Artificial intelligence: training data must be good, fair and balanced
Aspects to consider with AI
When using AI, how can it be ensured that not only the data quality is right, but also that ethical and legal requirements are met? A free online course that will be launched on April 19 on openHPI, the open education platform of the Hasso Plattner Institute, will introduce this issue of machine learning.
It is led by HPI professor Felix Naumann and three other experts: Media ethicist Jessica Heesen from the University of Tübingen, criminal law professor Frauke Rostalski from the University of Cologne, and standards expert Sebastian Hallensleben from the Association for Electrical, Electronic & Information Technologies. Participants will learn how experts in the fields of computer science, law, ethics and standardization view the issues surrounding Big Data applications from different perspectives.
“We show newcomers to the topic which aspects to pay attention to in data collection and processing one should pay attention to in order to use good, fair and balanced training data and thus also develop fair AI systems.“
Felix Naumann,
Professor of Information Systems,
Hasso Plattner Institute, University of Potsdam
"We show newcomers to the topic which aspects to pay attention to in data collection and processing in order to use good, fair and balanced training data and thus also develop fair AI systems," says Felix Naumann. According to him, requirements such as non-discrimination, consideration of diversity or employee data protection have an impact on the data and processes with which AI models were previously trained. "Conversely, incomplete, erroneous, inappropriate or biased training data lead to unsafe models," Naumann warns.
The results could thus ultimately lead to wrong decisions. Together with the other course instructors, the computer scientist wants to show that the legal requirements for test, validation and training data in machine learning as well as their implementation in norms and standards are also still "largely unresolved". Anyone interested can register for the two-week free course "AI and Data Quality - Perspectives from Data Science, Ethics, Standardization and Law" on the official website: open.hpi.de
