Data profiling is a critical data management task that provides an informative view of data in the form of constraints, patterns, and trends within the data. Although data profiling is commonly used to understand and cleanse data, few have explored its potential to facilitate fair learning.
Detecting and mitigating bias in ML pipelines is complex, and data scientists have a strong need for system-level support for such tasks. Humans should be empowered to verify and control bias in these pipelines before decision-makers use such predictive pipelines to determine outcomes in critical domains.
Synthetic data is a valuable tool for machine learning researchers and data practitioners, providing a cost-effective, flexible, fair, and privacy-preserving alternative to real-world data. There are three goals for generating synthetic tabular data: (1) the data should be statistically similar to the distribution of a real data set, (2) the data points should ideally be indistinguishable from real data points, and (3) the data points should be sufficiently different from each other. Achieving all three of these properties can be challenging.
Algorithmic recourse provides explanations to individuals, helps them understand why a particular decision was made, and recommends actionable feedback on how to change the results of machine learning models. However, it is difficult to provide recommendations to groups, where changes now would affect the models in unexpected ways.
As ML models are increasingly integrated into data science pipelines, we believe that data-focused and model-focused explanations are complementary for understanding the behavior of ML models. Data-focused explanations offer three advantages: (1) they support complex queries on model results, (2) they are easy to interpret for non-experts, and (3) they capture the assumptions about data and models in a declarative way.
EthiCSEdu serves as a dynamic learning environment that complements ethics education in computer science. It aims to engage students in simulations that help them consider the potential ethical consequences of the computer systems they design and work with. It aims to equip students with the critical thinking skills needed to address the increasingly complex ethical issues they will encounter in their future careers.
Irresponsible use of AI has the potential to cause harm on an unprecedented scale. As we develop and deploy AI systems, we are compelled to think about the impact of these systems on individuals, populations, and society at large. Responsible AI is a tutorial that addresses issues of ethics, legal compliance, data quality, algorithmic fairness and diversity, data and algorithm transparency, privacy, and data protection.