The transcript below is edited and condensed for clarity.
Lately, I have been trying to focus not just on the hype, but the operational nature of how to deploy generative AI. It's more about how we take advantage of these capabilities without necessarily having to get caught up in all the hype. How can I look at it from a more practical perspective? If we're looking at things like content management, for example, it's much easier for us to focus on what kinds of things we want to achieve productivity-wise rather than around the idea of replacing jobs. Let’s focus on productivity enhancements, which is where a lot of organizations are going today. As things evolve over time, we can start to take advantage of additional capabilities within the context of the business.
There’s a lot of low-hanging fruit within many organizations with very manual-type processes. Content generation as I mentioned earlier is just one example, but there are other business use cases because over time organizations collect data that can be consolidated and used to produce knowledge with generative AI solutions. All the time spent collecting, verifying and validating information can be faster in organizations. Clients also want the benefit of consistency, like highly accurate and automated content moderation.
I am getting the question a lot: “Does this mean I don’t need to worry about the data anymore?” But that couldn’t be further from the truth. Throwing everything into a data lake and then expecting generative AI to rationalize it is not what it was built to do. The emphasis needs to be on a well-curated dataset that is relevant to the question(s) you’re asking. We have to start with data governance, and if you’re not getting the results you expected, you need to go back and evaluate the data. Organizations may need to curate it better or even add more data for improved results. The saying “garbage in, garbage out” really applies to generative AI, and if you want strong results, you need to be intentional with what you put into the solution to ensure it is high quality.
Traditionally with machine learning and AI technologies, we’ve had data scientists being at the forefront of those decisions. But with generative AI today, the business people are really the ones determining the quality. We don’t need someone who is an expert at error rates that might be part of a training model — we need someone who knows the subject area and can determine if the particular response is correct. I’d call it the democratization of AI, where we’re looking elsewhere for expertise around these AI solutions.
With this, it’s important that organizations know what [the] outcome needs to be when we use the generative AI solution. With the right strategy of determining the business impact and the problem we want to solve, we can ensure the data is high quality and therefore the responses are what we are looking for in the end. Part of this process is iterating until we get to the outcome we set out to find.
What do organizations need to ensure consistency of their generative AI solution?
With pretrained models, there is a certain amount of data that may contain biases. But what organizations really need to focus on is how they’ll be augmenting a pretrained model to fit their needs. Essentially there is a knowledge base that the organization will embed into the Large Language Model (LLM), and with a highly curated set of data, they can drive better performance of that LLM. I like to call this process of ensuring the right data is curated and embedded GenAIOps, which helps ensure the efficacy of the solution. In order to make these solutions work, we need to move out of science project mode and start operationalizing.
It’s all about the data. Data and its governance are central to leveraging generative AI capabilities. If you don’t have a center of excellence around data governance, it’s something you should be investing in. When organizations kick the can down the road and think they’ll do it eventually, they miss out on a lot of value and the acceleration of that value. If your data is not high quality, there will be immediate consequences to your use case. I would say the center of excellence around generative AI may be broader and include people across the spectrum of expertise from machine learning to data, but ultimately in order to be successful with generative AI, organizations need to prioritize quality data.