Jim Goodnight is the CEO and co-founder of SAS, set up in 1976 as Statistical Analysis Systems. Computer Weekly has interviewed him many times over the past decades, as have our data management specialist colleagues on the US TechTarget network. He spoke to Brian McKenna at the company’s recent Analytics Experience user conference in Amsterdam. What follows is an edited transcript of that interview.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
What’s your take on the vogue for artificial intelligence (AI) and machine learning in enterprise software?
Goodnight: As SAS, we’ve been doing machine learning for almost 20 years. In one of our biggest applications, back in 2002, we were doing credit card fraud detection for one of the big banks in London, using neural networks. If you include logistic regression, one of the most useful iterative tools for machine learning, we’ve been doing that since 1977. Any non-linear model you have iterates over and over again, each time improving your results. That is machine learning.
Is the current novelty due to the enormity of the data volumes?
Goodnight: We now do these big jobs using massively parallel computing. Machines are going to have to get bigger and faster if we are to develop these complex models with millions of parameters. We have turned to GPUs [graphics processing units], which we will support in our December release, and they allow us thousands of multipliers and additions in every cycle. There are two billion cycles a second.
So is the vogue about quantity or quality, in the sense of different data types?
Goodnight: Many different models have been tried for, say, voice recognition. For that, we use recurrent neural networks, and we use convolutional neural networks for pictures. It’s just about feeding a neural network in a special way. It’s taken many years to construct neural networks that mimic vision or recognise voices.
Read more about SAS’s Jim Goodnight
- A podcast Q&A with SAS execs Jim Goodnight and Jim Davis looks at big data enablers, including in-memory analytics, and the role of data scientists.
- In an interview, SAS CEO Jim Goodnight advocates analytics over BI query and reporting, explains text analytics use cases and touches on predictive asset management.
- SAS, with Jim Goodnight piped in, opens a research and development (R&D) centre in Glasgow, focused on combatting fraud and financial crimes.
It is it useful to make a distinction between automation and autonomisation, where a machine exhibits human-like cognition?
Goodnight: If you ask Google “what is artificial intelligence?” it mentions decision making. We’ve been using AI, in that sense, for years. We are on that next four year period where we have to talk about something new. It was big data, then cloud, and now it is AI.
Let’s shift to a subject I know is dear to your heart – the education of the young. Do you still the think the balance between STEM and non-STEM education in the West is wrong?
Goodnight: I’d like to see 10% more going into STEM than we have now. But we still need liberal arts; we still need lawyers. The demand for STEM graduates continues to grow.
Of course, AI may start to automate legal tasks. Now, China and Russia are saying they are prioritising AI. It is reminiscent, is it not, of the space race, which generated a slew of mathematics PhDs in the US?
Goodnight: Yes, it did. There was Sputnik on 4 October 1957, and two years later, I was off to study applied mathematics.
I’m not a dyed-in-the-wool believer that AI is the next great thing. It will be a tough slog, with a lot of model building and lots of trials and errors. We are years away from a machine being able to think. We can train models to forecast what words you are seeing or estimate an image on a TV screen. But, even then, you have to train it to identify every single object, so it’s down to probability.
SAS as a company
Let’s turn to SAS as a company. How are you planning to make your market position, in advanced analytics, strategically defensible for the next 10 years and beyond?
Goodnight: Our Viya single platform view is important to that, and we are just at the beginning of what we can do with it. We’ve got this massively parallel computing platform and we can deploy a lot of our solutions on top of that – more data and dealt with faster. Our Visual Investigator product is already running on Viya. You can make enormous investigations with that. Our group in Scotland is taking all their intelligence agencies data and analysing it in almost real time using Viya.
Being from Glasgow myself, that is nice to hear. Now, you make much of your revenue from financial services. What’s your reading of how that sector – especially in London and on Wall Street has recovered from the 2008-9 crisis?
Goodnight: It’s doing very well. In the US, all the government bail-out money has been handed back and they are making near-record profits. Banking is back, but banks did need some new regulation put on them in terms of risk requirements. They were essentially taking people’s money and gambling with it. They had to be slowed down. It is a stronger sector now, and it is about 41% of our business – insurance and banking.
Talking to those important financial services customers in London, what is your sense of their sense of Brexit? There is much talk of relocating staff, for instance.
Goodnight: I have not talked to them about that. There has always been strong banking here in Amsterdam, too, and in Frankfurt. But we’re more down at the risk level. A lot of the bigger banks have tried to do their own thing, when they could have bought it from us for a million dollars.
London has got more of a Hadoop problem. There are not enough people in London who know it very well. With our Viya platform, we are reading Hadoop files in parallel directly into memory. We had a problem with one of the banks here [in Europe]. They kept losing their Hadoop people because they wouldn’t pay them enough, and they’d go somewhere else for better money.
Storage likely to become cheaper
Do you think Hadoop is becoming less central to big data projects than it was a few years ago?
Goodnight: I think we’re probably going to see storage getting cheaper, so Hadoop may not remain so relevant. We’re already using flash drives more, which have got algorithms on them for data compression.
We can essentially store a terabyte of data more cheaply on a flash drive than in memory in RAM [random-access memory]. RAM costs about $10,000 a terabyte, whereas now you can store a terabyte on some of the flash devices for $1,000. Hopefully, Dell will bring down a lot of the EMC pricing, which would help a lot. Hadoop has always been a play to get rid of SAN [storage area network] devices.
What else are you personally keeping track of, in technical terms?
Goodnight: I have recently been studying a lot of the neural network methodologies and written some code myself – just to make sure I understand it. I need to understand those deep learning algorithms. You’ve essentially got a model you are trying to optimise. They are just models.
From a business strategy point of view, what is at the top of your mind?
Goodnight: We’re concentrating a lot on the IoT [internet of things] and administering processing software there. We’ve got three or four major agreements in place to distribute that. We are working a lot at moving our model computation out to the edge versus bringing it all back to the cloud. And at the edge, we have our ASP engine running on Intel and Cisco devices, as well as on Raspberry Pi.
Finally, on the conference itself: you’ve done this event in the US, too. Are the conversations with customers any different?
Goodnight: They are the same conversations. We’ve got very good uptake on our Viya product in Europe, which is the strategically important product for us. It extends us into this massively parallel computing. Because all the computational algorithms are up in the cloud, to evoke one of these you just have to send some simple messages, such as, “run a regression with this target variable and these input variables”.
It is therefore just a simple message stream. Anyone using Python, Java or anything else can use their language, and we will translate that into commands for Viya. The results are then sent back down. The customer conversations are the same.