Analysing eighteen thousand chicken videos
The indispensable work of participation employees in the AI & Animal Welfare Lab

Artificial intelligence can take over a lot of work from us. But what is still occasionally forgotten is that training all these AI models can require very much manual labour. One of these AI trainers is participation employee Armanda den Exter, whose work in the AI Lab for Animal Welfare included labelling thousands of chicken videos. “Without Armanda and her colleague, this research would be barely feasible.”
Skillfully, Armanda den Exter picks up a chicken from the henhouses in the Farm Animals Building of Veterinary Medicine. It is clear she has held chickens before. “Hundreds,” she calculates, while looking at the photographer with the chicken in her arms. “Because I've helped undoing special little backpacks with QR codes.”
The little backpacks with QR codes are a part of a research project within the AI & Animal Welfare Lab, one of the fifteen AI Labs of the university. In these Labs, various kinds of researchers collaborate with experts from the professional field on AI solutions to societal challenges. Some examples of the work done in the AI Lab is on ways to detect pain in dogs, or movement problems in horses. The little backpacks with QR codes are used in a project to monitor the well-being of laying hens.
AI measures behaviour of individual chickens
The use of cameras and sensors has already been a well-tested method to measure animal behaviour for some time, but this was mostly about group behaviour and simple motion detection until recently. Thanks to AI, the behaviour of individual chickens in a big group can now be analysed. “The AI models are trained to differentiate between chickens and to recognise various behavioural patterns, such as scratching or eating,” Professor Bas Rodenburg, one of the head researchers of the AI & Animal Welfare Lab, tells. This way, there can be a timely intervention if a chicken is showing problematic behaviour, such as pecking or ‘piling’. In the latter, the chickens get together in a big pile, resulting in the suffocation of the animals at the bottom. A big problem in practice, Rodenburg notes.
The AI models are trained to differentiate between chickens and to recognise various behavioural patterns, such as scratching or eating.
The final goal of the research is to select chickens who feel comfortable in big groups and display positive behaviour. Rodenburg says: “These chickens can be used to provide the next generation of laying hens; this is why the AI Lab collaborates with breeding company Hendrix Genetics for this project.”

Back to the little backpacks with QR codes. The QR code enables individual recognition with camera footage, even in crowded groups. But a big part of the labelling in order to train the system is done manually, by watching the footage, drawing a box around each chicken in the image and describing their behaviour. And that is an enormous job, Armanda den Exter knows from experience. “In total, I've watched 18,000 frames of chicken videos. At one point, I couldn't stand to see any more chickens.”
It is often the researchers themselves who have to watch and label the enormous amounts of data. A time-consuming job, which whittles away precious hours from the time available for the research on top of that. This is why at Veterinary Medicine, they collaborate with participation employees; people who have difficulty finding regular jobs for a variety of reasons, but are looking for meaningful work and want to be involved in society.
In total, I've watched 18,000 frames of chicken videos. At one point, I couldn't stand to see any more chickens.
After degree programmes to become an animal caretaker and a laboratory employee, Den Exter was looking for an opportunity at the labour market too. Her mental health played an important role in that search; she was diagnosed with a Borderline personality disorder. Den Exter says: “I see everything very much black and white, which is why I have difficulty handling criticism and quickly draw negative conclusions. Besides this, I feel emotions much more intensely than other people; I could – especially in the past – react very extremely in certain situations.”

Four years of intensive therapy helped her to deal with that and to better understand what she needs, also in a work environment. Den Exter says: “Thanks to the therapy, I'm more aware that reality can be different from how I see it. There is now also, so to speak, space to think on how I really feel first, before I react. That can sometimes take a couple of hours.”
Den Exter came to the conclusion that a traditional nine-to-five job does not suit her. But she did really want to work. At Veterinary Medicine, she can schedule her own time and work from home if she needs it. She has since become a valuable employee, because researchers at Veterinary Medicine also know where to find her for other tasks, just like the other participation employee at the department, who do important work in the fields of data collection and literature research.
Concentration and focus
“To us, it's great to have participation employees who want to do this work,” says Den Exter's supervisor Rodenburg. “Participation employees are often people who are very good at working with concentration and focus. Labelling data is essential for the research, so you want that to be done very neatly and precisely, and in line with a set routine. And that turned out to match very well with Armanda's qualities. Without Armanda and her colleague, this research would be barely feasible.”
According to Rodenburg, many research-supporting tasks are suitable for participation employees. “These are tasks which can be planned very well, which wait for you. If you don't feel good one day, or there's something else going on, they'll be there tomorrow too. So that links up very well with what they can handle and like. It's priceless to us that these jobs get done and the mountain of data is processed, and it's great for them to have the connection with the professional field.”
Once everything is up and running, and they work more and more independently and develop themselves further, that's then very pleasant to see.
As a supervisor, it does require some additional effort to work with participation employees, Rodenburg acknowledges. “You have to be available for questions and communicate a lot: can you still do it? Are you doing okay? It takes time to make clear agreements and to provide a defined task. But once everything is up and running, and they work more and more independently and develop themselves further, that's then very pleasant to see. I can recommend it to everyone.”
Den Exter calls her work for the university her ‘dream job’. “I'm an enormous animal lover – I have an entire zoo at home – so it's fun then to also contribute to animal well-being in your work. Besides that, everyone here exudes that I'm allowed to make mistakes, that it's okay if it's not perfect. The appreciation can be seen from all sides, that's simply really nice.”
Utrecht AI Labs
In the Utrecht AI Labs, Utrecht University unites science and practice by intensively associating with the corporate world, the public sector and other partners. Within the Labs, researchers work on responsible applications of AI and the AI talent of the future is being educated at the same time.
In the AI & Animal Welfare Lab, animal scientists, biologists and IT researchers collaborate with experts from, for instance, breeding companies, sports associations and NGOs to look for possible applications of AI-based solutions in practice. Examples of current projects are pain detection in dogs and horses, smart sensors in stables and computer-controlled design of horseshoes.
The goal of the AI & Animal Welfare Lab is to work from the animal's perspective and to look into how it experiences its environment and how it feels. AI has added value for the animal itself in this and does not serve to utilise animals more efficiently for human purposes.
Would you like to know more about the AI & Animal Welfare Lab? Take a look at our other projects.