The supercomputer-powered AI age is just beginning
9 July 2018
Per Nyberg, Vice President, Artificial Intelligence
Artificial intelligence, or AI, is all around us. It’s in hospitals, smart factories, banks, insurance companies, and recommendation systems that help you choose the next hit series to binge-watch. But would you believe we’re just entering the dawn of the AI age? It’s true. Even though the science around artificial intelligence has been evolving and maturing for more than 60 years, it’s only since 2011 that AI and one of its advanced technologies – deep learning – has captured the imagination of industry, driven by advances in big data and analytics, access to unprecedented computation power, and newer algorithms that mimic the structure and predictive power of the human brain.
Cray AI VP Per Nyberg recently spoke with Business Reporter’s Alastair Greener, answering a range of questions about the state of AI adoption by businesses.
Nyberg’s observations about the future of AI in business included:
- Skills – or, more specifically, the shortage of skilled practitioners – are the immediate challenge for organisations that want to leverage AI in business or science.
- Supercomputers will be an inevitable ingredient for successful deep-learning deployments, and not just because of their computational power. Why supercomputing? Mainly because of, as Nyberg describes it, “the sum of the parts that are necessary to solve one of these difficult problems, everything from data-ingest all the way through to the scaling of the actual computation.”
- Finally, while AI has emerged as a key business trend and there’s a sense of urgency about implementation, patience is required. The successful adoption of AI “is going to play out over a number of years,” as Nyberg describes, and we’ll need “a degree of patience to continue to invest – even if there are some stumbles along the way.”
Let Cray show you how to exploit supercomputing for your toughest AI problems.
Welcome to Business Reporter's digital economy campaign. I'm Alastair Greener. The concept of artificial intelligence, long in the domain of science fiction, has rapidly entered the lexicon of the digital economy. Those who apply AI expect that big data, which to date has been unmanageably complex and vast, will become an opportunity. Artificial intelligence is expected to make businesses faster, more efficient, and even more responsive to customers. But for it to be faster and better, do we have the necessary hardware, software, human capital, knowledge, skills, and security that a mature AI project needs? It looks like that the promises around AI are far bigger and brighter than previous transformational technologies like big data.
But is there a gap between the promise of AI and the reality businesses face when trying to make AI work for them? We've invited Per Nyberg from Cray, the leader in supercomputers, to help us understand how we can get the very best out of this cutting-edge technology. Good morning.
What's driving interest today in artificial intelligence?
Well, it's really that organisations are looking for solutions to solve problems which they otherwise thought intractable, and to really address this data deluge that they have. This is something that all organisations are facing in, and it's a pretty severe challenge. There's a lot of value in that data, and they see that they can gain a competitive advantage by using new technologies like artificial intelligence.
Are there different approaches to AI? Are people coming at it from different directions?
Yeah, that's a good question. So AI, artificial intelligence, I mean, it's one of these terms that now kind of means anything and everything, and most people look at AI as having kind of three components to it. One is certainly advanced analytics, predictive analytics, which became very popular in the industry a number of years ago around big data.
The more, I guess, specific areas that people are looking at today are machine learning and deep learning, right? And in both cases, it's really all about machines training themselves essentially on data, past data in order to be able to predict the future in some form. So deep learning is a form of machine learning, and more specifically looks at types of data like text or video or images, time series data.
Now we know that it's about patterns, but to recognise patterns, we need to know these patterns in the first place. So who's actually providing them? Where are they coming from?
Yeah, so a lot of the deep learning that you hear about today is what's called supervised learning. And in that case, you've got data that's in one form or other labelled by a human. The kind of silly cat, not a cat example where you would tell a machine or give it enough examples of what is or is not something, and it would train itself based on that data. So in that case, you know, the human was helping the machine to essentially at least give it a starting point upon which it can train itself and then use data to make its own predictions.
There's also unsupervised learning, which in some ways is even more exciting, because it doesn't really need the human, if you will. There's a number of approaches in that case. You've got clustering, for example, where the systems will use clustering methods to kind of group data and look for patterns on its own and then make inferences from there. And so that's a growing area. There's still a lot of research across both areas. But ultimately, it all starts with the data.
If an organisation starts using AI, is it something that just can be used, or is there a whole process to go through?
In one form, everybody can use AI today, typically as a consumer, let's say, right? I mean, a lot of the apps on your phone might have some form of AI that's been built in. But you know, it's kind of less interesting at some degree.
If you really look at organisations that are trying to leverage AI in their lines of business, for example, or in their areas of science, you know, the organisations have some degree of experience in just data analytics in general. They have strong sort of infrastructure and experience in data management.
So you have some kind of table stakes areas that will allow you to start exploring some of these more advanced methods. Because ultimately, again, without the data and good data, if you will, you know, you're not going to get the accuracy or the results that you would need out of machine or deep learning.
You talked about expansion and experimentation. Tell us a little bit more about how Cray is involved in that, and how you help organisations.
Yeah, so today, we look at-- well, first of all, I mean, deep learning and machine learning are applications that we see across our entire customer base. Doesn't matter which discipline that they may be in.
But in terms of some example areas, industries, we have autonomous technologies, health care, pharma, aerospace, personal insurance, oil and gas. So as you can see, it's a very broad spectrum of users that are leveraging deep learning specifically for their particular problems in their lines of business.
So really, what Cray provides are the tools and the capabilities to solve these types of problems. And this is not only from a performance perspective, which Cray is very well known for, but also from earlier stages in their journey in terms of exploring AI. We kind of call this the ideation stage.
So this is really all about putting the tools in the hands of those data scientists and researchers who are exploring the potential and power of AI. And then as they grow in their journey from ideation to prototype to production, be able to work with them along that journey, provide systems that scale and are able to solve their problems as they become increasingly complex, and provide a lot of the expertise, which in high performance computing and supercomputing, that many organisations don't have, but that are necessary in order for them to solve these AI problems.
When it comes to supercomputing, what sort of power is missing there that you've been talking about?
Yeah. I would actually-- instead of the word power, I'd use kind of capabilities. And the reason why I say that is that supercomputing by its definition is more than just speed of the processors, let's say. That's a really interesting kind of area. It's a fun area that people point at, but it really is more than that.
Supercomputing is all about the sum of the parts that are necessary to solve one of these difficult problems. So that is everything from data ingest, let's say, all the way through to, you know, scaling of the actual computational portions of these applications. Communication within the machines, output, running large sets of jobs, the tools, the system scheduling. So without getting into too many of the technology features, it really is that sum of the parts.
Now from a gaps perspective, I think the supercomputing industry-- and I can say for ourselves, I mean, we're learning a lot from those leading customers that we have today or leaders in the industry today that are developing these next-generation applications in deep learning. A lot of the challenges simply haven't been identified yet, and that's a good thing, right? It's a fun thing. I mean, our engineers love that.
And you know, we've really targeted a number of these leading organisations and are working closely with them on a day-to-day basis to really understand what they need today and what they will need in the future. Again, this is going to play out over many years.
As you say, things are changing rapidly at the moment, and we're suddenly living in very exciting times. So when do you think that AI and supercomputers will become the norm, part of our everyday lives?
Yeah, that's a good question. It's a little bit of a crystal ball question, but I mean, there's certainly problems today, AI problems today. And I don't know, they give you something like a chat bot, for example, which they're pretty ubiquitous already. They don't really need supercomputing per se. But there already is some today.
I think that as the industry as a whole-- again, across industry and science, as their use cases mature, this need for supercomputing is inevitable, right? But it's going to take some years. We've done a lot of research into a number of different industries and kind of asked the question to these business leaders. How long do you think it'll be before you see a return on investment, let's say, to your lines of business? And in some cases, they're saying four to five years, right?
So they are absolutely investing today. They are working very diligently and urgently today. But some of these problems are so complex that it's going to take that long. And in some cases, even as much data as there exists today, there still isn't enough to really gain the benefit that they need.
So there are a few other factors that come into play as well. But I would expect certainly over the next few years that it'll become more ubiquitous, the use of supercomputing, supercomputing capabilities to solve a greater number of these AI problems.
If you wanted our audience to take away maybe three things from this interview that will really help decision makers when it comes to their future thoughts and interpretation of AI, what would those be?
Well, first of all, I'd start with skills. This is an absolute essential starting point, and it's the number one enabler. The second would be around supercomputing as well. This is something that is inevitable as organisations scale up, and is a fundamental building block to the ultimate realisation of the promise of AI. And then the third one would be-- I kind of call it this degree of patience. This is going to play out over a number of years. There's certainly a sense of urgency. But that degree of patience, continue to invest even if there's some stumbles along the way, that the promise is absolutely there.
As you say, the promise is there. I mean, it's exciting times when it comes to AI when we look into the future and see all the incredible developments that are happening. It's absolutely fascinating. It's been great to get a further insight into AI and how it's going to affect all of us in the future. Per Nyberg from Cray, thank you very much, indeed.