With artificial intelligence, machine learning, cognitive computing — call it what you will — all the rage at the moment, Tech Pro Research took a recent opportunity to speak with Gayle Sheppard, CEO of Saffron Technology. This Intel-owned company’s brain-inspired Natural Intelligence Platform helps enterprises analyse dynamic multidimensional data in pursuit of valuable insights.

Saffron’s history

How did Saffron get started?

“The company was established in 1999 by two gentlemen who were working in IBM’s Intelligent Agent Lab in Raleigh NC — one of IBM’s largest facilities at the time. Our chief neuroscientist is Manny Aparicio, and he wanted to build a technology platform that mimicked the way the human brain works, using associative memory. His cofounder is Jim Fleming, who is our chief engineer today, and they both left IBM to start Saffron Technology and build this platform.”

This was not long after the so-called ‘AI winter‘, so leaving IBM for an AI startup must have taken some faith and determination…

“It did — and think about the timing: in 1999 they were working on personalisation, designing software that allows computers to adapt to humans, rather than the other way around, the internet was skyrocketing and off they went. And then, ‘boom’, the dot-com crash happened, followed by 9/11. It was the perfect storm: there was no money to be raised, and a lot of problems for a fledgling startup like this.”

“Fortunately the phone rang, and it was the US Defense Department, so the company got started on national security at that point — a journey that continues to this day. I met Manny and Jim in 2004, having spent my career in enterprise software, building and implementing systems, and creating teams to do that. I ended up in North Carolina after leaving the enterprise software industry when Oracle acquired PeopleSoft [where Sheppard was vice-president and managing director]. They [Aparicio and Fleming] told me what they were doing and I got very excited, became an investor and got very involved — working as a volunteer within the company for a long time. But the barriers for the adoption of our technology were still very significant at this time.”

“We had a robust learning engine, with inputs from all kinds of structured and unstructured data, and our goal was to be model- and role-free, not to be reductionist, and to learn on the fly in real time — because at any point, anything can matter.”

That’s a challenging brief…

“Yes, we needed natural language processing to improve, we needed more structured and unstructured data — and we needed a realisation inside businesses that there are different ways to do analytics other than traditional statistics-based machine learning. So we had a lot of work to do, and we also needed a more mature platform that scaled and performed at enterprise grade.”

So how did Saffron’s early years play out?

“We worked with companies like the world’s largest private foundation, one of the largest aerospace companies, banks and so on — a small but important portfolio of customers who were solving a variety of problems, and proving out our platform. We did that until about 2012, when we felt that the market had made a fundamental shift: AI was no longer a dirty word — we could use it, and ‘cognitive computing’ was becoming a term, thanks to IBM, that kind of fitted naturally with how we thought of ourselves. And businesses, although they were not yet investing in cognitive computing, were beginning to understand it, explore it and research it. At this point, I became the CEO and we raised money, and Intel Capital happened to be one of the companies we talked to.”

Inside Intel

“Intel Capital had an aspiration about artificial intelligence and the future of autonomous learning, so they invested in our company and we implemented our technology inside Intel, helping it solve some important operational issues. We also became a strong partner to Intel Capital and their strategic investors: USAA is one that became an investor as well as a customer, introducing us to companies in financial services, banking and insurance, and becoming a wonderful partner and laboratory for us. About fifteen months later [in 2015], Intel acquired us, and for us it was a marriage made in heaven, so to speak, because we’re solving some specific problems, but underneath that is a fundamentally robust platform — and Intel is a platform company.”

“We have many partner opportunities inside our own company, to take our technology and embed it — in IoT analytics or client computing, or whatever it might be. Fortunately our management committee and my boss, Josh Walden, are very much invested in Saffron continuing to pursue customers directly, working to solve some very difficult problems in industry. That brings us up to today, where we operate inside the New Technology Group as independently as you can within a large corporation.”

How big is Intel’s AI portfolio?

“It’s pretty robust. Of course we have Intel Labs, which is very much on the forefront of the future; we have Movidius, which is our optical processing capability; we have RealSense, which is the vision part of that; we have drone technology, which is a big data-capturing component; and we have Nervana, an important acquisition on the deep learning side, which is really about learning what objects and things look like, and curating that knowledge — and we can take Nervana’s knowledge and use that as well.”

How much cross-fertilisation is there within Intel’s AI portfolio generally?

“We’re still mapping that out, but as we see it today there’s a tremendous amount of complementary learning going on there. I think ‘complementary learning’ is probably a new phrase we’re going to work on: there’s no form of learning that we should feel isn’t appropriate for our system, so we’re bringing that into Saffron. We have examples where we already do that, in sports analytics where we use our open-source TAP platform to do some pre-work on athletic performance, which we then take the results from to use for further analysis.”

The Natural Intelligence Platform

Saffron’s bedrock technology is the Natural Intelligence Platform (NIP), which it describes as ‘a key:value, incremental learning, fast-query, graph-oriented, matrix-implemented, semantic, and statistical knowledge store inspired by the associative structure and function of real neural systems’. Here’s what it looks like:

How is the NIP implemented in practice?

“The Natural Intelligence Platform is nicely wrapped in an API layer that customers and others can develop on top of, and we have twenty use cases that are proven. We also have Issues and Defects Resolution (IDR), which is really for product engineers, to identify duplicate and similar issues, and resolve those as quickly as possible — 35 percent of an engineer’s time is spent on duplication, according to research. Intel and Accenture — an important customer of ours — use that product today. We’re working on the next generation of anti-money laundering and fraud investigation — financial services and financial crimes are a really important product area for us. In each use case, the underlying platform will be the same: the main change will be the data sources themselves, making sure we have as many connectors and integration points as possible, and then on the top side customers will use different UIs — they may use ours, or they can plug in their own through our API structure.”

“What’s important is, we unify structured and unstructured data at an entity level: a lot of people talk about ‘oh, we store it together’, but they don’t unify it. Unifying at an entity level means every comment written by you, and every transaction booked by you, is now together in one memory — and when they’re associated with each other, that’s also connected. So you can do a transaction with your bank and write a comment saying ‘that was the worst teller I’ve ever worked with’: now you get the comment with the transaction, and that’s valuable data that just grows and grows.”

“My first complex project used 45 different data sources, doing predictive maintenance: we had manuals on how to operate things, and what the parts list looked like — all of this data coming in. The most complex project I’ve worked on was in defence, combating drug trafficking, with 145 data sources in different languages. One of the most important things to emerge from this was not just identifying that there’s a pattern occurring in the field, but it’s also the policing process you have to go through in order to take somebody to court: that process of evaluating all the evidence can take weeks — but if you can get it down to hours, there’s a greater chance of catching the bad guys.”

New use cases & digital transformation

Turning to future use-case developments, how does the Internet of Things (IoT) fit into Saffron’s plans? — after all, patterns in the data generated by IoT devices will be exceedingly valuable…

“When you have autonomous learning, like we do, you can learn from big data sources and small data sources, and there’s an opportunity to work very closely with IoT organisations and help them with the analysis part. As we mature inside Intel, that’s a great area of opportunity: when you’ve got to learn in real time on the edge of the network and yet you also want the experience to be captured across the system overall, then we’re perfect for that.”

Would a push on the IoT — with emphasis on real-time edge-of-network analysis — require much retooling of Saffron’s platform?

“I would say, hang onto that idea.”

OK, so I’m a CIO in possession of significant amounts of structured and unstructured data requiring analysis, and I’ve heard of Saffron Technology: what’s the typical timescale from greenlighting a project to receiving insights from a live deployment?

“I love this question! It’s faster than you’d think, and the key variable is always the customer’s access to data — is it available, and is there permission for it to be used? That’s probably the greatest time lag. But once we’re in data, then we’ve had results within 24-72 hours in terms of finding fraud rings, for example, so we can go fast. A typical implementation for production use, from the data design point — where we just try and understand what kind of knowledge base the customer is trying to build — to production, would take between three and six months, depending on the customer’s operating environment — we have to do change management, train people, standardise operating procedures and so on.”

In terms of winning new business, is Saffron actively reaching out to enterprises via a sales team and saying ‘You need us’, or do you have as many customers as you can handle at the moment?

“I would say the latter, and I also think that 2017 is an interesting year. We’re beginning to move away from customer discovery — ‘what is it [AI]?’, ‘how does it work?’, ‘would I ever want to use it?’, ‘what can I use it for?’, to ‘now let me build out some adoption plans’. I absolutely know that customers have been evaluating this technology for the last two to three years, and there’s been some implementation, of course, but not on a broad scale. We’re starting to move away from the very early adoption to the next phase of this.”

And is AI, going forward, the key to unlocking ‘digital transformation’?

“I think it’s certainly a great enabler. We’re truly moving towards the dream of mass individualisation — not personalisation, but individualisation. Instead of having to class people into groups, reducing them to statistics, we can literally serve customers as individuals — whether in healthcare, retail or whatever — and that should be exciting for everybody.”

Is the much-discussed shortage of trained data scientists a problem that you, or your customers, run into?

“I guess I talk to customers who are hoping they don’t have to do the data science. It’s time for vendors to build solutions that bring the data science inside the application — the customer has specialists, yes, but doesn’t have to have armies of data scientists. That’s where we are, and it’s our responsibility to move from platforms to applications and solutions as fast as possible.”

Ethics and other AI issues

There’s a lot of concern about ethics and AI — particularly as AI becomes ‘stronger’ and reaches ever further into peoples’ lives. What’s your position in this debate?

“First of all, strong ethics and social responsibility are at Intel’s core — we believe that’s just a demonstration of great leadership, which you have to have at the core of your corporation. That’s number one, where we start from. Number two is, there are organisations that are coming together to look at this topic — whether they’re academic institutions, or some combination of industry and academia — and they will explore how ethics and morals come into AI, and I think these are important things to participate in. And finally, with Saffron’s enterprise business focus, we focus on what jobs humans can’t do, or are difficult or dangerous for humans to do: how do we help workers get their jobs done better, more accurately, faster, with less stress? I think about AML (anti-money-laundering) and fraud investigators, and how stressed these people get, because criminals are using technology faster than their companies can adopt it — they don’t have the rigour of disciplined implementation and regulatory surveillance of this technology, like the banks do, so they can do it at will. So, I see a lot of stress in the workforce that I hope to eliminate — making machines work for them, instead of the other way around.”

Another worry that’s widely expressed is the ‘black box’ element of AI algorithms that increasingly affect our lives — can their workings be explained intelligibly, and will the companies behind them always be willing to reveal such information?

“Great question, and explanation is one of the most important topics — why are you making this recommendation, why do you think this is a suspicious activity, why is this fraud?

So if someone came to you and said ‘Saffron, please explain exactly how you came to this conclusion’, you could show your workings?

“Yes, all the way to the raw data — we keep that on record. And not only do we point back to it, our goal is also to provide that explanation in human-expressed language, so it’s easy for an investigator to see it — you don’t require a data scientist or an engineer to read what the machine is saying.”

Chaos and opportunity

Although Saffron Technology emerged from IBM at the turn of the millennium, its experiential reasoning approach differs from the curated knowledge system that underpins Big Blue’s Watson. Other software giants such as Google, Microsoft and Amazon all put their own emphases on the different techniques that comprise artificial intelligence/machine learning/cognitive computing, and there’s a burgeoning startup scene in this space too. So what does Sheppard identify as Saffron’s USP?

“There’s a lot of chaos in the world, and we thrive on high dimensionality, rapid change and chaos — we humans do well on it, and Saffron handles that really well.”

Chaos is unlikely to be in short supply in the next few years, and the need for the ‘good guys’ to find an edge will be more important than ever. Sheppard certainly sees plenty of opportunity for herself and her company.

“The next three years will be some of the most important I’ve ever worked. I’ve done some pretty interesting things, but the next three years will be the most exciting ever. How lucky can you be when that happens? The combination of Nervana, Movidius and Saffron is powerful, and we’re also very extraordinary on our own.”

Read more