HPC meets AI - a world of possibilities
- 28 September, 2017 13:23
Artificial intelligence (AI) will transform the technology landscape and touch almost every industry over the next 10 years.
That’s according to a group of experts (from government, industry and academia), who gathered in Sydney for an exclusive Computerworld and Lenovo Roundtable lunch to discuss the implications of AI and the emerging use cases.
Certainly, no domain of computing will escape the transformation, including high-performance computing (HPC), and there’s a definite link between AI and HPC, according to Lenovo enterprise solutions architect Joao Almeida.
Indeed, the marriage of HPC and AI has already taken place, he noted. The rapid emergence and adoption of AI is creating a world of opportunities for both scientific research and businesses.
Asked ‘why now’ and why it’s important, particularly for business, Almeida said companies are now starting to see the true business value of the technology - particularly the ability to automate routine processes, improve overall business efficiencies, upskill the workforce and transform the user experience.
“It became important quite a long time ago, but the markets and the enterprise, in particular, has only now started to come across it and are seeing possible business value out of it. Research in AI has been going on for a very long time and it has always been very close to HPC.”
Asked just how close, Almeida said HPC is the engine that helps deliver the promise of AI - the technological building blocks that help make it all happen and fuel the possibilities.
“To actually build an AI something - and there are a lot of definitions for AI and machine learning (where it stops and when it starts) - you need to do a lot of work before you can actually use it,” he explained.
“Normally, you have a lot of unstructured data - or structured data - that you put through a model and that requires a lot of computation. And techniques that came from HPC are used a lot in that part of the workflow. Technology developments like GPU usage on that modelling and training phase of AI has been using HPC for a very long time.”
And while there’s opportunity in every industry sector, the oil and gas market, in particular, is already latching onto the technology and realising tangible operational cost benefits, Almeida noted.
“Efficiencies, for example, is probably one of the areas where companies in the oil and gas industry can see the benefits. Improving efficiencies by one per cent is only possible if they have a very good AI model looking at millions and thousands of sensor data and creating models that can improve efficiencies.”
From the doomsday ‘science fiction’ use case scenarios (where robots create havoc and force robotic domination) to the real-world everyday uses, AI is a hot topic, and one that sparks conversation on all levels from the ethical and moral issues, to implications for upskilling the current workforce (and concerns over job losses), as well as the practicalities around how businesses can embrace this new form of technology.
And while attendees discussed some real-world applications that are likely to benefit from the AI computing transformation, many at the table admitted they’re still either considering the technology or slowly dipping their toe into the water as they continue to progress on their digital transformation journeys.
Indeed, it is an growing market that has enormous opportunities for business, even for the ones starting on the journey, according to Lenovo ANZ head of technical sales, Shayne Harris.
He said the intersection of HPC and AI is creating a vibrant new market and fueling the growth of AI platforms and products.
“Lenovo inherited a large legacy of high performance computing, which is effectively the engine for AI and many other things. One of the great things for us is we’ve been able to build on that over the last three years, which is a fantastic opportunity to talk to our wider customer network about,” he said.
“AI is being driven by the demand in the market and historically it has largely been science and technology, research and education, and now large industry, like banking and finance and energy, are starting to really delve into it. We’re starting to see that take place in the wider market.”
He said the wider large enterprise market is starting to realise the applications of AI and the efficiencies they can gain from that. It’s either one of two things: problems that can be solved in the organisation or the ability to make the organisation or business better.
“We are effectively the engine behind high performance computing and the engine that enables organisations to develop solutions using AI. We are just one of multiple parts of that engine. But for us, it is an important part. We are building this high performance computing platform to enable organisations to develop solutions for artificial intelligence. It is a massive ecosystem that’s only starting to get bigger really quickly now.”
AI gets real
Attendees agreed operational efficiencies present some of the biggest opportunities associated with AI. In other arenas, like healthcare and aged care services, meanwhile, AI can truly have an impact that goes beyond convenience and positively affects human lives.
Take for example The Salvation Army. Paul Berryman, head of IT, aged care plus support services, said AI is the use of machines to learn behaviours and make decisions and/or recommendations for humans. It almost always involves large and/or complex data sets, which would cause humans to take too long.
He offered up some real-world examples. “Applicable use cases of AI would include pattern recognition and then detecting exceptions or trends. In a home care case, this might be detecting a trend that Mum is getting up progressively earlier by a few minutes each week - this could spark a conversation with the doctor and Mum about what this means - is it related to bladder function, that she needs to get up progressively earlier.
“Alternatively, it could be that Mum always makes her tea between 7:30am and 7:40 am - the system has detected that the fridge door opens at this time each morning. I might get an alert to say it's 7:45 and Mum hasn't opened the fridge door, alerting me to give her a call and ask how she's going,” he said.
“Opening or closing doors and managing behaviours of dementia residents might also be a good case for AI. Machines can learn patterns and allow us to manage particular behaviours, such as getting up in the night or wandering into the wrong room.”
Over at Bupa ANZ, Syed Ahmed, head of digital products and operations, said proper AI is when the software can move beyond simple pattern matching and regression analysis-based prediction to autonomous decision-making in open, unpredictable and ambiguous systems.
Asked some of the applicable use cases of AI in business, Ahmed said it revolves around analysis and advisory.
“Within the current legislative and socially acceptable frameworks, most use cases of AI in business relate to assistive analysis and advisory. Examples of this are financial portfolio optimisation, financial fraud prediction, and medical diagnosis.”
Meanwhile, over at DHL Supply Chain, AI is an extension to robotic process automation (RPA), according to Kasi Kolla, global leader in outsourcing infrastructure, ITSM and cloud strategy.
Kolla said AI promises to give machines flexibility to learn and evolve to help humans.
“In my opinion, in the next few years humans in most positions in the world of work will be nearly 100 per cent replaced by or partnered with smart software and robots (AI).”
Over at JLL, a global real estate services firm specialising in commercial property and investment management, AI is process automation or efficiency undertaken by machines in replacement of human interaction, according to regional director of integrated portfolio services, Jordan Berryman.
“AI in business is utilised for repetitive action roles, trend analysis or cognitive matching, At JLL and in reference to our clients that we assist, AI is employed from a range of roles including our robotic receptionist ‘JiLL’, trend analysis from data analytics and forecasting, to environment and security control of buildings and automation of supply chain functions,” Berryman said.
Meanwhile, over in the education space, Neil Fraser, Macquarie University director of information, said most successful AI tasks (be they supervised or unsupervised), are replacing activities beyond our human capability.
“Plenty of definitions are out there on AI but for me this is about the age of machine learning (supervised and unsupervised) which is where we’re applying statistical approaches and modelling to data. The transformational shift is the massive drop in the cost of computing in parallel with an unprecedented explosion in data volume and variety.”
He said the opportunities for the education space come “anywhere where this transformation shift threatens the underlying value chain of the business and data products can find a footing.”
Connecting the dots
No matter the industry, Lenovo’s Almeida said businesses need help putting the pieces together.
Asked how companies can prepare to get onboard the journey, he said businesses need to consider: How does AI benefit my business? How does it apply to my business? How can businesses use it? And how can they turn what’s typically considered a start-up scenario into a business outcome?
“We engage with customers, we show them examples of AI in businesses similar to them, then we develop that idea, which applies to their business, and customise it to show them how that would apply to their business, and then we help them deploy that by creating a system. Training the system and delivering a workload or a workflow, or a system that they can actually utilise.”
When looking at the AI workflow, he said several things come to light. “One is you need to have a lot of data because the data is used to train the system - that’s the training phase of AI. If you have unstructured information or unstructured data, or structured data, it doesn’t matter. Then you train the system.
“Then you have an inference point where you actually consume the information. The information from that trained system. And we stop right there. We help the customer create that training environment, create the model to train. Then when it comes to inference and utilising, that’s where we hand over to the business. That’s now part of the business to move along and create whatever it is that they want to be doing with it.”