Dr John Zakos

Griffith PhD on three decades at the forefront of cognitive computing

Dr John Zakos was 14 when he authored his first artificial intelligence software, a “virtual person AI” he dubbed Let’s Talk.

“I was fascinated with how we could program computers to be humanlike and came across AI without knowing it was called AI,” he laughs.

“I then did postgrad research in the field, looking at AI for breast cancer diagnosis, and took it from there.”

Where the technologist and entrepreneur “took it” was co-founding Cognea, an AI company specialising in chat and virtual agent products that was bought by IBM in a high-profile 2014 acquisition.

Cognea developed a cognitive computing and conversational AI platform to drive a virtual assistant that related to its human users through realistic personalities.

In 2008, Cognea was recognised as one of Australia's top 100 web companies by BRW magazine's influential Fast 100 list.

As part of the sale to IBM, Dr Zakos also became a program director for the tech giant, where he helped develop the next generation of cognitive computing and integrated Cognea with IBM’s own chat platform, Watson.

He’s often asked about “the future of AI” and says some of the most exciting work in the fast-growing sector is in deep learning, where a platform processes information like a human mind.

“Ideally, we will build AI systems only be giving them raw data and the goals we want them to achieve and leave the rest to the AI platform,” says Dr Zakos.

In 2017 he was made an honorary professor at the University of Technology Sydney, where he currently advises on R&D, entrepreneurship and commercialisation, and industry collaboration.

“One of the most rewarding aspects of my career is being able to inspire young people to pursue excellence in the [IT and AI] field,” says Dr Zakos.

"You can read books and learn all day, but if you’re not dreaming and imagining things, you’re not doing much in terms of innovating for the future.”

At Griffith, Dr Zakos graduated first with a 1997 Bachelor of Science and was back the next year for a Bachelor of Information Technology and subsequent PhD.

He holds several tech patents and was published internationally throughout his academic career on topics including information retrieval on the web, chatbots and AI language recognition.

The key, he says, to a successful AI is its ability to ‘generalise’, or successfully extrapolate from a conceptual baseline.

“If you can achieve generalisation, then you have something truly intelligent,” he explains.

“For example, if we want to build a system to recognise furniture and we train it by showing it three-legged chairs, it should have no problem recognising four-legged chairs as chairs also.

Generalisation, he says, is achieved by coupling the right methodology with the right data to build the AI.

“And it's never a one-round process,” he laughs, “we are forever retraining and fine-tuning to get it just right.”