AI can cause unconscious bias – but it could also be the solution

20 Jul 2018

Alexa Gorman, global vice-president, SAP.iO Fund and Foundry Image: Conor McCabe Photography

Without diversity in machine learning, AI is doomed to have unconscious bias. But, it doesn’t have to be this way.

“I want you to tell me how many labradoodles you see.”

Alexa Gorman had the entire Inspirefest audience giggling within three minutes of walking on stage. She used a series of pictures split into a grid. Some had labradoodles, some had fried chicken.

She did this twice more to show to show the audience what machine learning is and how algorithms can be trained to identify things the way humans can.

Gorman is the global vice-president of the SAP.iO Fund and Foundry and her work has led her towards a particular focus on AI and machine-learning start-ups. She told the Inspirefest audience that both diversity and AI are very important to SAP.iO.

“The goal of machine learning is to teach machines basically to be able to do the tasks that we humans can do.”

Through explaining the process of feeding machines huge amounts of data in order for it to learn, Gorman was quickly able to highlight the problems that can easily occur.

If the data that the machine is fed contains unconscious bias, then the outcome of the machine learning will also contain that bias.

“It’s not that the programmer sits down and does this on purpose, it’s something that he or she just doesn’t know and that way, the algorithms of these outcomes become flawed.”

To hammer home the problem, Gorman showed an image search that showed mostly white men. The search term was simply ‘engineer’. On another one, where the word was ‘assistant’, the image search showed women.

“Here, you actually see that the only male on this picture is a cartoon character, which is even worse,” said Gorman.

She pointed out that machines and algorithms will only be as diverse as the engineers and data scientists who build them. These biases within algorithms can then perpetuate stereotypes.

It’s not all bad

Luckily, Gorman was not at Inspirefest to be the bearer of bad news. While she said that the fact that AI built with bias will not reflect the world around us, she added that this can still be flipped to become a positive thing.

“We also see opportunities for AI to recognise biases, to dispel stereotypes,” she said. “The technology is that good that we can actually use it to audit. We can audit websites, we can audit companies, we can audit governments to see: How diverse are they? What kind of biases do they include?”

Inspirefest is Silicon Republic’s international event celebrating the point where science, technology and the arts collide. Ultra Early Bird tickets for Inspirefest 2019 are available now.

Jenny Darmody is the editor of Silicon Republic