Overcoming the pitfalls and perils of algorithms: A classification of machine learning biases and mitigation methods

Giffen, Benjamin van; Herhausen, Dennis; Fahse, Tobias

doi:10.1016/j.jbusres.2022.01.076

Cited by 99 publications

(41 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…T HE recent advances in computing power and machine learning (ML) have given rise to several innovations and developments in many areas of research and human lives. In the last few years, advancements in machine learning, a subset of artificial intelligence (AI), have transformed and inherently changed almost every area of our lives [1], [2]. Particularly, ML has been applied in disease diagnosis [3], fraud detection [4], text classification [5], and image recognition [6], among others.…”

mentioning

confidence: 99%

A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects

2022

View full text Add to dashboard Cite

Ensemble learning techniques have achieved state-of-the-art performance in diverse machine learning applications by combining the predictions from two or more base models. This paper presents a concise overview of ensemble learning, covering the three main ensemble methods: bagging, boosting, and stacking, their early development to the recent state-of-the-art algorithms. The study focuses on the widely used ensemble algorithms, including random forest, adaptive boosting (AdaBoost), gradient boosting, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost). An attempt is made to concisely cover their mathematical and algorithmic representations, which is lacking in the existing literature and would be beneficial to machine learning researchers and practitioners.

show abstract

mentioning

confidence: 99%

A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects

2022

View full text Add to dashboard Cite

show abstract

“…Despite the benefits of algorithmic decision making, algorithms also pose at least four serious risks. First, they can actually amplify social biases, thus giving rise to DELIBERATE IGNORANCE 6 unfair decisions and outcomes that reinforce existing discrimination and inequalities (van Giffen et al, 2022;Wachter et al, 2021). Second, they can be used to manipulate people-citizens, voters, consumers-and distort economic and political competitions (Lewandowsky et al, 2020;Lorenz-Spreen et al, 2020).…”

Section: Algorithms As a New Methods Of Social Choicementioning

confidence: 99%

“…Why then do they have the potential to produce and amplify social biases rather than remove them? In general, biases can enter algorithms in two ways: via the implicit or explicit biases of their developers, and via the historical data used to train the algorithms ("biases in, biases out"; for examples and a classification of biases, see Mehrabi et al, 2021;van Giffen et al, 2022).…”

Section: Algorithms As a New Methods Of Social Choicementioning

confidence: 99%

Blinding to circumvent human biases: Deliberate ignorance in humans, institutions, and machines

Hertwig¹,

Kozyreva²,

Herzog³

2022

Preprint

View full text Add to dashboard Cite

Persistent inequalities and injustices are a blight on modern liberal societies. Examples abound, from the gender gap in pay to sentencing disparities between Black, Hispanic, and White defendants to allocation disparities in medical resources between Black and White patients. One cause of these and other inequalities is implicit social biases. In a process thought to be outside conscious control, human cognition is assumed to make associations between social groups and attributions such as "nurturing," "lazy," or "uneducated." Such associations can result in explicit unequal treatment. In theory, one way to circumvent implicit but, of course, also explicit biases is to delegate important decisions—for instance, on allocating benefits, resources, or opportunities—to algorithms, which are assumed to be free of human biases. However, evidence shows that algorithms can perpetuate and even amplify existing inequalities and injustices. We discuss how a philosophical thought experiment, Rawls's "veil of ignorance," and a psychological phenomenon, "deliberate ignorance" (the choice to not know), can shield people, institutions, and algorithms from biases. We explore the advantages and ways of blinding human and artificial decision makers to information that could result in biases and consider the practical problems of successfully blinding algorithms.

show abstract

“…While AI is supposed to assist us in overcoming our challenges, research reveals that with the advancement of AIoT, there is a major risk that the assumptions and biases incorporated during the design and development of AI systems might adversely influence the AIoT decisions (Giffen et al , 2022). The unintended effects induced by biases and the opacity of the results constitute a hurdle for its integration into the public sector (Alon-Barkat and Busuioc, 2022).…”

Section: Integration Of Artificial Intelligence Of Things In the Publ...mentioning

confidence: 99%

Integration of artificial intelligence of things (AIoT) in the public sector: drivers, barriers and future research agenda

Ishengoma

Shao²,

Alexopoulos³

et al. 2022

DPRG

View full text Add to dashboard Cite

Purpose With the development of information technology (IT), governments around the globe are using state-of-the-art IT interfaces to implement the so-called 3E’s in public service delivery, that is, economy, efficiency and effectiveness. Two of these IT interfaces relate to Artificial Intelligence (AI) and Internet of Things (IoT). While AI focuses on providing a “human” garb for computing devices, thereby making them “intelligent” devices, IoT relies on interfaces between sensors and the environment to make “intelligent” decisions. Recently, the convergence of AI and IoT – also referred to as Artificial Intelligence of Things (AIoT) – is seen as a real opportunity to refurbish the public service delivery formats. However, there is limited understanding as to how AIoT could contribute to the improvisation of public service delivery. This study aims to create a modular framework for AIoT in addition to highlighting the drivers and barriers for its integration in the public sector. Design/methodology/approach This descriptive-explanatory study takes a qualitative approach. It entails a thorough examination of the drivers and barriers of integrating AI and IoT in the public sector. A review of literature has led to the development of a conceptual framework outlining the various factors that contribute to creating public value. Findings Value creation occurs when AI and IoT coalesce in the public service delivery mechanisms. Originality/value AIoT is a cutting-edge technology revolutionizing health care, agriculture, infrastructure and all other industrial domains. This study adds to the growing body of knowledge on the public sector's use of AI and IoT. Understanding these disruptive technologies is critical to formulating policies and regulations that can maximize the potential benefits for the public-sector organizations.

show abstract

Overcoming the pitfalls and perils of algorithms: A classification of machine learning biases and mitigation methods

Cited by 99 publications

References 36 publications

A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects

A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects

Blinding to circumvent human biases: Deliberate ignorance in humans, institutions, and machines

Integration of artificial intelligence of things (AIoT) in the public sector: drivers, barriers and future research agenda

Contact Info

Product

Resources

About