Navigating AI’s ethical aspects — Insights from the latest EMERALDS webinar

As AI systems become increasingly integral to our lives, the importance of ethical considerations and legal compliance in their design cannot be overstated. That’s why EMERALDS proposed as its latest webinar topic one that explores AI’s ethical and legal aspects, particularly based on how ethical considerations are addressed and managed within the project. From optimising transportation routes, to enhancing public safety, the possibilities of integrating AI technologies and data-driven decision-making systems within urban mobility processes seem endless. Yet, there are risks to be taken into account: what happens if these data are misused? What if privacy is compromised, leading to what we may experience as a surveillance dystopia?

During the webinar “Navigating AI’s ethical aspects”, we focused on best practices and regulatory frameworks to assist developers in designing ethical systems for urban mobility data analytics, and connect the dots with EU regulations such as the EU Data Act and the recent AI Act. And to discuss these complexities, we were joined by a pool of ethical experts and representatives of the EMERALDS project: Mr. Konstantinos Ntzoufas, legal ethics advisor on AI, specialised in civil law, who discussed AI trustworthiness and some key ethical issues arising from the recent EU AI Act; Jeroen Steenbakkers, owner and CEO of Argaleo , who presented the insights of The Hague Use Case; an insightful presentation on 荷兰代尔夫特理工大学 Human Research Ethics Practices from the academic point of view given by Sascha Hoogendorn-Lanser; from the software development perspective, a presentation on from Anita Graser, Data Science Researcher from AIT Austrian Institute of Technology , discussing the ethics of mobility AI development; and lastly, we hosted a presentation from Mr. Nassos Fanos, also a LLM in EU law, who presented the Data Governance Act and explored GDPR considerations in the scope of developing EMERALDS algorithms.

Mr Konstantinos Ntzoufas started his presentation with an overview of the new EU AI Act that was recently endorsed, including the shifted definition of AI systems and risk-based regulatory frameworks. He also explored the classification of AI systems into three risk categories: prohibited, high-risk, and minimal risk, with different requirements for each.

“The overarching characteristics of AI systems must be the trustworthiness and the human-centric character, which is a finishing touch to this legislation act. We have to observe that there has been a serious shift in the definition of AI systems. AI system means machine-based system, whereas at the beginning, there was a reference to software systems, which are designed to operate with varying levels of autonomy and they may exhibit adaptiveness after.”

Up to Jeroen Steenbakkers next, who gave a thorough introduction to his own company, Argaleo, and their digital twin project to create data visualisations and predictions for crowd levels and traffic in the Scheveningen beach area. The challenges the area presented were more than just a few, with it being one of the busiest beaches in The Netherlands— namely, high visitor numbers and capacity issues for crowds and traffic. Jeroen described the Digital Twin Dashboard created at Argaleo using open and local data sources, with machine learning algorithms being developed for crowd forecast, all privacy-proof according to GDPR.

“What they [the municipality, editor's note] asked us to do is to create a sort of a digital twin data dashboard, in which, on the one hand, we use a lot of open data available, but we also connect to local data sets. For instance, there are numbers of cameras on the boulevard, which we use to get insights on the crowdedness and the amount of people that walk on the boulevard.”

Sascha Hoogendoorn-Lanser gave us insights on dealing with ethical challenges in a university setting. While presenting the Human Research Ethics Committee (HAREC) existing at TU Delft, she indicated the composition and mandatory screening by the Committee for any research involving human subjects to ensure privacy, security and ethics. She noted that the approval process for this kind of research requires documentation and explanation for adhering to ethical standards when questions arise.

“The HAREC works together with a privacy team at TU Delft, but also with data management services. And they have a number of tools that can help you in planning, minimising, and managing possible risks that arise from research activities that we do at the university.”

Next up, we had AIT Data Science Researcher Anita Graser, who gave some considerations for developing ethical and public AI tools. Anita highlighted the large energy usage of AI and strategies for green AI development, something that balances costs and performance, and emphasised the importance of data protection measures to avoid privacy issues while still utilising mobility data.

“We just know that even training relatively small models is easily spending more energy than, for example, flying from between continents on an intercontinental flight. And this is constantly growing because one strategy that is being pursued in making better AI is to just give it more training data, which means we need more computational resources. This is what can be referred to as the red AI approach. So, you just throw more data, more computational power at the problem, and you hope that your AI model gets better by doing that, disregarding all the costs, energy-wise, society-wise.”

Anita proposes the alternative to this approach, which is known as green AI, an AI research approach where we take computational time and costs into account, and we try to reduce those costs while still achieving good results. She also emphasised the importance of balancing model accuracy and robustness with the need for transparency.

Finally, our last speaker for the day, Mr. Nassos Fanos, examined how the EU Data Governance Regulation and GDPR apply to the development of AI algorithms in the EMERALDS project, emphasising the importance of data sovereignty and accessibility. He also highlighted the need for a holistic approach that takes into account all the relevant regulatory requirements.

“In the case of not explicitly anonymised data sets, regarding anonymisation, access controls, personal data, recommendation, legal guidance, and training, all these two regulative instruments will have to be reviewed as per their intersections as well as per their exclusive scope so that all of them are invoked at the end of our job as ethics advisors, only for ethics issues, not for data protection issues because these have been covered by the DPOs who are involved in the data collection.”

With all that was discussed during the webinar, we experienced and explored some best practices, got insights on regulatory frameworks, and real-world strategies that can be utilised to ensure that innovation in urban mobility data analytics is not just effective but also ethical and responsible. A good reminder to all of us, our Project Coordinator Foivos Galatoulas closed the webinar:

“The?key?takeout?today?is?that?we?need?to?remind?ourselves?that?with?technological?advancements?and?innovation?comes?even?greater?responsibility?towards?society.”