The EU-funded SPATIAL project, led by TU Delft, held its first Consortium meeting in person on the 21st and 22nd of September, 2022, hosted by the partner Fraunhofer Fokus at the Fraunhofer HHI premises in Berlin.
The first day began with a series of short sessions where leaders from each Work Package summarised their current status and progress. Two workshop sessions followed this – the first one, focused on metrics for accountability and resilience in AI, and the second on the motivations behind the consortium’s work on Explainable AI. This was also the first time the consortium got properly introduced to their new partner, MinnaLearn.
Work Package 5 kicked off, marking the beginning of the use case deployment phase of the project. Each of the four SPATIAL pilots was introduced, and the discussion circled how these use cases should be tied to and integrated into the SPATIAL platform. This is a crucial question as it determines how the project’s different components will flow together in the coming year.
The afternoon’s workshop sessions were preceded by a summary of the previous consortium meeting in May. The main takeaway was that while the May meeting introduced many new ideas, few conclusions were made by the end of it. For this reason, it was established that this meeting would focus on a primary goal: setting the architecture blueprint of the SPATIAL platform.
The following workshop session focused on the topics of metrics in accountability, transparency, resilience and privacy. Two main areas of concern were metric evaluation of clarity and metric computation concerning confidentiality.
The last day’s workshop surrounded the topic of motivations for Explainable AI and the abstract concept of the SPATIAL platform. The University of Tartu presented an example of a trustworthy city-scale deployment of autonomous vehicles to visualise the challenges that XAI aims to help solve. The discussion here revolved around what alternatives there are for the SPATIAL platform – such as application generic versus application specific – and what challenges these options entail for the architecture, such as the problem of scalability.
In the end, the consortium members left with the impression that meeting each other in person was a positive and fruitful experience.
The second day of the consortium meeting was divided into two main parts. The first was a workshop led by Erasmus University of Rotterdam and MinnaLearn. This workshop started with a presentation introducing the concept of the Education Module that MinnaLearn is developing in conjunction with the project.
The main topic of this presentation was the inspiration behind the module’s vision and content creation plan and how consortium partners can contribute to it. After this, the workshop revolved around discussions in smaller breakout groups concerning the societal impacts of AI from the angle of cybersecurity and privacy in the realms of governance, infrastructure, business, equality and democracy.
The consortium’s main takeaways from these discussions were that AI can be a great tool to empower citizens and communities but that an increase in AI prevalence also carries the risk of an equal increase in privacy and transparency concerns.
For example, opaque AI designs can erode citizen trust in AI-based infrastructures, limiting engagement and creating societal gaps. In contrast, harnessing the power of explainability in AI can help bring clarity to inherent biases and increase fairness and security in society.
Following that, the fourth and final workshop of the consortium meeting aimed to conceptualise the architecture of the explainability component of the SPATIAL platform. Much of the discussion revolved around two different aspects of explainability, one being the functionality and performance of an AI system about the metrics established the previous day, the other being how explanations affect users and how to close the sociotechnical gap.
Each partner also had the opportunity to present and discuss their planned component contributions to the SPATIAL platform to find ways for these applications to streamline toward a shared architecture – a roof under which all components can stand. In the end, no conclusions were reached about what shape the top encompassing all these applications would take, although suggestions were made. There was, however, an explicit agreement among all involved parties that the goal going forward was defining what makes the SPATIAL platform more significant than the sum of its parts.