University buys supercomputer to keep NZ in the AI race.
New Zealand is scrambling to stay ahead of the artificial intelligence curve as it loses its early advantage and others ramp up their investments.
Academics argue we can’t ever hope to be at the front of a pack led by China and the United States, but they say falling too far behind could mean losing control of key systems in a future where artificial intelligence (AI) will be critical to national prosperity.
Nationally and globally AI is being seen as a major growth opportunity. Management consultants at PwC and McKinsey have touted massive numbers around what this technology could add to global GDP – US$18 trillion (NZ$25t) by 2030 in PwC’s case and US$13t in McKinsey’s.
An online conference hosted by Bloomberg last week revealed chief executives around the world ranked AI ahead of Covid19 in terms of the potential longterm disruption it could cause to their businesses.
A draft strategy on AI is floating around the corridors of power in New Zealand with a discussion document and then a final strategy likely to be released by the end of this year.
University of Waikato Professor Alfred Bifet says New Zealand needs to be strategic about its investment in AI because more automation will make these sorts of technologies a lot more important in future.
‘‘This could change how many things are done. And for countries this is really, really important . . . in Europe now they’ve realised this and they are also investing a lot of money but I think they are a bit delayed.
‘‘This is why it is really important for New Zealand to have a national strategy because it’s very important to decide where to focus.
‘‘It’s clear that with the size of New Zealand we cannot compete with the US and China in everything, so we need to decide what are the most important things that we could make a difference in.’’
As part of New Zealand Inc’s quest to keep up, the University of Waikato has bought the country’s most powerful supercomputer dedicated to machine learning, and it also plans to set up an institute dedicated to artificial intelligence which Bifet will head.
Two weeks ago representatives of seven universities converged on Hobbiton for a discussion about the new artificial intelligence institute and the university’s supercomputer purchase.
Bifet pulled an old party trick at the meeting that he often uses at gatherings like these.
He pulls up Google’s data on most-searched terms in the US and compares the number of people who mention Hobbiton,to the number who mention a piece of domestically-developed machine learning software.
The Waikato Environment for Knowledge Analysis (Weka), developed at the university, always wins.
‘‘It is really, really prestigious in all the universities around the world, but this is something that is not very well-known in New Zealand,’’ Bifet says.
The university has used the sale of Weka commercial licences to fund the purchase of the Nvidia DGX A100. Other supercomputers like Niwa’s machines are more powerful but dedicated to analysing things like metereological patterns. This will be the most powerful computer in the country dedicated to artificial intelligence.
It sits in near the bottom of an unassuming stack of nondescript black boxes underneath the university’s library. In some ways you can’t miss it. The machine’s exterior is gold-plated with a large Nvidia logo on it.
Is there any reason why it’s coloured in gold? Not really, it just looks better.
Then there’s a deafening roar. Something akin to a jet engine. That’s the sound of two doctoral students logging onto the server, Bifet explains.
The noise comes courtesy of a network of powerful fans which fire up to cool the computer down as it works through various AI models.
Which is why it’s chilly when you stand in front of the machine, but neartropical when you step into the small area between racks where the temperature jumps to a coat-shedding 30 degrees Celsius.
The computer is using artificial neural networks – computer software styled on the human brain – to learn how to make predictions in particular areas using a process called ‘‘deep learning’’. The model makes its own predictions, tests these against real-world results, then is trained by humans to recognise what went wrong..
Artificial Intelligence institute associate director Jannat Maqbool says that if we don’t retain some AI capacity domestically, our companies will be at a disadvantage, especially if overseas players create products which don’t really suit our own needs and priorities.
Some of these could be environmental, such as local weather forecasts or predicting demand for renewable electricity.
Maqbool says New Zealand also has specific needs because of digital connectivity issues around the country.
‘‘We don’t always have cloud access to be able to send data out of a forest or even 45 minutes out of Hamilton.’’
So, if our primary industries are going to be able to use these technologies they won’t be stored in the cloud, they’ll have to be located closer to where the crops are being harvested or the cows milked – which might not be a development priority for AI specialists overseas.
Computer and mathematical sciences Associate Professor Te Taka Keegan says a top priority for him is ensuring AI delivers solutions for Ma¯ ori as well.
Overseas players aren’t going to bother designing neurolinguistic models which might allow machines to respond in Ma¯ ori or to predict how people from different iwi groups might react to situations. They won’t be too interested in creating healthsystem AI which might produce tailored health solutions for Ma¯ ori either.
Despite the enthusiasm, Bifet knows there is also a vein of scepticism around AI as well, thanks to the industry’s history of ‘‘AI winters’’.
These are where overoptimistic predictions and overinvestment in AI lead to massive underinvestment after its initial promise falls away. Usually this is because development hits an unforeseen technological ‘‘wall’’ of some kind.
His counter-argument to whether we will hit a similar wall this time is that many of these deep learning technologies are already being used to power common consumer applications on phones and computers.
Bifet acknowledges one wall we could hit is around the machine’s ability to reason.
We might be coming to grips with deep learning models, but we have been less successful at producing computer programmes which have a real understanding of why things are happening. Still, he’s optimistic we can find some way to increase the capabilities of AI models in this respect too.
‘‘It’s the future. We cannot predict what’s going to happen.’’
Weka, a piece of machine learning software developed at Waikato, is used by universities around the world but is not wellknown here.