P2P

72 P E E R T O P E E R : I L T A ' S Q U A R T E R L Y M A G A Z I N E | W I N T E R 2 0 2 4 Robust data governance entails documenting the pedigree of its sources, data types, how it was collected and processed, and intended usage. While this level of detail may be invisible to end-users, Gen AI applications can use it to present precise citations and direct links to source materials next to the AI-generated responses. Explainable AI (XAI) tools can also provide users (or regulators) with insights into the factors influencing a model's output, allowing for a deeper understanding of how the AI system generated its outputs. These tools can identify what specific data or source material had the most significant impact or influence on the model's production. This ability enhances contextual understanding of and confidence in how the model generated its results and could potentially illuminate any existing biases, leading to better-informed decision-making. • Responsible AI Use: Organizations need to fully understand and consider the real-world impact of Gen AI on employees, customers, partners, and end-users to prevent the creation or reinforcement of unfair bias, discriminatory outputs, or other harmful consequences. This process entails a deep understanding of how Gen AI models are created, trained, implemented, and used – by whom and for what purpose – and extensive internal planning. Data governance policies and procedures ensure that data used to train AI models is accurate, diverse, unbiased, and handled responsibly. This helps mitigate risks like privacy or copyright violations, discrimination, and inaccurate or misleading information. These are examples of ways organizations can leverage data governance policies to improve Gen AI output and protect their organizations. Other steps they can take to enhance their organization's Gen AI applications include: • Employing Retrieval-Augmented Generation (RAG) technologies to keep Gen AI focused on specific datasets for specific queries or use cases. Incorporating RAG allows the model to dynamically retrieve relevant information from a designated data source before generating a response, ensuring the output aligns with the specific context of the query and the specified dataset. • Establishing IT frameworks that enable the easy/ rapid switching of AI models without extensive retooling of applications or data, as well as regulatory frameworks to facilitate compliance and streamline reporting. • Collaborating with trusted partners that leverage proven tools, Gen AI-approved data, and licensed

• Cover