Cognitive computing comes to Canberra

Cognitive computing has come to the Australian Public Service this year, first in a customer service role with IP Australia and now as part of the profound transformation of the Immigration and Border Protection portfolio and its focus on expanding intelligence capabilities.

Big data is a big part of the consolidation process that is merging the traditional immigration and customs functions into a single entity to manage flows of people and things in and out of the country, with the Australian Border Force soon to emerge as its uniformed operational arm, relying on intelligence it gathers from a wide range of public and non-public sources.

The government says the ABF, which received a $50 million funding boost in yesterday’s budget, will be “intelligence-led, mobile and technology-enabled” and IBM’s Watson — a computer that can think, learn, process normal language and beat anyone at the TV quiz show Jeopardy! — has been drafted in to help with that.

DIBP first assistant secretary Randall Brugeaud made the announcement last month, a few days before he finished acting as a deputy secretary heading up the Intelligence and Capability division, a role now taken on by former Australian Geospatial-intelligence Organisation director Maria Fernandez. The former Customs chief information officer said DIBP was taking the lead and “working closely with a number of partner agencies” to see what else Watson could do in the APS.

Brugeaud explained that our new border guards hope the technology can “rapidly expose connections between otherwise isolated threads” in the “very noisy world” of the internet. They want to vacuum up vast amounts of information designed for human eyes and mine it to get the drop on smugglers and other criminals, but there’s too much of it. Watson won’t suffer what Brugeaud calls “information overload” but can still sift through this unstructured data in a thoughtful way to help the ABF find needles in haystacks.

Interest in the cognitive computing platform’s commercial and government applications came soon after it demonstrated it could wipe the floor with game show champions in 2011. Jason Leonard, who leads the IBM Watson business in the Asia-Pacific, says oncologists were first to make use of its capacity to support evidence-based decision making, and they were soon followed by large enterprises who saw its customer service potential.

“The real world is a little bit more complicated than the quiz-show world, and we’ve been pretty busy in the past few years working that out,” Leonard told The Mandarin. Along with IP Australia, which took up the system in February, early adopters of this aspect of Watson’s abilities include the Singapore Government and Deakin University, who both use it to get around the problem of information siloes. Patricia Kelly, IP Australia director general, said it was an “opportunity to improve how we deliver efficient, timely and high quality services to the public.”

Give Watson access to all the information spread across hundreds of government websites — far more than one call centre can handle — and Leonard says it could become the “concierge” for all government services.

“Singapore and Deakin are both at about the same stage of maturity,” he explained. “They’re both live, they’re both up and running and we’re now working to expand the range of questions and the level of personalisation of the answers, so the way it’ll evolve in the short term — the next 3-4 months — is that rather than giving an answer that may be useful, but maybe isn’t very specific to your circumstances, [Watson will start] taking into account maybe the previous questions that you’ve asked in order to give you a more specific answer, or any other information that you might want to volunteer.”

Listening, looking, learning

Watson demonstrated an ability to answer verbal questions on Jeopardy! but most people prefer to interact with the artificial intelligence by tapping a smartphone screen or tablet. “It’s like if you talk to some large organisation with an online chat facility, where there might be a person at the back end of that, so you’re asking questions but the entity answering you is Watson,” says Leonard.

Understanding normal language, whether in published reports or conversation, is pretty impressive. But Watson has also begun to peer at images to absorb visual information as well, from medical applications like looking at x-rays to recognising familiar people on sight.

Watson is a fast learner with a can-do attitude, but still needs training to make use of its skills in a new specialist job like border protection intelligence. “A programmable computer on its first day is having its best ever day, because after that it might start to decline; it will become obsolete some time in the future,” explained Leonard. “From Watson’s point of view, its first day is probably its worst day, because it hasn’t had much experience; it hasn’t seen as many different circumstances pop up so therefore it hasn’t learnt as much.”

The Watson system does not live in a world of black and white, either. It can make educated guesses about things it doesn’t quite understand, and tell you how confident it is about an answer to a question you ask.

“Now, if it hasn’t received any training, the confidence level it has could be quite low because maybe it doesn’t really understand your question as well as it might after a while longer,” said Leonard. “It could come back and say: ‘I’ve got low confidence on that, can you tell me this extra piece of information?’ In which case, you give it that piece of information and the confidence will fire up. That’s one possibility — you haven’t given it enough information [about the specific question] — or the second possibility is you haven’t given it enough training, and it might have low confidence because it hasn’t seen enough examples to make a really strong match with what you’re looking for.”

Of course, there’s less theatrics in Watson’s serious roles than when it appeared on television and its physical footprint is much less grandiose. “It’s probably better to think of Watson as a whole range of services that can do interesting things,” said Leonard.

Those individual services — it can also gauge the “sentiment” of words, for example — are growing in number all the time and generally offered via the cloud, but can also be applied to sensitive data stored at secure sites, he says.

“IBM will create some of those applications for particular industries and organisations that we’re primarily focused on, but then we have a whole range of ecosystem partners, who have their own ideas about the applications they want to build using these cognitive services.

“And then of course organisations that have enough capacity to do so … can put together their own applications using these cognitive services, so it’s probably best to think of Watson as a platform of services that you can mix and match according to what you’re trying to do.”

Read more at The Mandarin: Australia has a ‘false understanding of privacy’, says Europe’s top rated CIO