"AI: The biggest challenges are the biases and lack of transparency of algorithms"
The biggest challenges are the biases and the lack of transparency of the algorithms embedded in existing AI solutions.
Most current AI models are closed. It’s not clear how they were trained. They are like a black box in which you provide input, then magic happens and you obtain a certain output. The more you train it, the better it becomes, but the output will always depend on your input. And those bringing in the input are often biased.
The problem with existing models is you cannot even know if they are biased, or how they are biased because they are black boxes. You cannot know what’s inside and training data and processes are not traceable.
HOT seeks to tackle biases by localising models, meaning not looking at the general model that works everywhere. And we counter the lack of transparency by using fully open-source AI models. In our case, it’s clear how our AI systems are trained and who is training them. The training data is available, so anyone could check how we get a certain output. And those bringing in the input are local people doing the mapping of their own space for their own purposes. The input is relevant in their context.
Map data features like buildings or roads are extracted automatically from satellite or drone imagery and validated by humans – local humans who train the AI models by applying them in their own regions. HOT adds this local knowledge to the extracted data. For instance, if I am working in the health sector in Zimbabwe and the imagery shows a building that appears to be a hospital, locals will not only confirm whether this is the case but will also note whether this is a facility where pregnant women can get certain services – the kind of data that is not easily extractable from imagery with current technologies.
It is the local community who defines what they want to map and why. We don’t give directions. We have regional hubs that provide support, but the actual ask comes from communities. This could be mapping fishponds in India, water points in Niger, or buildings in Brazil. It’s very often about mapping certain buildings so that people know where to go in an emergency, but it very much depends on whether the maps are going to be used for development work or in the event of a disaster.
HOT grew and became known as a result of its mapping of buildings and roads, which was very useful in responding to earthquakes such as those in Haiti and Nepal. But we then expanded our focus to enable not only post-disaster action but also anticipatory action, which is increasingly related to climate change-related challenges such as flooding or drought. At the end of the day, it’s about what each community wants to take action on – we basically enable them to do it. We’re more of a catalyser; we don’t want to do everything ourselves, and that’s why all our software is free and open source.