National Conference: Autonomy in a Digital Era

Frans Oliehoek is Full Professor Interactive Learning and Decision Making at the Department of Intelligent Systems at TU Delft.

'Adaptive agents and interactive learning'

Artificial intelligence has a long history of formalizing intelligent systems as 'agents', i.e., entities that interact with an environments by actions (actuators) and make observations (via sensors). These agents could automate tasks, save time, optimize logistics, and even accelerate scientific discovery. However, building such agents is typically complex, and most use cases would require such agents to be adaptive, such that they can learn from their experience, adding to this complexity burden.

I will give an overview of some different approaches that people have taken towards creating these intelligent agents. Specifically, I will zoom in on a particularly promising approach called 'reinforcement learning' (RL) that potentially enables agents themselves to learn how to act, and that led to successes in the game of Go, robotic manipulation, and training of language models. Finally, I will also touch upon the recent 'LLM agents' and the theoretical implications of training these via RL.

Pinar Yolum is Full Professor Trustworthy AI at the Department of Information and Computing Sciences at Utrecht University.

'Responsible Autonomy in Human-AI Collaboration'

In human–AI collaboration, both people and AI agents need to exercise autonomy at the right moments and within appropriate boundaries. This relies on each agent understanding what actions they can take, when they need help or resources from others, and how to interact effectively to obtain them. 

To support responsible autonomy in these settings, we propose using consent as a unifying abstraction. Although consent has been a part of the software ecosystem, there has been little work to understand its dynamics formally, and to devise mechanisms to use consent in facilitating autonomy. 

This talk will discuss a representation of consent as well as a life-cycle to monitor its evolution. We will discuss how this representation can be useful in modeling and detecting autonomy violations in human-AI collaboration.

Rense Corten is Associate Professor at the Department of Sociology at Utrecht University.

'Trust, reputation and autonomy in the platform economy: computational and experimental approaches'

Trust problems are inherent in many markets, but especially prominent the emerging Platform Economy. How do buyers of second-hand products on E-bay or drugs on the Dark web, guests on Airbnb, or employers on Taskrabbit know that the seller, host or worker will not take advantage of them? 

A well-known 'institutional' solution is the implementation of rating- or reputation systems, in which users of platforms share their satisfaction using star ratings or written reviews. While this solutions generally seems to work well, there are many open questions. For example, to what extent are such systems capable of solving in-group biases? And to what extent are users able to 'carry' their reputation from one platform to another, avoiding lock-in? 
In this talk, I demonstrate how such questions can be addressed using a variety of (computational) social science methods, including the use of digital traces and behavioral experiments.

Cecilia Rikap is Associate Professor in Economics and the Head of Research at the Institute for Innovation and Public Purpose- University College London.

 'Digital crossroads: corporate power, sovereignty and the ecological breakdown'

How did we arrive at a world in which a handful of megacorporations from the United States -and to a lesser extent, from China- control the production and use of the technologies that define contemporary capitalism? Thousands of companies, universities, public research organizations, and even governments, located all around the world are part of the new peripheries, creating value and knowledge for others. Those who accumulate the benefits are not entire countries but Google, Meta, Amazon, Microsoft. They are the new core of global capitalism. Resulting relations of dependency cannot be described through the binary opposites of imperial power and colony, or feudal lord and serfs, as suggested by frameworks such as digital colonialism and techno-feudalism. Dependency today instead means a web of complicities, tensions and hierarchies. 

Who are the peripheries’ local accomplices of digital dependency? How are nature’s extractivism and data and knowledge extractivisms connected? How is work being transformed in a world in which new technologies are a powerful mechanism of control and indoctrination? And why is it that, faced with the need to build an alternative, attempts to expand digital sovereignty equally fail? These key questions will structure my talk at the “Autonomy in the Digital Era” Conference.

Niels Kerssens is Assistant Professor in the Department of Media and Culture Studies at Utrecht University. 

'Reclaiming autonomy in the digital classroom: securing public value in European education in the era of Big Tech and AI'

How can European primary and secondary schools (re)claim control over public education amid the growing dominance of Big Tech and AI? While schools struggle to assess the potential and risks of AI-driven educational technologies (EdTech), global tech companies aggressively push these tools into the classroom. 
This rapid influx of (AI-)EdTech by Big Tech companies, poses unprecedented challenges to democratic control over public education, fueling societal demands for digital autonomy in Europe. 
In response to these developments, this talk outlines three pathways toward digital autonomy in European primary and secondary education: co-design practices; impact assessment frameworks, and public digital infrastructures. 

Rianne Riemens is a postdoctoral researcher studying Big Tech and processes of platformization in relation to the climate crisis at the Department of Media and Culture Studies at Utrecht University. 

'Big AI’s climate futures: an environmental perspective on the autonomy-dependency debate'

The development of AI services has quickly become a core activity of Big Tech companies. This has led to a consolidation of their economic power, but also to a large increase in their carbon emissions and electricity use. Companies such as Microsoft and Google have responded to critiques by readjusting their sustainability goals, downplaying their impact, and promoting “green AI” solutions. 
Nevertheless, their plans for decarbonization are jeopardized by massive infrastructural and energy investments needed to sustain AI growth. 

In this talk, I highlight the political and environmental implications of these new developments. What new dependencies emerge when tech companies gain more ownership over energy resources and infrastructures, finite resources, or the carbon market? And how might this affect the plans for a digitally autonomous and sustainable Europe?

Jeroen de Ridder is Principal Investigator / Full Professor in the Center for Molecular Medicine of the UMC Utrecht. 

'AI and ML in molecular diagnostics'

The de Ridder lab specialises in developing advanced machine learning and artificial intelligence algorithms to achieve critical (pre-)clinical objectives using modern omics data. 
For instance, with Sturgeon we’ve cut down drastically on the time required for (pediatric) CNS cancer diagnostics with intraoperative native DNA Nanopore sequencing and methylation classification using deep learning. 
This is a clear example of our constant quest for actual, clinical, bedside impact. Interest areas in the lab are broad and range from GWAS, to liquid biopsy classification under data sparsity, to 3D genome conformation and epigenetics. In all these varied endeavors, translating from fundamental research to actual patient benefit is a guiding principle.

Joel Anderson is chairholder of Moral Psychology & Social Philosophy at Utrecht University.

'Data-ethical autonomy as a social skill'

 Governance of digitalising societies requires data-ethical autonomy – not just digital literacy. In this lecture, I argue that to be autonomous as participants in numerous data-related practices and institutions – and as citizens – we need three packages of skills: self-determination (recognized opportunities to shape algorithmic governance), self-governance (dialogical competence to critically assess data practices), and self-authorization (the sense of standing to contest algorithmic decisions).