Machine learning in tutorials – Universal applicability, underinformed application, and other misconceptions

Abstract

Machine learning has become a key component of contemporary information systems. Unlike prior information systems explicitly programmed in formal languages, ML systems infer rules from data. This paper shows what this difference means for the critical analysis of socio-technical systems based on machine learning. To provide a foundation for future critical analysis of machine learning-based systems, we engage with how the term is framed and constructed in self-education resources. For this, we analyze machine learning tutorials, an important information source for self-learners and a key tool for the formation of the practices of the machine learning community. Our analysis identifies canonical examples of machine learning as well as important misconceptions and problematic framings. Our results show that machine learning is presented as being universally applicable and that the application of machine learning without special expertise is actively encouraged. Explanations of machine learning algorithms are missing or strongly limited. Meanwhile, the importance of data is vastly understated. This has implications for the manifestation of (new) social inequalities through machine learning-based systems.

Audit, Don't Explain -- Recommendations Based on a Socio-Technical Understanding of ML-Based Systems

“…The findings in [11] add to this by recognizing the role of 'ML practitioners' in developing and evaluating ML-based systems. However, the investigation also revealed essential limitations regarding how the significance of data is discussed.…”

“…'ML practitioners' are those who train and develop ML-based systems. They are another group of people that directly influence the ML-based systems [11]. Their objective is to develop and evaluate the ML-based systems as well as the interface with which users interact.…”

Auditing the Biases Enacted by YouTube for Political Topics in Germany

“…While YouTube's published research describes the general idea of their recommendation system in a paper by Covington, Adams & Sargin [13], it remains unclear how the system works and what factors it takes into account. Despite the growing importance of recommendation systems that are based on machine learning (ML), designers and developers' understanding of ML and its applications is only emerging [19,34].…”

“…An ML algorithm merely describes how the ML model is inferred from data. For this reason, ML-based systems cannot be studied using code audits that investigate the source code [34]. This paper applies sock-puppet audits to scrutinize public relevance algorithms like YouTube's recommendation system.…”

The Explanatory Gap in Algorithmic News Curation

“…On social media sites like Facebook and Twitter, ML-based curation systems solve the challenging tasks of selecting, organizing, and presenting news from a variety of sources [12]. While curation is necessary considering the large number of users of social media sites and the immense number of available news stories, ML-based curation systems pose important challenges regarding algorithmic transparency and algorithmic experience [8,36,3,24]. In the past, news curation was a task predominantly performed by skilled journalists, who assessed the newsworthiness of content [48].…”