The disturbing rise of racist robots

Are robots racist? AI computers are showing sexist and racist bias. A growing number of experts fear that we have created AI in our own worst image, and it could have awful consequences. You’ve probably seen the images popping up on social media. Pictures of your friends and acquaintances, their faces framed by green boxes and strange labels. It’s all down to ImageNet Roulette, an artificial intelligence (AI) craze sweeping the internet. It’s a simple idea. Upload a selfie to an AI computer, and it will categorise you based on its knowledge of over 14 million photographs from across the internet. Some of the labels it gives out are flattering — who doesn’t want to recognised as “leader” or “president”? — and others are less so: “swot”, “nerd”, “dweeb”. But some are disturbing. Chinese-Jewish writer Julia Carrie Wong wrote in The Guardian that the programme called her a racist slur. Many more people of colour have reported offensive labels. You might think something went horribly wrong but, in fact, ImageNet Roulette is working exactly as its creators intended. “We want to shed light on what happens when technical systems are trained on problematic training data,” they wrote. AI computers are fed data to help them learn and perform their tasks better, but that data comes from a world peppered with prejudice and discrimination. As they learn the data, robots are also learning to mimic our biases. There are other examples: a LinkedIn search that prioritised male profiles; a ChatBot fired out anti-semitic messages after learning to “speak” from Twitter. “People expected AI to be unbiased; that’s just wrong. If the underlying data reflects stereotypes, or if you train AI from human culture, you will find these things,” says Joanna Bryson, an AI researcher. “It’s racism in, racism out,” is how activist Hamid Khan sums up the problem. And it is already having disturbing consequences. A US court started using a computer programme to assess how likely prisoners were to re-offend. Last year, a study found that the system was much more likely to mistakenly label black defendants as likely reoffenders. This happened to black prisoners twice as much as it did to white. But why? The research suggested that the data (arrest records, postcodes, social connections and income) the programme was fed can all reflect problematic stereotypes, and further ingrain them. “Police can say: ‘We’re not being biased, we’re just doing what the math tells us’,” warns Kristian Lum, who works for the Human Rights Data Analysis Group. This year, tech firms are expected to spend $35.8 billion (£31.8 billion) on AI development. What kind of future are we heading towards? Heart of darkness Are robots racist? For all the talk of machine learning, the bias in AI computers proves that they cannot learn like humans: they don’t understand morality, or institutional discrimination, or other unequal forces that shape society. We must beware of giving power and knowledge to something that is so hard and coldly rational, without a shred of human warmth. But surely it is wrong to say that it’s the robots that are racist? They are a blank slate, ruled by logic. Of course they can’t be biased. They are taught to be so by us. Corners of the internet teem with hatred; unemployment and income figures show constant racial inequality. AI is mimicking our worst instincts: it is we who need to reform. KeywordsArtificial Intelligence - Artificial intelligence, or “AI,” is the ability for a computer to think and learn. With AI, computers can perform tasks that are typically done by people, including processing language, problem-solving, and learning.