Online supercomputers that learn from mistakes will drive the next hi-tech revolution

An online supercomputer with software capable of learning from its mistakes will soon be available for businesses and individuals. Jamie Carter takes a look at a hi-tech revolution

PUBLISHED : Friday, 04 April, 2014, 9:56am
UPDATED : Friday, 04 April, 2014, 10:25am

Data is everywhere. We live our lives online, and voluntarily give websites and retailers our details in exchange for something. It's how we live our lives, and it's fast creating the age of "big data" where websites can track and predict our every action and desire.

But how do we crunch all that data to truly give the internet a mind of its own that we can all use? Step forward the supercomputer, the coal in the furnace of the internet that will drive the next revolution.

Supercomputers are what, one day, will make Google provide accurate answers to questions, and perhaps even guess the question.

When future generations look back to where the internet started, it's likely that they'll pick November 2013 as the point where a tangle of information became a living, breathing service indistinguishable from modern life. For last year saw the arrival online of IBM's Watson, a supercomputer that promises to bring artificial intelligence (AI) into the mainstream by making it into an app.

IBM has created a developer cloud for Watson so that apps for both people and businesses can be created that rely on its unique cognitive computing skills. Forget about robots and sentient machines - that isn't going to happen. Watson is all about software, but it goes way beyond mere logic.

Watson could be used to search through millions of pages of academic research and drug trials, something a human could never do or keep up with, and in double-quick time
Joe Peppard, professor at the European School of Management and Technology in Berlin, Germany

Reasoning, perception, social and cultural awareness are all within its grasp, but the kingpin of the new age of the internet is Watson's ability to learn from its mistakes. The end result could be computers not with AI, but GI - general intelligence.

Watson is based on a serious piece of hardware. A modular supercomputer running on 16 terabytes of RAM, it's comprised of a cluster of 90 servers all networked both together and, since November, to the internet at large.

So-called dual core or even quad core smartphones, laptops and TVs are common, but Watson has 2,880 cores and runs at 80 teraflops, making it capable of about 80 trillion operations per second. That, and around 300 algorithms operating inside, enable the real miracle - DeepQA.

IBM's smart learning software, DeepQA allows Watson not just to understand, but also to interpret written or spoken questions. DeepQA also allows it to search databases and nuggets of information, as well as spot patterns and make predictions.

Such intelligence in the cloud, which could theoretically be called upon via apps on a smartphone or tablet within a few years, are seen by many as crucial - and long overdue.

The financial and medical industries are currently drowning in data, most of which they lack the processing power to analyse.

"Watson could be used to search through millions of pages of academic research and drug trials, something a human could never do or keep up with, and in double-quick time," says Joe Peppard, a professor at the European School of Management and Technology in Berlin, Germany, which has a strong focus on the management of technology in business, and consults for large corporations on IT strategy. "It could provide a doctor or nurse with the required diagnosis and even the most appropriate medication plan, having taken into consideration all the latest thinking on any specific health problem."

The "Watson effect" is predicted to spread throughout the world of work. "I can't imagine a profession that isn't going to use this," says technology commentator Peter Cochrane, former chief technology officer and head of research at British Telecom.

"If you look at physicists the half-life of their knowledge is 13 years, for marine biologists it's 2½ years, and in the technology industry it's nine months. There will be a bunch of services where professionals can call up and ask a question. I hope it wipes out all the investment bankers." Or just their jobs, perhaps.

An "Ask Watson" app that allows working professionals in data-driven industries to access the very latest research - predicted to go online within a year - will spread in society.

One of the first places we'll probably see evidence of supercomputers is in call centres.

"Currently, about 60 per cent of call centre inquiries do not result in a satisfactory answer, but Watson is able to search a far greater amount of data in a much faster time," says Peppard.

"By doing so it would be able to assist with inquiries and, in many cases, pre-empt those inquiries so that the answer is available before the question has been asked," Peppard says.

What do you do with a supercomputer that has the ability to accurately guess what a searcher is looking for? You turn it into a web search engine, of course. So it should come as no surprise that Google is showing intense interest in the kind of AI exhibited by Watson.

Cochrane thinks it's about time AI was applied to web searches to upgrade the existing dumb technology. "Google is simultaneously absolutely brilliant yet also quite useless since it only tells you what the top 10 searches are," he says.

"What you really want is for Google to watch everything you do, understand what you do, anticipate what you might need, and bring you the 10 most relevant search items from the 37 million items it's found. I think we'll start to see this kind of thing trickle out over the next year, and it will be a little bit at a time, and it will be fascinating."

The foundations for the "machine learning" era have been laid. "The combination of big data, the cloud and the storage of personal information in the form of cookies means that internet searches are changing - and have been for a while," says Peppard.

"This is really about the dawn of a new age in cognitive computing - it's about computers being able to learn."

Perhaps the most obvious area where machine learning and AI can play a part is in speech-based services such as the iPhone's Siri personal assistant and Android handsets' Google Voice. What would a supercomputer-fuelled Siri look like?

"Instead of tapping a search into your phone and the internet coming up with an answer, you give verbal commands that the handset can action. An example would be 'Pay my water bill' - and it's done," says Peppard. "But it could also learn when these bills need to be paid and do it automatically - or remind you to do it."

Where Watson will really differ from Siri is in its ability to learn and improve. The first apps to use Watson are likely to be costly and limited to professionals, although the online supercomputer is here to stay.

Watson is just one of 252 supercomputers in the US; China has 66 - including the world's fastest.

It's a power play that's taking place nearby; housed in the National Supercomputing Centre in Guangzhou is Tianhe-2, which runs at about 33.86 petaflops - that's over 33 thousand trillion actions per second compared to Watson's trifling 80 trillion - and the exaflop computer isn't far away.

An exaflop is one thousand quadrillion, although by the time that gets online we'll all be reminiscing about when we all used Wikipedia to make ourselves look clever.

[email protected]