Public ADD Lecture: What are the values in value alignment?

Join ADD and the TANTlab on November 6th at 15:00 – 16:00 for a public lecture with David Moats from the University of Helsinki
Public ADD lecture with David Moats

Event host: The Techno-Anthropology Lab

When and where: November 6th, 2023 from 15:00-16:00 at Aalborg University Copenhagen, A.C. Meyers Vænge 15, Copenhagen, Danmark 2450

Recent work in machine learning under the heading of ‘value alignment’ seeks to align autonomous systems with ‘human values’. Some of this happens through the mathematical formalisation of values like ‘fairness’, while approaches like Inverse Reinforcement Learning (IRL) seek to extract a reward function (seen as a proxy for value) from human preferences or behaviours. But how do these computational approaches understand ‘human values’ or ‘alignment’?

Through an analysis of recent debates about ‘value alignment’, I will argue that many of the dominant approaches understand values in a narrow and individualistic way, often conflating them with goals or preferences. I will contrast these understandings with theories of collective values from the social sciences and empirical studies of valuation practices. This work suggests that values are not drivers of action in our heads but tools we use to justify and make sense of new scenarios. What is unique about collective values is the way they are invoked and negotiated in public, something which is not captured in laboratory experiments and simulated debates.

I will go on to describe the work we have been doing in a European project to study values in relation to algorithmic systems empirically. These include mining past cases for values and value tensions using situational maps 2) analysing value judgements about algorithmic systems on social media, in policy documents and in parliamentary discussions 3) developing participatory approaches to LLM evaluation.

Looking forward to seeing you there!

Join Zoom Meeting

Meeting ID: 617 8278 4212
Passcode: 945855