Artificial Intelligence is fundamentally changing various fields within society and the economy. As promising as these possibilities are, the digital transformation also raises concerns and the need for regulation, such as the EU’s AI Regulation, which came into effect last year and which aims to promote responsible use of AI.

Interdisciplinary study

How can this be implemented in practice? “We have to consider the influence of people throughout the entire life cycle of an AI,” says Elia, research assistant at the Institute of Computer Science at the 威尼斯赌博游戏_威尼斯赌博app-【官网】 of Augsburg. She has developed a holistic, ethical approach for responsible use of AI.

Elia was supported by Paula Ziethmann (CReAITech), Prof Kerstin Schl?gl-Flierl from the Chair of 威尼斯赌博游戏_威尼斯赌博app-【官网】 Moral Theology, her doctoral supervisor Prof Bernhard Bauer, Professor for Software Methodology for Distributed Systems at the 威尼斯赌博游戏_威尼斯赌博app-【官网】 of Augsburg and member of the CReAITech executive board, as well as by Dr Julia Krumme from the 威尼斯赌博游戏_威尼斯赌博app-【官网】 of Applied Sciences Augsburg.

Thinking about the big picture

“What mindset should people who are developing, designing, implementing, and evaluating AI have?” This is how Elia summarises the guiding question of her research. One answer is that it is important to be aware of the bigger picture because developers often have little connection to the domain where the AI will be used.

The question of how AI and the real world influence each other is currently the subject of much research. “Here, it is helpful to ask the right questions. With our approach, we would like to point out some directions and possibilities for implementation,” explains Elia. ?

Four criteria for responsible use of AI

Elia's analysis identifies four criteria that could help raise awareness of this:

• Generalisability: The AI should be able to reliably function in different situations. For example, a medical AI that helps to diagnose skin diseases should not only be able to recognise these in specific skin types (e.g., fair skin) but in all skin types (e.g., dark skin as well). The AI therefore needs to be trained using a wide range of data.
• Adaptability: The AI should be able to adapt to new circumstances. “The world in which we live is constantly changing. The AI needs to be able to adapt, like us humans,” explains Elia. Environmental influences, climate change, wars, pandemics: An AI that helps to evaluate environmental risks, for example, needs to be able to take such events into account.
• Translationality: The AI should be able to be integrated into existing circumstances: “Where should the AI be used? What workflows and IT infrastructure is available? What skills do the users already have and what kind of training is necessary?” explains Elia. When a new medical tool is to be used, medical personnel should be trained to avoid inadvertent misuse.
• Transversality: Anyone who develops AI should consider social and ethical aspects to avoid distortions. “We are introducing this term for AI and software development for the first time,” explains Elia. “It is about thinking about the society in which we live and about how we understand fairness, for example.” From Elia’s perspective, many terms used like “fairness” and “bias” are complex, loaded, and have another focus. “Therefore, we are using a neutral term that describes, first of all, the underlying mechanism more objectively,” says Elia. ?

For example, to develop a medical AI that treats male and female patients equally more data from female patients is in many cases needed. This data gap is real, and it influences AI. Developers should consider this and introduce a pop-up box in the application to remind medical personnel that the results for female patients are probably less accurate. ?

Easy to use

What makes their approach unique is that the entire life cycle of an AI is considered and aimed directly at interdisciplinary developers. Online materials are also provided for them to interact with where it concerns ethics in the context of individual projects. “We have tried to translate the concepts in a clear and easy to understand way,” says Elia.

It is clear to her that dealing with ethical criteria should not mean a lot of extra work for users. “Everyone is always short on time. Therefore, ethics must be as easy as possible to integrate,” says Elia. Her publication and training materials should help. She is particularly focused on areas with a high level of social responsibility and risk, such as medicine, education, and the public sector.

?

cg

Further reading

Miriam Elia, Paula Ziethmann, Julia Krumme, Kerstin Schl?gl-Flierl and Bernhard Bauer. in press. Responsible AI, ethics, and the AI lifecycle: how to consider the human influence?, AI and Ethics. DOI: 10.1007/s43681-025-00666-z
URL: https://link.springer.com/article/10.1007/s43681-025-00666-z

Training material on Responsible AI at GitHub (Compiled by Dr Julia Krumme): https://github.com/miriamelia/MQG4AI/blob/main/MQG4DesignKnowledge/1_System/Application/Ethics_Specific/Ethics_Specific.md