From pixels to practice: the human side of AI
Yesterday I was a guest at Mirto, a social enterprise with sites in Drongen and Eeklo that focuses on inclusive employment, using digitalisation as a lever. I gave a training session for an AI labelling project, where we label photos to use later for training a new computer vision model. If we want AI to go beyond recognising dogs, cats, cars, buses or zebra crossings, people need to help make that possible.
Why Mirto? Clear processes, attention to detail and care for people. They give talent that gets fewer opportunities the space to do meaningful work. For us, they are the ideal partner to join forces with and deliver a strong project together.
Why labelling? AI does not learn "in general", but from concrete, consistent examples. Labels make explicit what matters (e.g. this is a button, that is a display). Without those definitions, you cannot apply supervised learning.
How are we approaching this?
🔹Concrete goals: what do we want to recognise, in what kind of photos?
🔹Clear instructions: what to label, what not to label and how, with "do's & don'ts" and plenty of visual examples
🔹Quality control: close collaboration, follow-up and scheduled feedback moments
🔹Tooling: we chose CVAT, quick to learn, easy to use and convenient for collaboration and follow-up
Measuring means knowing (and adjusting)
During and after labelling, we test whether the agreements also work in practice. After a first small batch, you quickly see what is going well and what needs to be adjusted. Based on that, we refine the rules and expand the examples, so the AI model visibly improves with every round.
From labels to a workable product
What starts with careful labels from the people at Mirto will end as a practical tool on the work floor that increases efficiency and productivity. Proud of this collaboration, where human talent and technology reinforce each other. At a time when AI is sometimes framed negatively, this is a story of AI built with people, for people.
Thanks to Luc Bulckaert and colleagues for the pleasant collaboration!
