AI-fueled efficiency a focus for SAS analytics platform – TechTarget

author
10 minutes, 25 seconds Read

SAS on Wednesday unveiled a host of new features aimed at helping analytics users more efficiently access and analyze data with AI, including a generative AI assistant and prebuilt AI models.

In addition, the vendor introduced synthetic data generation capabilities, a new environment for developing AI models and applications, and model cards that provide governance details about AI applications.

All were revealed during SAS Innovate, the vendor’s user conference in Las Vegas.

Based in Cary, N.C., SAS is a long-established analytics vendor that offers Viya as its primary platform for data exploration and analysis. It also provides industry-specific tools tailored to the needs of customers using data for such applications as risk management and fraud detection.

Unlike many data management and analytics vendors that in late 2022 and early 2023 quickly integrated with generative AI models such as ChatGPT and Google Bard, now known as Gemini, SAS took a measured approach to integrating GenAI with its platform.

SAS has been a dedicated developer of traditional AI, committing to invest $1 billion in AI in 2019 and another $1 billion in May 2023. But as peers including Microsoft and Tableau unveiled plans to build generative AI capabilities during the first half of 2023, the vendor held back on investing in generative AI over concerns related to the accuracy and security of generative AI models.

That changed last September, and now SAS is planning to develop some capabilities such as a copilot that are similar to what others are building, along with GenAI tools that are more distinct.

The part that stands out for me is Data Maker. SAS is one of the few vendors that is talking about and addressing the need for synthetic data generation.
Doug HenschenAnalyst, Constellation Research

For example, Data Maker, a synthetic data generator now in preview that will help organizations that don’t have enough data to train AI, is somewhat unique among analytics vendors, according to Doug Henschen, an analyst at Constellation Research.

“The part that stands out for me is Data Maker,” he said. “SAS is one of the few vendors that is talking about and addressing the need for synthetic data generation. Constellation believes data scarcity will limit the accuracy and effectiveness of AI-based systems. SAS is one of the few companies talking about this capability in the context of their generative AI capabilities.”

Copilot capabilities

In the 18 months since OpenAI released ChatGPT — a launch that marked a significant improvement in generative AI capabilities — many data management and analytics vendors have prioritized generative AI development.

One main reason is that large language models (LLMs) such as ChatGPT and Google Gemini enable true natural language interactions rather than the limited natural language processing some vendors attempted to develop. True natural language interactions, meanwhile, enable nontechnical employees within organizations to query and analyze data, which has the potential to broaden analytics use within organizations.

Another main reason data management and analytics vendors have focused on generative AI is its potential to make anyone who works with data more efficient. By enabling natural language interactions, data scientists, data engineers and other data experts are relieved of some of the time-consuming coding previously needed to develop and analyze data products. In addition, LLMs are able to generate code when programmed to do so, enabling data experts to automate previously time-consuming processes.

Viya Copilot addresses improved efficiency, according to SAS. Now in private preview, the feature is designed to reduce time-consuming tasks such as code generation and surfacing insights such as knowledge gaps in applications.

“The capabilities delivered via Viya Copilot … are very important,” said Mike Leone, an analyst at TechTarget’s Enterprise Strategy Group. He added that the firm’s research shows about 1 in 4 organizations pursuing GenAI use cases cite chatbots as their top priority.

In particular, many organizations view AI assistants as a means of enabling more employees to explore data, he continued.

However, while having an AI assistant will be new to SAS customers, similar tools are not new to users of other data management and analytics platforms.

For example, Microsoft unveiled plans to add copilot capabilities in Power BI in May 2023 and did the same for its new Fabric platform in November. In addition, Domo and Tableau recently introduced their AI assistants, while MicroStrategy not only unveiled such capabilities in October, but also made them generally available.

But while many other vendors have already introduced AI assistants, most, like Viya Copilot, are in some stage of preview. In addition, the use of such tools is in the future plans of most organizations rather than their present plans, according to Henschen.

“Large customers doing analytics and AI at scale do not switch horses quickly based on this or that hot new feature,” he said. “In fact, plenty of companies and [chief experience officers] are being very cautious about GenAI.”

With respect to the perception that SAS is perhaps trailing the generative AI innovation speed of other vendors, SAS CTO Bryan Harris said it’s more important that SAS get generative AI right rather than be first.

SAS caters to a customer audience of large enterprises often engaged in highly regulated industries, whereas some analytics vendors, while also serving large enterprise customers, also have users that are smaller businesses in perhaps less regulated industries. As a result, before SAS can introduce new features, it has to make sure they are secure and deliver accurate results.

“We’re not a company that prides itself on selling hype,” Harris said. “We’re a company that prides itself on results. The success of a 47-year-old company is about delivering results, not hype.”

Toward that end, some generative AI assistants and other generative AI tools introduced by SAS competitors have been in preview for as much as a year.

“We saw a lot of competitors who are not in regulated industries and don’t have to worry about the same things we have to worry about could speak … about generative AI,” Harris said. “Now, they’re realizing they aren’t going to work the way they thought they were.”

Seven benefits of generative AI for the enterprise.

Other new capabilities

While Viya Copilot aims to make data workers more efficient, SAS Data Maker is a synthetic data generation tool within Viya designed to help enterprises train generative AI models.

LLMs such as ChatGPT are trained on public data and can accurately respond to queries related to that data. They can answer all kinds of questions about World War II and even generate text that mimics a poem by William Wordsworth or a song by The Who. But they have no clue what an organization’s weekly sales were in Washington over the winter months.

To respond to queries about an individual business, the business has to use its own proprietary data to either fine-tune an LLM or develop its own domain-specific language model. Furthermore, to train those generative AI models to deliver accurate query responses related to an individual business, a huge amount of data is required.

Without enough data, generative AI models are prone to delivering incorrect outputs called AI hallucinations that can be misleading if not checked for accuracy.

Some organizations, however, don’t have that requisite amount of data to at least reduce the likelihood of AI hallucinations to an acceptable level. As a result, SAS Data Maker is a significant addition to the vendor’s analytics platform, according to Henschen.

The tool, which like Viya Copilot is in private preview, creates synthetic data that statistically mimics an organization’s actual training data to give organizations more data with which to work. In addition, it protects sensitive information so that data such as personally identifiable information is not accidentally replicated and exposed.

Leone noted that Enterprise Strategy Group’s research shows nearly half of all organizations regularly use synthetic data when training models, either as a substitute for their real data or to augment that real data. As a result, SAS Data Maker is a meaningful new feature to the vendor’s analytics platform.

“It aligns very well with what the market is searching for,” Leone said.

In addition to Viya Copilot and SAS Data Maker, the vendor introduced the following:

  • Industry-specific AI models expected to be generally available before the end of the year that are designed to address real-world use cases such as healthcare and manufacturing, available to buy on an individual basis.
  • Model cards in Viya, an auto-generated feature in private preview that provides details about models such as accuracy, fairness, model drift and model lineage so that customers can know which models can be trusted.
  • Viya Workbench, an environment for building AI applications first introduced in September that will be made generally available before the end of the second quarter. It includes data preparation, exploratory analysis and machine learning model development capabilities.
  • The formation of an AI governance advisory board to assist customers as they increasingly rely on AI models and applications to inform decisions.

Combined, SAS’ additions demonstrate a concerted effort to provide customers with capabilities that result in trusted AI that will lead to buy-in and increased adoption, according to Leone.

Model cards and prebuilt AI models, in particular, have the potential to help customers confidently work with AI, he noted.

“[Model cards] are all about trust and reliability,” Leone said. “Not only are they showing AI metrics like accuracy, fairness and model drift, but also key governance details. The idea of specialized models as a service will really empower organizations to ramp up AI initiatives faster.”

Udo Sglavo, SAS’ vice president of applied AI and modeling R&D, similarly highlighted the prebuilt AI models as one of the more significant projects the vendor has in development.

SAS has long provided industry-specific tools that sit on top of Viya and previous SAS analytics platforms, tailored to the needs of customers in specific industries. The prebuilt AI models are an extension of those industry-specific tools whose use by customers has provided SAS with years of intellectual property, and likewise they will sit on top of Viya.

SAS plans to initially launch AI models built for banking, finance, healthcare and government, and subsequently follow with dozens more.

“The one unique capability we are delivering is our plan to release models as standalone offerings,” Sglavo said. “These are an extension of our portfolio. Over many years, we successfully created a platform including industry solutions. We will target specific industry problems and solve them with a model. … The goal is to focus on a business application.”

Next steps

As SAS plots product development, its guiding principles have been productivity, performance and trust, according to Harris.

At their core, the features the vendor unveiled on Wednesday combine to address each. Given that Viya Copilot, SAS Data Maker, prebuilt models, model cards and Viya Workbench are all in preview, they essentially represent SAS’ roadmap for the rest of 2024.

Looking beyond what was unveiled during SAS Innovate, the vendor is planning to add more generative AI capabilities such as using natural language to develop data flows, models and dashboards, Harris said. In addition, using AI to improve data management with a specific focus on data quality is a point of emphasis.

Perhaps more big-picture, SAS is also planning to do more toward implementing quantum computing. Data management and analytics workloads are growing quickly, as is the amount of data organizations now collect and try to operationalize. Quantum computing provides substantially greater computing power than traditional computing.

“I believe that in the next 24 months, quantum computing will have the same kind of moment that LLMs and generative AI are having now,” Harris said.

SAS isn’t planning to build its own quantum computers, but it’s already developing hybrid architectures that add quantum computing to classical computing as an accelerator, he continued.

“We’re seeing promising results,” Harris said. “As we become more hyperconnected, the complexity of some problems is demanding some out-there approaches on how to identify solutions. Not everything can be done by a single algorithm.”

Henschen, meanwhile, said he’d like to see SAS not simply introduce AI capabilities, but also make them generally available.

AI development is happening quickly across the data landscape, he noted. By the time tools held in preview for lengthy periods are finally released, they might no longer be the vanguard.

“The pace of innovation is constantly accelerating, particularly in GenAI, so I’d like to see these private-preview announcements move into public preview and general availability as quickly as possible,” Henschen said.

Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.

This post was originally published on this site

Similar Posts