Supervisors

Suitable for

Abstract

Abstract

Ticking without reading “I have read and agree to the Terms of Service” when registering online accounts is known as “the biggest lie on the Internet”. We are all guilty of ticking without reading it, but we recognize the necessity of the existence of these terms.

Data Terms of Use (DToU) is a similar but more approachable concept. It specifies what the data users need to follow before, during, and after using the data. In a decentralised web context - where everyone becomes a data provider, everyone will need to write their own
DToU. In addition, not only do they need to govern the direct use of data, the use of derived data is also subject to such DToU.

Handling such DToU is often labour-consuming and error-prone. Formalisation and automated reasoning have the potential to improve the handling and understanding of such DToU. We have previously developed a formal DToU language and the reasoning engine for
decentralised contexts, with also integration into Solid (https://solidproject.org), with potential to be used in wider scenarios. That is based on symbolic AI techniques, featuring interoperability, explainability and accountability. However, usability of symbolic AI systems
can be a major barrier for wider user communities such as normal citizens. In this project, the student is expected to explore mechanism facilitating the expression and utilisation of DToU, such as investigating how modern machine learning-based AI methods can facilitate
DToU expression and utilisation for wider user communities. Topics of interest include but are not limited to:
- How can NLP/LLM technology facilitate the expression and comprehension of DToU policies?
- How to utilise agents to perform preference learning through users’ activities?
- Mechanisms for better accountability or trustworthiness, such as provenance, instantiation to ODRL, etc

Interested students are welcome to contact Rui Zhao and Jun Zhao to discuss or propose their own ideas related to above (rui.zhao@cs.ox.ac.uk, jun.zhao@cs.ox.ac.uk).
Prerequisites:
- Appropriate foundation of first-order logic
- Although the student is not required to extend the DToU language, appropriate foundation of FOL is needed to understand the language, and may be useful for the machine learning part for explainability
- Practical and theoretical knowledge of machine learning, were the student interested in the modern AI aspects
- Technical experience with at least one of the following will be a big plus:
- Solid
- RDF / Linked Data / Semantic Web / Knowledge Graph
- NLP / LLM
- Confident with an appropriate language to work with AI and/or Solid, especially Python or JS/TS

References:
1. https://arxiv.org/abs/2403.07587