Home page > Netherlands > Amsterdam

The Value Connection: Identifying, Implementing, and Assessing Values for AI Alignment

The Value Connection: Identifying, Implementing, and Assessing Values for AI Alignment Royal Netherlands Academy of Arts and Sciences (KNAW) Early Career Partnership, KNAW Trippenhuis Amsterdam, 15 – 17 March October 2023. ***Organizers*** Michael Klenk (Delft University of Technology) ***Theme*** There is ample discussion of the risks, benefits, and impacts of Artificial Intelligence (AI). Although the exact effects of AI on society are neither clear nor certain, AI is doubtlessly having a profound effect on overall human development and progress, and it will continue to do so in the future. The development of AI raises urgent questions about which and whose values it should be aligned with. This has come to be known as the value alignment problem (Russell 2019; Gabriel 2020, Christian 2020). Important stakeholders such as the IEEE or the European Union are increasingly alert to the issue and call for technology in general, including AI-based technology, to be aligned with or designed for ethical values (EU 2020; IEEE 2019). Thus, AI technology requires us to identify, implement, and assess appropriate values for the technology to optimise for. However, it is far from clear how appropriate targets for alignment would look like and how they could be identified in light of persistent conceptual and empirical challenges about the nature and accessibility of valuesThe aim of the conference is to present and discuss research on the interrelations of values and technology. Specifically, we aim to explore how novel technological developments lead to changes in moral values and, conversely, how changing moral values affect the developments of new technologies. ***Aims*** The partnership will contribute to the solution of value alignment in AI by integrating insights into (moral) value from philosophy and the empirical social sciences with current computer science approaches to value alignment. The leading questions are: *1. Identify: What are sound assumptions about (moral) values relevant for reinforcement learning? *2. Implement: Should value embedding in AI remain uncertain, and what role should democratic participation play? *3. Assess: How can embedded values be assessed? The partnership will look across disciplinary boundaries to inventorise insights pertinent to value alignment that currently remain hidden behind disciplinary boundaries. It will then synthesise the gathered expertise toward novel solutions for identifying, implementing, and assessing value alignment. Much of the current discourse has been focused on explorations in (inverse) reinforcement learning and related machine learning techniques needed to teach AIs values. Thus, the conversation has so far primarily been one between technical AI researchers. However, the value alignment problem extends into essentially all domains of human life and inquiry by its very nature. So, what has been largely ignored are deep and important social, political, and philosophical questions that have crucial implications in technical AI safety research itself and the implementation of value-aligned AIs. What is more, Important ideas for the solution of the alignment problem remain hidden behind disciplinary boundaries because the debate is siloed and suffers from a lack of integration across disciplines. An much more interdisciplinary approach is needed to identify values, implementing them in AI, and assessing the result. The ‘Value Connection’ will thus be the nucleus of a new network of 30 capable researchers working on value alignment from its essential perspectives: computer science, philosophy, political theory, and social science. The Value Connection will enter into and develop interdisciplinary collaborations to arrive at new fundamental research insights in the field of AI value alignment. ***Scientific Committee*** Michael Klenk (TU Delft) Marija Slavkovik (Bergen) Catholijn Jonker (TU Delft) Ibo van de Poel (TU Delft) ***Format*** Interactive workshop with the following themes: Introducing the alignment problem (day 1, afternoon), risks and opportunities in interdisciplinary collaborations (day 1, evening session), brief presentation pitches (ca 10min) from different disciplines on the topics of identifying, measuring, implementing, and assessing values in alignment with discussion (ca 20min for each pitch) (day 2), collaboration time and plans for follow-up activities with other members of the Value Connection (day 3). There will be ample time for discussion and collaboration throughout the workshop. Seven spots at the Value Connection are open for application. We invite applications for participation from early to mid-career researchers from all disciplines who want to join the ‘Value Connection.’ Applications should include a CV and a statement of motivation (max 500 words). The statement should outline the applicant’s planned contribution to the Value Connection, e.g. a proposal for identifying or measuring values. We especially seek contributions from political theory, economics, and sociology. Prior (research) experience specifically in value alignment is not a requirement, but a willingness to address this important topic with the tools from one’s discipline is a requirement.   A limited number of travel bursaries (ca 400€) will be made available to eligible Non-NL participants. The Value Connection will convene at the beautiful KNAW location ‘Trippenhuis’ in central Amsterdam: https://www.knaw.nl/en/het-trippenhuis. Catering and a conference dinner are generously provided by the KNAW Early Career Partnership. The event is also supported by the Delft Design for Values Institute. ***Timeline*** February 13 2023: Deadline for applications February 20: Decision about acceptance February 24: Confirmation of participation for chosen applicants March 15 midday – 17 evening: Conference ***Application details*** Please submit your application by email to Michael Klenk at m.b.o.t.klenk@tudelft with “KNAW Value Connection: Application” in the header of the email. The deadline for applications is February 13th.

WHERE WHERE

Het Trippenhuis, Kloveniersburgwal 29, Amsterdam, Netherlands, 1011 JV

WHEN WHEN

Anyone can use the "Add event" option
Source: PhilEvents

Leave a comment