Proper enforcement of hate speech laws is key for protecting groups of people against harmful and discriminatory language. However, determining what constitutes hate speech is a complex task that is highly open to subjective interpretations. Existing works do not align their systems with enforceable definitions of hate speech, which can make their outputs inconsistent with the goals of regulators. This research introduces a new perspective and task for enforceable hate speech detection centred around legal definitions, and a dataset annotated on violations of eleven possible definitions by legal experts. Given the Figure 1: A visualization of our proposed method to challenge of identifying clear, legally enforceable ground hate speech to specialized legal definitions. A instances of hate speech, we augment the legal professional reads external legal resources and dataset with expert-generated samples and an makes a judgement on some hate speech input, then automatically mined challenge set. We experiment identifies offences according to our definitions and with grounding the model decision in makes a judgement on violations.

Several European artificial intelligence projects rely on race without explicitly saying so. In February, El Confidencial revealed that Renfe, the Spanish railways operator, published a public tender for a system of cameras that could automatically analyze the behavior of passengers on train platforms. One characteristic that the system should be able to assess was "ethnic origin". Ethnic origin can mean many things. But in the context of an automated system that assigns a category to people based on their appearance captured by camera the term is misleading.

"We're not quite sure what features our algorithm is looking for," says Professor Laurenz Wiskott from the Institute for Neural Computation. This is because the system has learned to assess faces. The successful algorithm developed by the Bochum-based researchers is a hierarchical neural network with eleven levels. As input data, the researchers fed it with several thousand photos of faces of different ages. The age was known in each case.

Applications based on Machine Learning models have now become an indispensable part of the everyday life and the professional world. A critical question then recently arised among the population: Do algorithmic decisions convey any type of discrimination against specific groups of population or minorities? In this paper, we show the importance of understanding how a bias can be introduced into automatic decisions. We first present a mathematical framework for the fair learning problem, specifically in the binary classification setting. We then propose to quantify the presence of bias by using the standard Disparate Impact index on the real and well-known Adult income data set. Finally, we check the performance of different approaches aiming to reduce the bias in binary classification outcomes. Importantly, we show that some intuitive methods are ineffective. This sheds light on the fact trying to make fair machine learning models may be a particularly challenging task, in particular when the training observations contain a bias.