Two Postdocs in CS for Foundational Research into ‘Legal Tech’

This is your chance to dig into the fundamental assumptions underlying computer science, teasing out the implications they may have for real life applications, notably those of ‘legal tech’. The combination of research into the theory of computer science and the opportunity to make a difference in the legal domain provides a unique opening for those willing to address the societal impact of both machine learning and self-executing code, based on frontline research in the theory of computer science.

Both postdocs will be part of the cross-disciplinary research team of the ERC Advanced Grant project on ‘Counting as a Human Being in the Era of Computation Law (COHUBICOL)’. They will inquire into the architecture and foundational issues of: 

Machine learning applications, e.g. prediction of judgements.
Self-executing code, e.g. ensuring compliance by design, in the context of legal practice, while the focus will be on assumptions of computer science and their implications.
One of the postdocs will focus on machine learning, the other on self-driving code.

Relevant research questions for the computer science team could, for instance, be:

On the side of machine learning:

If machine learning enables compression of a dataset into a mathematical target function, how could lay persons and lawyers learn to test and contest the relevance of the compression?
What is the relationship between ML and statistics and how do their assumptions differ? How does this relate to the difference between exploratory and confirmatory research? How does this relate to the credibility and reproducibility crisis and the lack of theoretical rigour in applied ML research?
On the side of self-executing code (smart contracts or smart regulation):

Which relevant assumptions are key to the mathematical foundations of CS?
What are the implications of the inherent limitations of CS, considering e.g. Godel’s incompleteness theorem, the Church-Turing thesis on undecidability, and Wolpert’s NFL theorems - coupled with the issues of e.g. undefined behaviour and floating point numbers ?
What kind of transparency is required to ensure that self-executing code aligns with the claims made about their behaviour (mathematical, computational, domain expertise)?

Further details:
Two Postdocs in CS for Foundational Research into ‘Legal Tech’