AI Ethics Workshop
Date: Friday, May 17 2024
Time: 12.30-15.30
Place: Campus Kolding, room 31.43
Sign up: Open event, no sign up
Program:
12.30-12.35 Opening remarks – Anne Gerdes
12.35-13.05 Tobias Hyrup: Sharing is Caring: Balancing Data Sharing and Privacy with Synthetic Data
13.05-13.35 Søren Harnow Klausen: AI and Affordances for Personal and Cultural Development
13.35 – 13.55 Coffee break
13.55-14.25 Chunfang Zhou: Integrating Creativity and Ethics into AI Education
14.25-14.55 Evgenios Vlachos: Data in AI: Trust is not Enough
14.55-15.25 Anne Gerdes: Highlighting the Benefits of Open Foundation Models in Healthcare
15.25- 15.30 Closing remarks - Anne Gerdes
Tobias Hyrup: Sharing is Caring: Balancing Data Sharing and Privacy with Synthetic Data
In the realm of data sharing, striking the right balance between openness and privacy is crucial. One promising approach to address this challenge is leveraging synthetic data as a substitute for real data, enabling sharing while safeguarding privacy. However, privacy is not an inherent property of synthetic data and therefore necessitates rigorous evaluation of its privacy-preserving capabilities and utility. This presentation explores the critical question of when synthetic data can be considered sufficiently private for safe sharing. Through an exploration of synthetic data generation and privacy metrics, it provides insights into assessing privacy adequacy. Ultimately, the goal of this presentation is to shed light on this complex topic and encourage a nuanced discussion that is applicable across various domains.
Department of Mathematics and Computer Science, University of Southern Denmark
Email: hyrup@imada.sdu.dk
Søren Harnow Klausen: AI and Affordances for Personal and Cultural Development
What does it mean for personal development if (or rather when) AI comes to determine the informational and cultural environment to which individuals are exposed? Self-cultivation or personal formation (Danish dannelse; German Bildung) has been assumed to depend centrally on a free exploration of cultural values and practices within and outside one’s community – and on experiences of not (yet) being able to fully understand something or cope with a situation, which can force one to enlarge one’s conceptual and emotional repertoire and so enlarge one’s horizon. Critical discussions of the growing influence of digital communication and AI tend to focus on the risk of being exposed to problematic content. But there may also be reason to worry that it could make information acquisition and enculturation too simple and smooth, relieving individuals of tasks that may be important to their personal development or contributing to a streamlining or mainstreaming which may reduce cultural diversity.
Department of Department of Design, Media and Educational Science
Email: harnow@sdu.dk
Chunfang Zhou: Integrating Creativity and Ethics into AI Education
AI is reshaping our society. This drives us to rethink how to foster qualified AI professionals for the future. There has been growing attention to develop AI study programs and AI-related disciplines education such as data science, computer science, robotics, machine learning, and neural network, etc in Danish universities and other cultures. Various learning skills and competencies have been drawn much attention in AI education. Creativity has been viewed as one of key learning skills in 21st century; AI ethics should be viewed as the framework of creativity development. How can we integrate creativity and ethics into AI education? This question will be focused in this presentation that will also lead a discussion with all participants in this workshop.
Department of Mathematics and Computer Science, University of Southern Denmark
Email: chzh@sdu.dk
Evgenios Vlachos1,2: Data in AI: Trust is not Enough
Data is of elevated significance in artificial intelligence (AI). The absence of data trust may introduce possible adverse effects on multiple fronts by introducing unfair bias, by upholding human autonomy, by influencing negatively societal well-being, or by having negative environmental repercussions. According to the Cambridge dictionary the definition of trust is, “to hope and expect that something is true”. Trust is not enough when it comes to data used in AI. There need to be safeguard mechanisms, formal processes and principles in place to assure that data will be fair, accurate and unbiased throughout all the stages of the data lifecycle: collection/generation; verification; cleaning and processing; and, integration. However, data integrity and responsible data management are the most laborious and boring aspects of AI. The aim of the paper is to present approaches to responsible data management for increased data quality.
1University Library of Southern Denmark, SDU
2The Maersk Mc-Kinney Moller Institute, SDU
Email: evl@bib.sdu.dk
Anne Gerdes: Highlighting the Benefits of Open Foundation Models in Healthcare
In recent years, machine learning has undergone a significant transformation fueled by the rise of foundation models. Such large pre-trained models can be fine-tuned for various downstream tasks. Consequently, integrating large language models (LLMs) in healthcare may enhance clinical workflows, health communication, and diagnostic capabilities. At the same time, it is well-known that LLMS gives rise to epistemic and ethical concerns related to, e.g., factual inconsistency, domain-specific data scarcity, bias, interpretability, explainability, and privacy. Against this setting, it is worth highlighting the role of open foundation models in accelerating research, increasing transparency and diversity, and encouraging well-documented developmental practices within the field.
Department of Department of Design, Media and Educational Science
Email: gerdes@sdu.dk