[agents] PhD position on Decoding Value Structures through Computational Exploration at the University of Amsterdam

[Giovanni Sileno]

PhD position on Decoding Value Structures through Computational Exploration at the University of Amsterdam

Are you passionate about investigating fundamental research questions in computer science, information and/or systems theory, and natural/artificial forms of intelligence? Are you motivated by the idea of understanding both social and artificial systems and enhancing their interconnected influence? This project aims to investigate methods and computational models to improve our understanding of values embedded in social systems, to guide socio-technical interventions. The PhD candidate position is embedded within the Socially Intelligent Artificial Systems (SIAS) group of the Informatics Institute (IvI) of the University of Amsterdam (UvA).

Project description
--------------------------------
In AI research and development, system evaluation is typically guided by a single core question: Does this solution outperform its alternatives? Yet, critical AI discourse makes clear that there is a prior question that need to be answered: What does this solution work for? As computational systems become deeply intertwined in global socio-technical ecosystems, having a better understanding of the network of distributed expectations, perceptions, and transformations of value(s) revealed through social interactions is an essential basis for addressing adequate interventions. From a computational perspective, this "reverse engineering" pursuit can be supported by several lines of research. Two alternative directions that could be investigated through this PhD position are:

Value structures discovery on simulated social systems. To what extent can contemporary computational methods help us to identify the inner functioning of distributed socio-technical systems? Constructs issued from agent-based modelling, complex adaptive systems, multi-agent and normative systems literature allow us to create increasingly sophisticated informational, motivational, and governance mechanisms. These models can mirror established socio-economic models, drawn for instance from historical reconstructions or model-based theoretical frameworks, and can be executed to generate synthetic data. Data-driven computational techniques can then be applied for the detection of various value translations inherent to or emerging from the underlying socio-economic model. These techniques may rely on approaches currently developed in AI (eg. causal discovery, reinforcement learning, explainable AI methods), in computer science (eg. model construction through compression), or in computational science (eg. neural differential equations, Markov's blankets). Limitations observed in this benchmark will be instrumental in proposing methods tailored to the reverse engineering task.

Processing user narratives for value ascriptions. How can human experiences be leveraged as primary source of knowledge to investigate the behaviour of systems? Narratives play a primary role in constructing and securing the mechanisms of intentionality, both at individual and collective level; various contributions in cognitive science argue that they provide the highest level of integration of an individual’s knowledge system. Contemporary natural language processing (NLP) techniques may provide instruments to automatically extract insights from narratives of people detailing individual interactions with eg. a device, a service, an organization. By disentangling these micro-level scenarios in an adequate representational model, we may construct a broader qualitative behavioural model of the system. This artefact can serve as a platform for investigations, both with respect to the real system in itself (eg. for stress testing), and to involve people further (eg. to test expectations on scenarios not yet accounted, or to frame requests for other past experiences). This approach becomes particularly relevant when data or models are not available to direct inspections, and provides a more systematic role to socially-distributed "anecdotal evidence" for computational research and development, in analogy to the use of case studies in medicine.

Embedding
--------------------------------
This PhD project will be conducted at the Socially Intelligent Artificial Systems group, under the supervision of Giovanni Sileno, within the theme People, Society and Technology of the Informatics Institute at the Science Faculty of the University of Amsterdam.

The University of Amsterdam (UvA) is the Netherlands' largest university (42,000 students, 6,000 staff members and 3,000 PhD candidates), offering the widest range of academic programmes. The Socially Intelligent Artificial Systems research group (SIAS) in the Informatics Institute (IvI) of the UvA studies how to advance people's everyday life and society in general through AI research, education and impact. The group engages in incremental trust building and value learning with stakeholders across various scientific disciplines and application domains, and on topics that are relevant both socially and academically.

Practical information
--------------------------------
More information, requirements and application instructions can be found here: https://vacatures.uva.nl/UvA/job/PhD-Candidate-on-Decoding-Value-Structures-through-Computational-Exploration/783019302/

Do you have any further questions on the project?
Please contact: Giovanni Sileno <g.sileno at uva.nl<mailto:g.sileno at uva.nl>>
If you feel the profile fits you, and you are interested in the job, we look forward to receiving your application. We accept applications until and including 5 January 2023.

---
Giovanni Sileno
Assistant Professor | Socially Intelligent Artificial Systems (SIAS)
Informatics Institute | Faculty of Science | University of Amsterdam
https://gsileno.net<https://gsileno.net/> || @gsileno
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cs.umbc.edu/pipermail/agents/attachments/20231127/1b2a6d41/attachment-0001.html>