One of the main aims of the project I have working been working on this past year in the Design Lab is adapting an animal classification algorithm to make it more accessible for people with no prior experience in machine learning. Researchers studying wildlife ecology who collect millions of camera trap images of animals sometimes do not have enough volunteers available to tag their images. Using machine learning for this task could lower the costs and the human labour involved in tagging the data on wildlife they collect.
Machine learning packages are currently very poorly documented: even though they are open-sourced and so abide by the principle that these technologies should be available to the public, the lack of documentation poses a huge barrier in using them.
In our searches, we came across ImageAI, a machine learning package built to allow developers, researchers and students to build applications using Computer Vision and Deep Learning. Every single application of the package was clearly documented in multiple languages, and with instructions that were appropriate for both advanced and less advanced users. The package was developed by Olafenwa and John Moses. In an article written by Olafenwa, he summarises the challenges in trying to democratise AI, from connectivity problems, to algorithmic bias. It is obvious that in their open source product the two founders have put a lot of thought not only into creating a state-of-the-art product, but also into ensuring that their product truly does reach a wide audience, and that it will also serve communities without extensive resources.
Their work is an amazing example of the level of accessibility that machine learning packages can reach, as a first step in bridging the gap in the digital divide.