Google DeepMind founder and leader in artificial intelligence returns to Hamilton

A leader in artificial intelligence first honed his skills at the University of Waikato.

Now, after launching a computer program with the ability to learn on its own, he has returned to accept a Distinguished Alumni Award.

Dr Shane Legg​ arrives at the Hamilton campus on Tuesday, and will trace the footsteps he first walked in 1993.

He graduated in 1996, when the internet was a relatively new mechanism, and soon after went on to co-found Google DeepMind.

READ MORE:
* How New Zealand can thrive in the age of AI
* Google shifts focus to AI
* An AI program is beating poker champions for the first time

Launched in 2010, the company is widely known for creating a computer system modelled on the human brain that learns how to play video games.

Their AlphaGo​ program beat a human professional Go player for the first time in 2015, while AlphaZero beat the most powerful programs playing Go and chess after only a few hours of playing against itself.

DeepMind​ was purchased by Google in 2014, with Legg​ as its chief scientist.

Legg's​ curiosity in computing began as a child, but it wasn't until his tertiary study that he was able to sharpen his skillset.

"I've been programming since I was 9-years-old, so back in 1982, I owned a pretty small computer, but we didn't have the internet.

"My first taste of the internet was at Waikato Uni. That's when I got my first email address and web browser and discovered there was a huge world out there," he said.

"It was all very new then, it was before Google or anything like that."

University of Waikato was the pioneer of bringing internet to New Zealand, with the first connection made by John Houlker​ in April 1989.

It also became the first university in the country to have cyber-graduates completing teaching degrees online, so it was fitting for the former Rotorua Lakes High School student to complete his undergraduate studies there.

Legg​ went on to study at the University of Auckland and obtained his PhD from the Dalle​ Molle Institute in Switzerland.

In 2009, he jetted off to the UK.

"After I finished my PhD in Switzerland I went to University College London and there's a research institute there called the Gatsby Unit. 

"When I was there, I was really interested in artificial intelligence, neuroscience and so on."

It was at the Gatsby​ Computational Neuroscience Unit where he met Demis Hassabis​, who introduced him to Mustafa​ Suleyman​, and DeepMind​ was formed shortly after from a shared passion and grand plans.

"We were believers that great things were going to happen in this area in the coming years," Legg​ said.

"When we started DeepMind​, we had grand plans. We really believed that machine learning and artificial intelligence were certainly going to take off, and if that did happen, we'd be able to grow DeepMind​ into quite a large organisation.

"We had big plans and, amazingly, the plans have worked out. You still have to pinch yourself when you see the reality of it."

Despite being at the forefront of artificial intelligence for some time, Legg​ said he was still regularly in awe by what machines were able to achieve.

When DeepMind​ program AlphaGo​ challenged 18-time world Go champion Lee Sedol​ to a five-game match, AlphaGo​ won all but the fourth game.

DeepMind​ states that their system learns from experience, and the Go victory was seen as being far ahead of its time.

"The Go results were amazing, it was incredible to watch that unfold," Legg​ said.

"We knew we had a powerful Go player and we were confident we would win most of the games, but we also knew it wasn't perfect.

"It was a good result but I can tell you we were really on-edge."

Legg​ also said he was often fascinated by the learning capabilities of artificial intelligence.

"It's the fact that you see something work at a problem and, with its fumbling and making blunders, it improves and learns.

"In not all cases but some, it will then go beyond the level we can play. It surprises you sometimes with the very creative moves that it comes up with."

Among the many areas Google DeepMind​ is prevalent in, Legg​ said they were working towards developing a program that could conquer StarCraft​, a real-time strategy game that no form of artificial intelligence has learned to play at a high level.

But even New Zealand's most influential global Kiwi in the technology field said there's no way of knowing what the future holds for artificial intellegence.

​"Nobody knows what it's going to look like in another 10-20 years. 

"I am quite confident, talking to people in hardware, that at least for the next few years, the performance of microprocesses​, the amount of information they can process, the amount of mathematical calculations they can perform in a second, will keep growing very rapidly for a few years," he​ said.

"After that, it's anybody's guess what will happen."