Businesses must design algorithms to take human flaws into account at every stage of the process, the mathematician Hannah Fry has said.
Speaking to NS Tech at IP Expo on Wednesday (3 October), the UCL academic and author of Hello World urged businesses applying machine learning not to view it in isolation.
“You need to think about the humans using algorithms and all of the pitfalls we have,” Fry said. “There are occasions when we just expect a little bit too much of our humans.”
Fry called out manufacturers who have built cruise control systems requiring drivers to step in to avoid a crash. “Being expected to operate at the highest skill level in the most critical scenario with no warning: that’s something we are incredibly bad at,” she said.
Tesla and Toyota, meanwhile, have recently built automated systems which reverse the process, taking control of the vehicle in emergencies. “These are really strong examples of where driverless car companies are taking into account humans’ flaws,” said Fry.
The complexities of adopting machine learning and artificial intelligence were a key theme at IP Expo. Speaking on a panel on the topic on Wednesday afternoon, Paul Ducklin, a senior technologist at security vendor Sophos, said machine learning was not a silver bullet for cyber security.
“You need to be careful of assuming that because you’ve ticked that one box [deploying machine learning], that one box is more important than everything else,” he told delegates.
Dan Yin, head of innovation at RingCentral in Europe, Middle East and Africa, added that some companies are branding their services as artificial intelligence to win funding and business. “It’s a fantastic marketing tool,” he told attendees.
Increased automation is regularly touted as a way of addressing the UK’s productivity crisis. But a Royal Society of Arts survey of business leaders last year indicated that just 14 per cent of companies in the UK had invested in AI or robotics, or were planning to in the near future.
“Many think that the technology is too costly or not yet proven,” the RSA’s researchers wrote. “For others, concepts such as machine learning, deep learning and cloud robotics appear to be completely new.”