Being mindful about generative AI: a conversation with Marinela Profi

By Jens, on Linkedin, as continuation of this series https://www.linkedin.com/in/jensfeilberg/recent-activity/articles/  

At SAS Explore on September 12, we heard about SAS’s plans to move into generative artificial intelligence (AI), with a focus on digital twins, synthetic data and large language models (LLMs). Generative AI has become a talking point since the launch of ChatGPT. SAS’s move into this area is very deliberate, and builds on many years of expertise in AI. I caught up with Marinela Profi, who is working on product marketing strategy for generative AI in SAS. 

Marinela, what were the key messages around generative AI that stood out for you at SAS Explore? 

I think the key message was that generative AI is about much more than just interacting with LLMs. This technology has the potential to transform the way we think about work. The real value is in developing industry focused applications using trusted GenAI that enable ultimately better decisioning. LLM are an important part of the GenAI but the real differentiator that SAS provides, in addition to our first class AI capabilities, is the decades of industry expertise.  

How will SAS be leveraging LLMs? 

The first way is to use LLMs to build applications more easily using SAS Viya. SAS Viya provides capabilities that you can use in your LLM app stack and tool chain to build applications more easily. Developers can access a lot of tools within SAS Viya, which facilitates the process of building applications. The second way is democratising the implementation of AI in analytics by using SAS Viya powered by AI assistants. This basically means that we incorporate generative AI to make SAS Viya even easier to use. I think we’ll see business analysts using this to ask questions through an interface, and having Viya detect the key tables for them.  

We’ve got a trial of SAS Viya starting soon. Is this going to be part of it? 

Yes, that’s the plan, although it’s a little way down the line. We’re already talking to the team running the trial about how we can incorporate generative AI. The trial is about raising awareness of SAS Viya, and the call to action is to try using it. That fits well with what we’re trying to do with generative AI. We want to find a few customers who are interested in starting to play around with early versions of our product and get involved in doing some tests.  

We know that there are issues about LLMs, especially accuracy. How are you dealing with these? 

This is absolutely the key point. Generative AI never gives you the same answer twice—it’s in the nature of the product. It’s therefore really important to consider all the accuracy aspects. We’re thinking of generative AI as a bit like an enthusiastic but not very good student—one that generally scores Cs across the board. We need to improve on that, because our customers need to be able to rely on the answers from analytics. What I’ve really learned from working on generative AI is that the human–machine partnership is going to be absolutely central. It’s very much a matter of using AI to augment human capabilities, not replace them.  

You said three ways that we would be leveraging LLMs. We’ve talked about two; what’s the third? 

The third is to execute industry-focused tasks using trusted SAS AI assistance. The other two ways I mentioned are really for those who are already using analytics and want to increase their productivity. We see them as active users of AI, or builders of AI. This third way we are leveraging LLMS is to allow people that are not AI experts to still benefit from the power of AI. I like to call them Consumers of AI. They don’t know much about AI or machine learning, but they have industry expertise and GenAI/AI can tremendously help them in doing their job faster and better. Suppose you are are investigator, and you are working on a case with thousands of witness statements. Instead of having to go through them all manually, you can use GenAI to provide a summary, or identify knowledge gaps, and then generate tasks to fill the gaps and trigger next steps. The idea is to use generative AI to build digital assistance. These are very much industry-focused solutions, and require industry expertise to build. It’s far more than just interacting with an LLM. It takes an LLM, then builds in trusted industry knowledge, regulations, or standards.  

Another example is in the medical field. Imagine you can generate synthetic medical images of a particular disease you want to diagnose, then use those images to train AI models, simulating medical scenarios, and enhancing diagnostic capabilities. 

 Effectively, SAS is bringing the industry knowledge in the Generative AI space, which I predict will be the real differentiator of the value creation process.   

Learn more in this free webinar: https://www.sas.com/en_us/webinars/generative-ai-strategy.html