data to model deployment. IBM has implemented its own governance team to oversee AI development.
-
Authenticity through user interactive interface: This pillar focuses on building an interface that provides end-users with confidence in the system’s integrity. The interface should provide evidence of tamper-evident verification and governance processes.
The speaker also discusses methods for implementing these controls, such as using crypto anchors for data set and model verification , establishing a governance group, and implementing on-chain governance. They mention that they are currently analyzing the effectiveness of using Hyperledger platform to implement blockchain technology in their system.
Some key points from the discussion include:
- The importance of establishing identities and workflows at the beginning of a project
- The use of blockchain technology for tamper-evident verification
- The need for governance and approval processes throughout AI development
- The importance of user interface in providing confidence in the system’s integrity
- The ongoing analysis of using Hyperledger platform to implement blockchain technology.