#206 Securing GenAI

In this episode, Darren continues his interview with Steve Orrin, the CTO of Intel Federal. They discuss the paradigm shift in DevSecOps to handle Artificial Intelligence and the dynamic nature of application development that AI requires.We find the transformative power of Digital Transformation, DevOps, and Artificial Intelligence (AI) at the fascinating intersection of technology and business leadership. In this realm, we will delve into two crucial aspects: the significance of securing the AI development process and the imperative of responsible and ethical data use. By understanding these, we can harness AI's potential to not only revolutionize our organizations but also inspire trust and confidence, driving digital transformation to new heights.  Ethical Data Sourcing and AI TrainingAI has revolutionized the way we engage with technology. The crux of every AI system lies in data diversity. Why? Because an AI system learns from data, feeds on data, and performs based on the information provided. The more diverse the data is, the better the AI system learns and performs. However, the ethical aspect of data sourcing and AI training must be considered with utmost urgency. The AI system must be deployed only on populations that align with the datasets used in the training phase. The ethical use of AI involves deep trust and transparency, which can only be garnered through thorough visibility and control throughout the AI's development lifecycle. The Golden Rule: TrustBuilding trust in AI systems is a direct result of their foundation on a diverse range of data. This approach prevents any single type or data source from dominating and diluting any biases that may exist in any dataset. The golden rule of trust in AI systems starts with diversifying data sources, thereby reducing undue dominance. In addition, data provenance visibility is integral to ethical AI. It provides transparency to the deploying institution, showing what information went into the AI's training and thus ensuring its optimal performance. Scalability and TraceabilityOne of the main challenges with AI development is managing the scalability of training data. The ability to rollback to well-known states in training is critical, but how do you do that with petabytes of data? Hash functions or blockchain methods become essential in managing large data pools. Traceability, accountability, and audibility also take center stage in the AI development process. In the case of untrustworthy data sources, a system that enables data extraction from the pipeline is necessary to prevent their usage in ongoing training. The Road AheadThe journey to secure AI development is guided by the principles of transparency, trust, and ethics. These are not mere suggestions, but essential elements in fostering trust in AI systems while ensuring their effectiveness. The path may seem challenging, but these steps provide a clear roadmap to navigate the complexities of AI DevSecOps.Be it through diverse data sourcing, treating data with the respect it deserves, or consistently documenting the data lifecycle process, the principles of trust, visibility, and a dogged commitment to ethical practices lie at the heart of burgeoning AI technologies.