Social Impact of AI Design

Introduction to the technical perspective of the societal impact of design decisions in AI development and deployment.

Details in implementation and design decisions during the training and deployment of artificial intelligence and machine learning systems can have far-reaching consequences for stakeholders as well as society.

This is not only a matter of company policy or regulatory requirements, but also directly affects the Data Scientists who develop and oversee the operation of these models due to the technical depth and expertise required.

This module teaches various pitfalls through technical examples and implementation tasks. The possible consequences are highlighted, and potential workaround strategies are taught.

Students will be able to identify the potential risks to individuals and society associated with datasets and algorithm development and mitigate them with measures.

This includes the different biases in datasets, unthought effects due to limitations of different algorithms, cultural differences and their influences in design decisions, conflicting optimisation metrics as fairness, and feedback effects in the training data pool.

Students have an overview of the possible problems regarding the risks for individuals and society when using AI or machine learning algorithms. Furthermore, students know measures to mitigate the problems.

These include the risk of data leakage due to the ill-considered use, the impact of the algorithms' lack of reflection capacity, the importance of the location and area of use (deployment context) of a system, the lack of neutrality and objectivity of the systems as well as technology credulity of the users and the effects of use, as well as distorting impact effects on the environment of the system.

Students are able to understand and apply the steps of a responsible development cycle for AI models. They have carried out the entire process, from the initial idea to development, operationalisation and long-term use, using a practical example.

This responsible development process consists of an initial assessment phase with deployment evaluation, metrics reflection and impact assessment, a development phase with prototype, data and model versioning, and a deployment phase where a multi-year deployment period is simulated. The aim is to use the knowledge acquired in the previous blocks to identify and avoid the pitfalls.

• GPR
• EDA
• WER
• LLR
• GDS