SAS Takes Pragmatic Route to GenAI

GenAI is an exciting phenomenon. Consumers are all excited about its generative capabilities, creating new use cases every other day.

Yet, in the enterprise world, CIOs are scratching their heads. It's not that GenAI does not offer significant productivity and creativity advantages; they just can't seem to see how that justifies the considerable investment they need to deploy it.

Bryan Harris, executive vice president and chief technology officer at SAS, understands this well.

"What's happening on the public side is when you have public information, generative AI enables people to do faster things productively with public data. But we must now deliver that same capability inside [the enterprise]," says Harris.

So, SAS is not launching a GenAI product (at least for the moment). Instead, it is adding new features and capabilities to SAS Viya and investing in its consultation so its enterprise customers can determine what that GenAI product is.

Making the business case for use cases

The challenge comes down to creating the right use case for generative AI. Yes, there are many reasons to adopt generative AI for productivity and decision-making reasons, but specifying how that is in business terms and ensuring the cost-benefit analysis is right might take a lot of work.

"I'm seeing this great set of capabilities and productivity in the public domain, but now I need to see how that translates into my business. And that's really what we're focused on," says Harris.

At SAS Innovate Singapore, that’s what the company focused on.

The analytics leader is investing USD1 billion in the next three years to create advanced analytics solutions for different industries—all running on SAS Viya, its cloud-native, massively parallel AI and analytics platform.

The company has been making significant inroads in finance, its major user base. It has "retooled" SAS Asset and Liability Management (ALM) for SAS Viya, helping banks mitigate the liquidity and balance sheet risks that triggered recent bank failures. It is also connecting siloed functions (e.g., onboarding, credit risk, fraud detection, marketing) on a unified, AI-powered architecture to streamline operational complexity while helping them make accurate decisions faster.

In insurance, the company launched SAS Dynamic Actuarial Modeling to help insurers offer fair pricing in a market facing enormous underwriting losses due to natural disasters. Meanwhile, its SAS Health and SAS Clinical Enrollment Simulation are helping healthcare players use AI to run their operations more intelligently.

Sizing up the use case

So, if you are an enterprise, how do you start creating the right use case. After all, each company can be unique regarding processes and actual data.

Harris suggests companies start with the outcomes. Then, he advises, you need to work backward. "It's then about how fast can I go through the AI lifecycle to create a capability that allows me to make that impact?" he explains.

To do this, especially with a shortage of AI and data engineers and data scientists, SAS is adding AI and automation to SAS Viya. But will this impact their core user base, expert users like data scientists?

Harris does not think so. He believes that automation will only help data scientists to focus on more complex tasks.

When data scientists think you're taking away or automating these things about their jobs, I will argue that there's more to do. There's not enough talent out there who still understand AI's core skills and capabilities, and AI is really analytics and statistics underneath it," says Harris.

Data scientists can also help business users understand whether the model is sound. "It's not magic, right? We need data scientists to help make sure that what you're creating is proper. You can run bad math on good data; you can run great math on bad data," he adds.

Closing the foundation model gap

One method to build industry-specific use cases for GenAI is to start with a foundation model. It is essentially an AI model trained with a wide range and variety of input (often with some sort of supervised learning like self or semi).

The term, popularized by the Stanford Institute for Human-Centered Artificial Intelligence, refers to models that can be fine-tuned to a wide range of downstream tasks.

While it makes perfect sense for any enterprise to build a foundation model (since they own a large enterprise data set that is often private), it's hard, expensive and needs some time.

But that's easier said than done. "It's a tough one because a foundation model requires this acceptance that as an industry, we're all gonna say these things," Harris argues.

Yet, he believes time will tell, and building a large language model inside an enterprise will only get less expensive.

"The question is if we're gonna see a bit more focus on how does one take the large language models that are out there that will then be the starting point for a narrower enterprise model that gets brought into the organization," says Harris.

This is a sweet spot that SAS intends to focus on.

"We want to leverage [the value from large-scale models] and bring them to a starting point. Then for industries, [we can use] domain-specific training. The trick [lies in] how we bring that in quickly, [and use] more constrained training to close the gap between the general public and the industry [outcomes]," says Harris.

SAS, he notes, is exploring ways to do those training. One way to shorten the learning time and expense is to create a foundation model for an industry.

Wait, there’s SAS

During the SAS Innovate event, SAS highlighted the various features of the latest SAS Viya. Many ease the burden on data science teams and help them to do dataops or ModelOps.

However, SAS still has a large group of SAS 9 (now 9.4) users. In fact, many use the latter for production and speed.

Harris points out that SAS is looking to add more capabilities to SAS Viya to give more reasons for customers to migrate. It also unifies the experience, like offering SAS Studio in version 9.4.

“It's very important for us, at least specifically for me in the R&D side of the house, to get our customer base to a unified platform so that everyone can experience the same innovation that we're delivering on that part of the business,” says Harris, who adds that SAS 9 is not “a small part of the business.”

But instead of pushing customers, SAS leaves it to them to decide when to migrate. "Although we want them to move to Viya eventually, we don't want to force them in an unreasonable timeline," says Harris.

In fact, SAS is looking to help SAS 9 customers understand what it will take to migrate through a set of tools. "It gives them an explainable set of reports that tells them what can translate straight over to Viya, what might need some changes and what might be a gap," Harris explains.

"It's about taking care of our customers where they are so that we can help them get to the future state, and then we work on the automation of that process," he adds.

It’s the same approach the company is following for GenAI—and that’s a good thing.

Winston Thomas is the editor-in-chief of CDOTrends and DigitalWorkforceTrends. He’s a singularity believer, a blockchain enthusiast, and believes we already live in a metaverse. You can reach him at [email protected].

Image credit: iStockphoto/BayramGurzogl