IN THIS LESSON

A deep dive into computer vision

Now that we’ve got some AI basics, we can start to get concrete on how people are applying AI to address real world problems. One of the most promising areas of application for AI in international development is computer vision. At its most basic, computer vision involves training computers to mimic a human understanding of visual content (Altamira, 2023). To get a bit clearer on what this looks like, consider the computer vision tasks laid out below:

  • In image classification, the system is trained to identify the category that an image belongs to; say identifying whether an image shows a dog or a cat (Boesch, n.d.).

  • Object detection involves picking out different objects within an image and identifying the boxes within which each object is bound. Suppose you have an image of a busy street, object detection would allow you to identify the unique objects in the street (shops, people, bikes, etc.) and the specific part of the image that they are bound in (Boesch, n.d.).

  • This involves dissecting an image into various regions, each belong to a different category. Suppose that you have an image of a landscape; using segmentation you can identify the regions of the image that can be considered land, water, and sky (Klingler, n.d.).

It is important to note that these represent discrete tasks which a computer vision system can be trained to perform. In any given use case, there may be multiple tasks working together to produce a desired result. Also, the underlying learning technique for each task can differ. For instance, you might train an algorithm to identify whether an image includes a cat or a dog with supervised learning by feeding it correctly labelled images of cats and dogs. Alternatively, you could feed it a set of unlabelled images, and using unsupervised learning techniques, have it identify different groups of images based on similar patterns of pixels, which it can then use to identify which animal is in each image.  

In international development, we have to do a lot more than tell the difference between cats and dogs. Throughout our work on the Frontier Tech programme, we’ve often found that technology only takes us so far. Successfully leveraging technology to influence real systems change and provide a sustainable difference to people’s lives involves a lot more than having a working use case. We need to be attentive to cost, human behaviour, the intricacies of existing systems, and the complexities of the challenges we’re addressing to create solutions that work for people.  

In the following two case studies, we’re going to explore how practitioners have navigated these issues while developing AI technologies to have a real impact on the world. We’ll dig into the solutions they’ve developed, the practical process they went through, the barriers they faced, and the ethical issues and risks which they addressed to develop impactful use-cases. Keep going to discover more about what we’ve been learning.