海角社区

This event will take place at the Center for Science and Thought at the University of Bonn in Germany. For more information, see the link below.

"General Ethical Requirements for AI Certification" 聽a conversation with聽Jocelyn Maclure and Markus Gabriel and launch of the whitepaper by Nicholas聽Kluge Corr锚a and Julia Maria M枚nig.

The research project "Certified AI" explores the question of what a聽certification of "trustworthy AI" could look like. The philosophy team, Nicholas聽Kluge Corr锚a and Julia Maria M枚nig, research this question from an ethical聽perspective.

While many arguments can be raised against certifying "ethical"聽AI, with our latest publication we offer guidance to computer scientists and聽programmers on how to try to make AI-based applications more "ethical". We do聽this by presenting (a list of) ethical principles that are consistent with聽other proposals in the literature. We briefly explain what the values are and聽why they are important, and give examples to align them with the risk聽categories of the European AI Act (2024). To operationalize the principles,聽for each step we propose methods and technical tools that can be used to put聽the ethical requirements into practice.聽In the long run, this could serve as a way to make the tech industry's current聽profit-driven motto of "move fast and break things" fairer, more privacy- preserving, safer and more robust, sustainable and transparent, while聽promoting accountability and truthfulness, if developers adhere to the聽proposed practices.聽

Arguments against ethics 'certification' (including by the authors) are that聽companies and other stakeholders could use it as 'ethics washing'. In聽addition, practices may evolve over time and therefore certification (as in聽other areas) could only be valid for a limited period of time, if at all,聽after which it would need to be renewed. In addition, ethical tools may not聽have the intended effects, as has been shown in the literature, e.g. for the聽example of fairness algorithms. Last but not least, it is difficult to聽'standardise' ethics, as this entails quantification, whereas ethics refers to聽and consists of deliberation, with values that may change over time, and in a聽liberal-democratic system must take into account the views of all聽stakeholders, especially vulnerable groups. We hence suggest that the efforts聽laid out in the paper would need to be accompanied for instance by workshops,聽guidance by professional ethicists and discussions with involved stakeholders.

While the values presented in the white paper resonate with those usually聽attributed to the so-called "Global North", we understand these values as聽being universal, with human dignity and respect at their core. The four聽overarching principles of autonomy, beneficence, no harm and accountability are聽intended to serve all humans on earth, and call for the ethical inclusion of聽all potentially affected stakeholders, 聽e.g. through the principle of聽sustainability that addresses problems caused by the actions of the "Global聽North" on the "Global South".

In this workshop we will present the guidelines in the presence of Markus聽Gabriel, the principal investigator of the philosophy subproject of聽"Zertifizierte KI", and Jocelyn Maclure who will comment on them from a聽philosophical perspective, drawing on his expertise and decade-long experience聽in AI ethics.