20 Jan 2017

The human intelligence behind artificial intelligence

2:50 pm on 20 January 2017

What would be your legal rights when coming up against a robot? Who would you sue if artificial intelligence denied you promotion, or even a job in the first place? What if your neighbour's future house-cleaning robot blew a fuse and crashed through your hedge or front window?

A legal team from Otago University is going to try to find the answers to these questions.

Robot 'Nao' addresses the opening ceremony of the 2016 RoboCup China Open in Hefei, China.

Robot 'Nao' addresses the opening ceremony of the 2016 RoboCup China Open in Hefei, China. Photo: AFP

Colin Gavaghan heads a research team that has just been paid $400,000 by the legal thinktank, the Law Foundation.

His task is to find out how legal practice can evolve to deal with new technologies such as driverless cars, crime prediction software and even virtual lawyers themselves.

Dr Gavaghan said when things went wrong with mechanised or automoted machinery, liability usually relates to either the manufacturer of the machine, or the user of the machine, getting something wrong.

"What is different about artificial intelligence, or what has been referred to as smart robots, is that in a weird kind of way, they are making decisions for themselves.

"They are capable of learning and they are capable of interacting in all kinds of complex ways with the environment, and that makes the whole chain of responsibility a bit more difficult to establish."

Other forms of artificial intelligence (AI) also needed careful analysis.

In some cases, algorithms had been developed that told police officers where crimes were likely to occur and even who was likely to commit them.

"Predictions about danger and risk are important, and it makes sense that they are as accurate as possible," he said.

"But there are possible downsides.

"AI technologies have a veneer of objectivity, because people think machines can't be biased, but their parameters are set by humans. This could result in biases being overlooked or even reinforced.

"Also, those parameters are often kept secret for commercial or other reasons, so it can be hard to assess the basis for some AI-based decisions."

Dr Gavaghan said an American law firm claimed to have hired its first AI lawyer to research precedents and make recommendations in bankruptcy law.

"Is the replacement of a human lawyer by an AI lawyer making the lawyer redundant, or more like replacing one lawyer with another one?

"Some professions - lawyers, doctors, teachers - also have ethical and pastoral obligations. Are we confident that an AI worker will be able to perform those roles?"

Dr Gavaghan suggested a couple of ideas that might work to contain these problems.

One had been proposed by the European parliament and involved keeping a register of smart robots so they could be properly monitored.

Another was to establish a rule of strict liability to make sure owners of robots were incentivised to ensure they had the money to pay for damages when robots go wrong.

Alternatively, there could be some kind of national insurance scheme to fund AI disasters from a common pool of money.

This research project will take three years to complete.

Get the new RNZ app

for ad-free news and current affairs