Many residents in the IJmond region are worried about air pollution from the steel plant Tata Steel. Therefore, residents of the IJmond region, the action group Frisse Wind and Greenpeace placed three cameras to monitor air pollution from Tata Steel. People can manually check those images for suspicious smoke clouds. With developments around AI, that process can be automated.
Currently, AI is often seen as an autonomous entity, an intelligent system that magically evolves itself. Large companies maintain that magic, making it difficult to get a grip on the developments. But the development of AI can also be done differently.
Waag Futurelab and computer scientists from the University of Amsterdam led by Yen-Chia Hsu invited residents of the IJmond region to jointly design an AI-tool that helps residents spot suspicious smoke clouds. Besides that an AI-tool designed by the community will be more aligned with citizens their values and wishes, people also learn some basics of how AI works. Because residents are involved in the design process, they get a grip on both the design and the use of the technology.
‘If people themselves are involved in design process, you build a more widely supported tool. It is not top-down and people see the results of their wishes. It's technology for and by society.’
- Jikke van den Ende, project manager at Waag Futurelab
Labelling the data
AI works with a ‘training dataset’: the ideal collection of examples, in this case examples of suspicious smoke. The system uses this training dataset to label new data as suspicious smoke clouds.
Waag and Yen-Chia Hsu and his team organised two meetings where residents could have a say in labelling the data. In other words, accurately determining what a suspicious smoke cloud is. The system shows a picture with a square marker around the suspicious smoke cloud. Residents can adjust this marking to make the system more accurate. Because the community labels data to create a training dataset, people have an influence on what choices the model makes.
“When we engage the community in the design process of AI, I think it makes people understand and makes the model more transparent.”
– Yen-Chia Hsu, computer scientist at University of Amsterdam
Jos, a resident of IJmond region, attended the workshop. ‘It was well explained how smoke is recognised. A computer does not have the ability to see and does not recognise images by itself. We discussed the parameters used for recognition.’ Residents indicated that a photo does not give a good picture, because you have to see how long a smoke cloud lasts. Hsu and his team have now modified the system: the system now shows a photo with a marker, but the marker can be checked with a video clip behind the photo.
Use of AI
Based on input from residents, the tool is being adapted. The adjustments now were mainly about labelling the data. In the future, the system could be developed further, for instance as an alert system. When detecting a suspicious cloud, the system could automatically send a message to residents or authorities. Further development of the AI could be designed in consultation with residents and various parties.
The AI-tool could also help citizen scientists with their research on air pollution: they could combine the signalled clouds with other data, such as particulate matter measurements and wind direction.
‘Not everyone can watch Tata's video footage all day to see what is happening. This tool helps to see what is going on, to come up with evidence. It helps shape public opinion and come to measures.’
- Jos, resident of the IJmond region
The AI-tool is available to use now! Help spot suspicious clouds with the AI-tool.
There is also a tutorial available on how to use the tool.
About Waag Futurelab
Waag Futurelab contributes to the research, design and development of a sustainable, just society for already 30 years. We strive for a public stack, an approach where all of the layers that contribute to technology – the foundational values, the design process, the technology itself, and the ways in which these layers position people – are based on public values. Our activities are focused on public research, which enables as many people as possible to design an open, fair and inclusive future.
About Yen-Chia Hsu
Yen-Chia Hsu is a computer scientist with an architectural design background. He examines how to embed technology in hyper-local contexts to facilitate civic engagement and community empowerment. Specifically, his research is focused on Community-Empowered Artificial Intelligence (AI), where he co-designs, implements, deploys, and evaluates interactive AI systems that empower communities, especially in addressing environmental and social issues. His research team includes assistant professor Katja Rogers and student assistant Tycho S. C. Stam.
About 'suspicious smoke clouds'
The colour of a cloud gives an indication of what gases or substances are in the cloud. The colours black, yellow, orange and brown give an indication that there may be toxic substances in the cloud. Therefore, the UvA trains the AI-tool to recognise black, yellow, orange or brown clouds. We did not measure the particles or gases in the clouds.
AI and sustainability
Waag Futurelab believes it is important to limit the climate impact of AI and carefully weigh up what AI could be used for and what it could not be used for. In this case, the use of AI serves an important public interest by automating the monitoring of Tata Steel's air pollution.