Title: Navigating Data Governance, Transparency, and Trust in the World of Generative AI
The implementation of artificial intelligence (AI) brings forth the issue of trust, especially when it comes to emerging technologies like generative AI. Hillary Ashton, Chief Product Officer at Teradata, sheds light on the complexities businesses face in navigating data governance, privacy, and maintaining transparency to ensure trust in their AI-driven operations during Transform 2023.
Generative AI has brought advanced analytics to the forefront of boardroom discussions, where it was previously limited to back-office conversations. Ashton emphasizes the importance of approaching this exciting space with caution, particularly from a trust perspective. With generative AI, it becomes crucial to determine how protected data, particularly personally identifiable information (PII), should be treated, especially in the context of large language models (LLMs) and generative AI.
To establish trust, enterprises must uphold their reputation for being trustworthy to their customers. Ashton emphasizes the need for a strong foundation of data quality and a clear separation of PII data. While these may seem like basic requirements, implementing them can be challenging for many enterprises. They require the right combination of technology, skilled personnel, and well-defined processes.
Robust data governance frameworks are essential for enterprises to treat data as a valuable product and provide clean, non-PII data to users. Adhering to regulatory compliance becomes paramount, and organizations must be transparent about the use of generative AI and its impact on data privacy. Safeguarding intellectual property (IP) and proprietary information is also crucial when collaborating with third-party vendors or utilizing LLMs.
Ashton highlights the significance of understanding how advanced analytics are used within an organization and the protection of not just PII but also IP. For instance, the prompts used by a senior data scientist to train an LLM become proprietary IP for the organization, valuable information that should not be given away freely to competitors. This necessitates new legal practices surrounding prompt protection and IP.
Even the structure of data becomes highly proprietary and competitive. Banks, for example, would not want to give their competitors an advantage by sharing their data structure, regardless of how sanitized it may be. Taking into account regulatory compliance and being transparent with customers about the use of generative AI are critical steps in building trust.
Trust extends beyond privacy; it encompasses the reliability and accuracy of model outcomes. Regular evaluation of models and proactive measures to rectify underperformance are essential in maintaining user trust.
Starting with clearly defined desired outcomes is crucial, according to Ashton. She categorizes these outcomes into two buckets: maintaining leadership advantage with advanced analytics in areas where a company already excels, and addressing fundamental challenges that may exist in competitors’ operations but not in one’s own.
Data governance and respect for IP and PII data sovereignty are vital considerations. The ability to execute these principles at scale is equally significant, as is managing the models efficiently as they go into production, ensuring they do not underperform or yield unacceptable results.
Finally, Ashton advises evaluating the return on investment and price performance when exploring the use of generative AI. While the technology is exciting, the cost of running certain applications may become prohibitive for low-value use cases. It is essential to avoid using LLMs unnecessarily when simpler solutions, such as a basic business intelligence pivot chart, can effectively fulfill the requirements.
As the world embraces the possibilities of generative AI, navigating data governance, transparency, and trust becomes paramount. By prioritizing foundational data quality, respecting privacy regulations, safeguarding IP, and continuously evaluating model performance, businesses can foster trust and successfully harness the power of generative AI.
