What’s the secret to making sure AI doesn’t steal your job? Work with it, not against it

Whether it’s athletes on a sporting field or celebrities in the jungle, nothing holds our attention like the drama of vying for a single prize. And when it comes to the evolution of artificial intelligence (AI), some of the most captivating moments have also been delivered in nailbiting finishes.

In 1997, IBM’s Deep Blue chess computer was pitted against grandmaster and reigning world champion Garry Kasparov, having lost to him the previous year.

But this time, the AI won. The popular Chinese game Go was next, in 2016, and again there was a collective intake of breath when Google’s AI was victorious. These competitions elegantly illustrate what is unique about AI: we can program it to do things we can’t do ourselves, such as beat a world champion.

But what if this framing obscures something vital – that human and artificial intelligence are not the same? AI can quickly process vast amounts of data and be trained to execute specific tasks; human intelligence is significantly more creative and adaptive.

The most interesting question is not who will win, but what can people and AI achieve together? Combining both forms of intelligence can provide a better outcome than either can achieve alone.

This is called collaborative intelligence. And this is the premise of CSIRO’s new A$12 million Collaborative Intelligence (CINTEL) Future Science Platform, which we are leading.

While chess has been used to illustrate AI-human competition, it also provides an example of collaborative intelligence. IBM’s Deep Blue beat the world champion, but did not render humans obsolete. Human chess players collaborating with AI have proven superior to both the best AI systems and human players.

And while such “freestyle” chess requires both excellent human skill and AI technology, the best results don’t come from simply combining the best AI with the best grandmaster. The process through which they collaborate is crucial.

So for many problems – particularly those that involve complex, variable and hard-to-define contexts – we’re likely to get better results if we design AI systems explicitly to work with human partners, and give humans the skills to interpret AI systems.