The question of trust is never far from the implementation of AI, especially in such a significant area as generative AI. At Transform 2023, Hillary Ashton, CPO of Teradata, dived into ways companies must address the complexities of data governance, privacy and maintaining transparency to ensure trust in their AI-powered operations.
“Generative AI has brought the whole advanced analytics arena into the boardroom as a discussion, whereas before it was a bit more of a back-office discussion, with a lot of untapped potential,” said Ashton. “It’s an exciting space, but also something people need to think carefully about from a trust perspective. Generative AI is really putting the focus on what is protected data, what is PII. How do you want to treat that data, especially with large language models and generative AI?
The role of trust in the adoption of generative AI
Companies have to be inherently trustworthy to their customers, Ashton said, and it all starts with the quality of the underlying data and a clear separation of PII data.
“It sounds very risky, but it’s very difficult for many companies to do it,” he added. “They need the technology, the people and the processes to be able to do that.”
Companies need to establish robust data governance structures, treating data as a commodity and ensuring that clean, non-PII data is made available to users. Compliance with regulatory compliance is key, with organizations being transparent about their use of generative AI and its impact on data privacy. It is also imperative to safeguard intellectual property (IP) and protect proprietary information when partnering with third-party vendors or using LLMs.
“This is where I come back to having a clear understanding of how you want to use advanced analytics – do you need to figure out what is not just your PII, but what your IP is?” she said. “How are you protecting him? You might consider that if you have your senior data scientist writing the prompts, this is your IP as an organization. It’s not IP you want to give away for free to a competitor. Maybe it’s not explicitly stated when we think about things like PII data. Now the prompts become your IP. You now have a whole new legal practice in timely protection and intellectual property.
This also includes how you have chosen to structure your data, which is highly proprietary. If you are Bank A, competing with Bank B, you don’t want to actually give your competitor an edge with a vendor that uses an LLM based on your data structure, no matter how sanitized it is.
From there, he said, it’s about “making sure you understand what market you’re in, what regulatory compliance is like, and building with that end state in mind, and then being transparent with your customers about how you’re using gen.” AI, as you are not using gen AI, so it can be trusted.
Trust isn’t limited to privacy either; extends to the reliability and accuracy of model results. Regular evaluation of models and proactive measures to correct underperformance are essential to maintain user trust.
So where to start?
Getting started means working backwards from the results you want to achieve, Ashton says, and they fall into two categories. The former is an area where you already have advanced capabilities and want to maintain that leadership edge with advanced analytics. The second is taking on challenges at the table that maybe the competition has but you don’t.
From there, the considerations are data governance and respecting IP and PII data sovereignty. The second piece is being able to do this at scale and finally the last piece is model operations i.e. managing models as they go into production and understanding when a model starts to underperform or returns unacceptable results .
And finally it is essential not to get lost in the novelty, but to evaluate the ROI and the price trend.
“We are all super excited about LLMs and gen AI in general,” Ashton said. “The cost of doing some of this could become prohibitive over time for use cases that aren’t high-value. Make sure you’re not solving something that could be done with a PivotChart BI with an LLM just because it’s cool – it sounds like a crazy example, but it’s not that crazy.