Q&A with SAS CEO Goodnight: Analytics Development from A to Z

CEO and co-founder built his own company, stayed with it and now is seeing business analytics, AI and machine learning become democratized across a legion of new mobile applications.

CP.Jim.Goodnight.SAS.2012

We're constantly hearing about how analytics and machine learning are turning the IT world upside down. The is due directly to the major convergence in the last five years of fast new networks, leaner software, more efficient software architectures, more powerful but less lower-hungry processors, ingenious mobile devices and virtually unlimited data storage.

All these factors dovetailing at the same time have enabled the rapid spread of heavier and more powerful applications though cloud services that do much more calculating than others that merely hit a couple of databases in order to find a cheap airline ticket.

SAS Institute (the original name was Statistical Analysis System) and its co-founder, Dr. Jim Goodnight, largely invented software-based business analytics back in 1976. While the years of rapid growth seem to be behind it, the company has continued to grow from a $2.87 billion in 2012 to an estimated $4 billion in 2017.

The company has poured 25 percent of its revenue back into research and development. The company is privately held, which mutes possible concerns regarding so much revenue going into R&D for relatively minor growth. The company has a commanding share in business analytics, a loyal customer base and a good track record of customer success stories.

SAS Uses All Its Own Home-Developed Components

Cary, N.C.-based SAS uses its own analytics, in-memory architecture, intuitive data exploration, Hadoop support and information-delivery options in its in-memory Visual Analytics product. For a long time it was the only in-memory engine designed for business visualization of big data that runs on inexpensive, non-proprietary hardware.

Goodnight, 75, whose company consistently ranks high in annual "Best Place to Work" polls in the U.S., is a modern Renaissance man. He is a computer science professor and software developer at North Carolina State University who founded his own business, grew it into the world's largest privately held company of its kind, became a real estate developer of epic proportions, and has become an icon of the tech industry.

Goodnight met with eWEEK for a question-and-answer session on May 2, 2018. He's pictured (at right) with eWEEK editor Chris Preimesberger.

eWEEK: Analytics applications are now available in about every sector of IT. Could you have foreseen this only a few years ago?

Goodnight:  Yes, a lot of people are talking about it. Actually they don't use the world 'analytics' anymore; they use the word 'AI.'

We're certainly seeing good strides in hardware, especially from Intel as it continues to shrink the circuitry inside its chips all the time. We've done quite a bit of work with Nvidia with their GPUs (graphical processing units). We're in the situation now with some of the deep learning and machine learning apps that we're having to estimate millions and millions of parameters or weights that control the outcome of the functions.

It takes a lot of multipliers in addition to be able to do these estimations, and some of the GPU chips are really proving to be useful in doing that. As machines have gotten faster and faster with more and more memory ... we like to keep our data in memory as much as possible, because on all of the machine-learning algorithms are all iterative, so you have to go through your data numerous times. If you can hold it in memory, that's the ideal thing.

We started on our in-memory journey about nine years ago. We've been writing in-memory, massively-parallel algorithms for all that time. From that standpoint, we have a pretty good jump on the marketplace, as far as the quantity and the depth and the payload capabilities.

Using our own internal control mechanism, we're able to actually switch controllers if one goes down as needed. Also, if one of our servers goes down, we can bring up another one to take its place. We've spent a lot of time in this area.

We have a new product called Viya, our massively parallel machine learning neural network for deep-learning systems.

[Editor's note: Viya is a cloud-based platform for data preparation, intelligence and investigation management, model managing and optimization that was released in 2016. Go here to view a video.]

eWEEK: Nine years ago was just about the time SAP came up with HANA, its own in-memory database, which completely changed--some people claim saved--their business. I see a parallel there.

Goodnight: Yes, we worked with them in the early days to make sure our in-memory products would run on their machines.

eWEEK: It's amazing to see chips like Nvidia's that were designed for consumer apps, such as video-gaming, taking their place in frontline enterprise IT.

Goodnight: Nvidia's whole purpose for starting was to be in the game-acceleration business to make graphics much, much faster for gamers. But it has ended up being the heart and soul of our deep-learning computing.

We've always done deep learning, or machine learning, as it's being called now--that goes back to some of our very first non-linear modeling routines that we developed about 40 years ago.

eWEEK: Sure, but those apps weren't nearly as available as they are now. They all were run on large mainframes 40 years ago.

Goodnight: They were, but the CPU in my iPhone is more powerful than those mainframes were back in those days.

eWEEK: What might we expect to see coming from SAS in terms of new products, revisions on current products and innovation in general over the next 12 to 24 months?

Goodnight: Well, as far as deep learning is concerned, we're going to see a lot more in terms of AI bots for call centers, where we can develop bots that can answer any question that you ask, based on our huge database of knowledge that we can search on for the right answer.

Another area that we're working on is medical image scanning, using neural networks to be able to determine whether a chemotherapy regimen is working or not. We're working with the Cancer Center on that. The goal is to be an adjunct to a technical doctor or radiologist who's reading the CT scans, to be a backup or second reference for that. There are some powerful capabilities for the medical field.

Where we've seen the most success today is in voice-to-text and text-to-voice. Google has done an incredibly good job in that. I pick up my Google phone all the time, and they recognize every word I speak, and I easily come back with a voice-response answer. That's one area that AI has really shined in. They're also getting very good at translations, from one language to another; that continues to improve year after year.

We use neural networks for things like fraud detection, but we've been doing that for over 20 years. A lot of the neural network startups are having money thrown at them by the venture capitalists, but you just can't take a bunch of data, throw it at a neural network and expect to come up with the most brilliant answer in the world. It still takes a knowledgeable person to understand what features and variables need to be created before you start doing the neural network analysis.

eWEEK: What is SAS working on in terms of AI for IoT apps?

Goodnight: We've been working on IoT for about four years now. We have one of the best event stream-processing engines in the industry. We're working with a lot of manufacturers who are putting sensors into their equipment to provide displays of current readings and to work on models in order to predict when things are going to fail. They also provide information about the efficiency of equipment that's running.

GE has about 1,000 locomotives right now that our IoT (software) is on, to be able to model about 600 different readings on those engines. They're predicting how to save fuel, how to improve performance, when parts are going to break down, and so on.

eWEEK: I'm still looking for the killer app for journalists, and for, say, lawyers doing depositions. That would be a high-quality voice-to-text app, so I wouldn't have to transcribe long interviews all the time--just edit them. Do you know of one?

Goodnight: Sure. Google's got one now, and you can find it on Github, I believe. [Go here to check it out.]

They also have text-to-voice that also is really good, called Tacotron 2. You can get any book and it'll read it to you.

But do we have an app yet that'll replace a lawyer?

eWEEK: Now wouldn't that make an interesting news story!

Chris Preimesberger

Chris J. Preimesberger

Chris J. Preimesberger is Editor of Features & Analysis at eWEEK, responsible in large part for the publication's coverage areas. In his 12 years and more than 3,900 stories at eWEEK, he...