Remember what I told you a few articles ago about the Kenyan minister of technology promoting and talking up artificial intelligence (AI) software development for his country, and how the U.S. government has increased the funding for it by forty percent since 2015? I also wrote that AI is a State priority in China, where it is being used for various government processes including catching law breakers.
In Africa, it’s not just Kenya that knows the value of AI. Aubra Anthony, a strategy and research lead for the Center for Digital Development within the U.S. Agency for International Development, wrote on 24 May 2018 in techcrunch.com about his discovery in Johannesburg, South Africa. This is what he says: “On a recent work trip, I found myself in a swanky-but-still-hip office of a private tech firm. I was drinking a freshly frothed cappuccino, eyeing a mini-fridge stocked with local beer and standing amidst a group of hoodie-clad software developers typing away diligently at their laptops against a backdrop of Star Wars and xkcd comic wallpaper. I wasn’t in Silicon Valley: I was in Johannesburg, South Africa, meeting with a firm that is designing machine learning (ML) tools for a local project backed by the U.S. Agency for International Development.”
Anthony continues, “Around the world, tech startups are partnering with NGOs to bring machine learning (ML) and artificial intelligence to bear on problems that the international aid sector has wrestled with for decades. ML is uncovering new ways to increase crop yields for rural farmers. Computer vision lets us leverage aerial imagery to improve crisis relief efforts.
Natural language processing helps us gauge community sentiment. I’m excited about what might come from all of this.”
So there are AI projects in South Africa as well, and many more countries in Africa and the developing world will probably follow suit. But there are risks and responsibilities here. A machine learns from the status quo – the data and the developer’s mindset. Train your AI tool on great data and you’ll win, or on lousy data and your tool will not be worth much. Also, any biases that AI tool developers have must be properly controlled so they don’t spill into the tool they develop. This obviously represents a checkpoint for AI tool development.
I wrote in the 29 February 2018 article in this column about how several popular software tools written for the face recognition type of AI discriminated against people with dark skin and also against women. The most popular FRT apps, such as those developed by Microsoft, IBM, and Megvii of China are grossly inaccurate for identifying the faces of people with dark skin, thereby complicating the racial discourse. (The app by Microsoft is reportedly more accurate than those from the other two companies.) This is quite sad because these apps have moved from development to deployment, with extensive usage on the field.
That is, your face will be correctly identified if you are white or yellow than if you have dark skin. And the discrimination is not just about skin pigmentation. It also has a gender component, meaning that you are significantly more likely to be misidentified if you are a woman – of any complexion! These errors probably have their roots from insufficient attention being paid to the victimized groups.
A few weeks ago in a barber shop, I engaged a student from Northern Sudan in a conversation, and asked him why Sudan was split into north and south. Without any hesitation, and seemingly with an air of pride, he said that it was based exclusively on people’s skin color – the people in the north have lighter complexion than those in the south. Not minding the irony that this guy from Northern Sudan is as dark as dark people can be, he had an air of superiority. He would not develop an AI tool that does justice to his brothers and sisters in the south.
This is another checkpoint for AI tool development: tribalism and all that stuff. (By the way the word Sudan means “black.”) We have to leave our biases behind at home to do justice to our facial recognition tools. Unfortunately, there are many of these status quo issues along tribal lines in Africa.
Anthony suggests five basic things to keep in mind when applying AI and ML, which I paraphrase as follows: a) ask who’s not at the table – meaning geographical and cultural sensitivities, b) let others check your work, c) doubt your data – that is, have a large population in your training data to avoid biases, d) pay attention to how things change in different use cases or regions, and e) automate with care by keeping humans in the loop, as the mental models of humans are more nuanced and flexible than your computer algorithms.
As you can see, developing a cool AI tool comes with some non-computer responsibilities packaged with the tool.
