What’s Behind SAS’s $1-Billion AI Investment
SAS’s announcement last week that it plans to invest $1 billion in artificial intelligence over the next three years may have taken some by surprise. The Cary, North Carolina company, after all, is the standard-bearer for the previous age of business analytics. But SAS refuses to cede ground to the new generation of AI startups, and has big plans to integrate AI with its proven analytical approach.
Behind the leadership of co-founder and CEO Jim Goodnight, SAS has been a trailblazer in the field of analytic software, not to mention a groundbreaker in corporate management in general. Perks like a 35-hour work week, unlimited sick days, and free M&M Wednesdays has garnered fierce loyalty and low turnover among a workforce that today numbers around 14,000. Meanwhile, its R&D investment is often twice what its cohorts spend, which helps to keep the $3-billion company’s analytic offerings on the cutting edge.
The new $1 billion investment in AI can be seen impacting both of these aspects of the company – its position as the preeminent brand for business analytics, which will help to attract the best and brightest to SAS, while also bolstering the underlying capability of its analytics software to give customers a competitive edge.
SAS plans to target three main areas with its $1 billion investment, including R&D, education, and services. During a conversation with Datanami at Nvidia‘s GPU Technology Conference (GTC) last week, Dr. Goodnight and SAS COO and CTO Oliver Schabenberger expanded on the investment, and discussed where they think the AI revolution will take us next.
On the software front, the goal really is not just to continue building AI tools, but instead to find ways to embed AI into workflows, Schabenberger said. “We see a lot of value in what I call the democratization of analytics, the democratization of AI, and embedding AI into our products and our solutions,” he said.
The education component is very big for SAS, Schabenberger said. “We can provide certification in data science and machine learning, and increase our presence in MOOCs [massive open online courses],” he said. “There’s still a huge skill gap out there in data science and machine learning, and I think with our history and our expertise in statistics and classical methods of analytics, there’s a lot we can bring to bear.”
Much of the emerging AI tech is complex, and companies need a helping hand. SAS intends to provide that through its services arm. “We see a big role for us in efficacy and advisory services and helping them get up the learning curve,” he said.
From Science Project to Production
By investing in improving AI products, AI education, and providing better AI services, SAS intends to help kickstart the AI practices of its clients, while turning AI science projects into full-blown production mode.
“I see many organizations building data lakes. They hire a hundred or 200 data scientists, and voila!” Schabenberger said. “No way. I believe that to truly scale analytics and data-driven insight at the breakthrough scale that we need, we have to should come down in the skill pyramid and enable the business analysts and software engineers to make use of that AI.”
Too many companies are stuck in AI science project mode, Schabenberger said. That’s because they underestimate the difficulty in making the big step from having a clever machine learning model that works in theory to actually having a machine learning model that impacts the business in a meaningful way.
With a 43-year legacy in providing business analytic solutions, SAS is arguably among the most well-qualified companies to help large Fortune 500 firms take AI from science project to production solution. To that end, a lot of SAS’s future success with AI hinges on its Viya platform, which it’s positioning as the main vehicle for helping clients move their data science and AI initiatives forward.
Viya, which SAS first unveiled in 2016, is a collection of data science development and runtime components, including Data Preparation and Model Manager, Visual Data Mining and Machine Learning (VDMML), and Event Stream Processing, among other modules. Users can interact with Viya in the traditional SAS language or program in a more visual, drag-and-drop manner if they choose. The software also supports both cloud and on-premise deployments.
Viya also supports emerging environments, like Python, which is arguably the most popular language in data science today. According to Dr. Goodnight, SAS has strived to make Viya welcoming to the burgeoning AI space, compatible with open source languages, emerging frameworks, and speedy GPUs, while still delivering the enterprise-level features and performance that longtime customers have come to expect from SAS.
“A lot of the emphasis on this [GTC19] conference is speed,” Dr. Goodnight said. “How much faster? 10x? Well, we bring that same capability to Python programmers who are writing in Python, which is an interpreted language. So any time you have to do any heavy lifting, like you have to do in AI and machine learning, you have to call out to somebody else’s libraries in C or C++…that have the performance. What we provide them is massively parallel in Viya, so instead of calling a single-threaded library, they can call a multi-threaded, massively parallel library and then get the results back.”
In addition to supporting Python, Viya customers can leverage GPUs, Dr. Goodnight said. “Not only are we massively parallel, but by putting Nvidia chips in there, we can make it even faster,” he said.
Neural Networks and Future of AI
Much of the excitement around AI these days focuses on neural network models and deep learning. Frameworks like Tensorflow, PyTorch, and MXnet help developers create complex machine learning models, often running on GPUs, that train against millions of features. These deep neural nets so far have mostly driven advances in computer vision and text processing applications, but other applications are beginning to emerge.
It may surprise the reader to know that SAS has been using neural networks in its products for decades. “A number of the large banks use us to screen all the purchases and credit cards and determine whether or not [it’s fraud],” Dr. Goodnight said. “That’s a neural network. We’ve been doing that for 20 years. It’s not real deep, but it’s a neural network.”
As deep learning becomes more applicable to non computer-vision and text-processing problems, SAS intends to be there to help customers figure out how to make it work for them. That’s another aspect of the $1 billion AI investment to watch out for. In particular, SAS is exploring how best to solve deep learning’s explainability problem.
“We can help our customers understand that, but we can’t do it for them,” Schabenberger said. “What we can do is provide tools that help with the interpretability and explainability and analysis of the models. For example, they have a complex machine learning model. Well, which variables really drive that? Which features are driving your decisions?… That’s what we’re providing.”
In many ways, the $1 billion investment in AI that SAS has pledged is not anything new. There will be no radical transformation of the privately held company, which has weathered recessions, technology booms, and busts without much drama, at least to the outside observer.
“It’s just a continuation of the last 10 years of investment in high performance computing,” Dr. Goodnight said of the $1 billion investment.
We are in the early days of AI, in many ways. While it may seem that the field has become quite sophisticated, there is room for a lot more automation, at least in the eyes of folks like Dr. Goodnight and Schabenberger. That leaves plenty of room for SAS to innovate on technology, differentiate on products, and claim its share of the emerging AI pie.
“I’m looking forward to the next wave of AI systems,” Schabenberger said. “To me, AI is really about data-driven automation, increasing amounts of data, data of different types, more unstructured data, and more streaming data. Our customers are drowning in data. So what can we do to help customers deal with data and automate around it? We will, and do call it, AI.”