How Algorithms Discriminate Based on Data They Lack: Challenges, Solutions, and Policy Implications

Williams, Betsy Anne; Brooks, Catherine; Shmargad, Yotam

doi:10.5325/jinfopoli.8.1.0078

Cited by 21 publications

(11 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…These individuals represent a distinct demographic (wealthy owners of premium houses), whose energy use is likely signi cantly different to other groups (such as renters or apartment dwellers). This biases in our data will have predictable as well as unknowable consequences that must to be acknowledged and ought to caution against generalisation 44 .…”

Section: Discussionmentioning

confidence: 93%

Applying responsible algorithm design to neighbourhood-scale batteries in Australia

et al. 2021

View full text Add to dashboard Cite

While the governance of algorithms is of growing societal concern, the energy sector has been slow to engage with this issue. We argue that there are at least three systemic concerns to the design and operation of algorithms in the new, digital energy era. Namely, reliance on algorithms can bias considerations towards the easily quanti able, that they can inhibit explainability, transparency and trust, and that they could undermine energy users' autonomy and control. We examine these tensions through an interdisciplinary study that reveals the diversity and materiality of algorithms. Our study focuses on neighbourhood-scale batteries (NSBs) in Australia as a case study of new energy algorithms. We conducted qualitative research with energy sector professionals and citizens to understand the range of perceived bene ts and risks of NSBs and the algorithms that drive their behaviour. Issues raised by stakeholders were integrated into our development of multiple NSB optimisation algorithms, whose impacts on NSB owners and customers we quanti ed through techno-economic modelling. Our resultsshow the allocation of bene ts and risks vary considerably between different algorithm designs. This insight a need to improve energy algorithm governance, enabling accountability and responsiveness across the design and use of algorithms so that the digitisation of energy technology does not lead to adverse public outcomes. Taken together, our study underscores the importance for researchers and developers of new algorithms to take a holistic view of stakeholders and public bene t, and demonstrates one method to practice responsible algorithm design. MainAlgorithms are playing ever more pervasive and critical roles in the operation of the electricity system as it becomes digitised and decentralised. The design of these algorithms -considering the values and biases they encode -has to date received scant attention by energy researchers and developers. We believe this to be an important omission given recent controversies regarding the reach of algorithmic authority in ever more areas including healthcare 1,2 , hiring practices 3,4 , law enforcement 5,6 and new media's role in social and political life 7 which have exposed deep ethical shortcomings that challenge the "arc of inevitability" of technical solutions 8 .Existing discussions of algorithms in the energy eld have tended to focus on questions of data privacy and the cybersecurity of smart meters and in-home devices 9,10 . While important, we argue that these overlook at least three systemic concerns. The rst is that algorithms tend to narrow considerations to factors that offer plentiful, easily quanti ed data, such as nancial values, excluding public values that may be harder to quantify. Such omissions narrow the conception of energy and have the potential to introduce structural biases 11 . Secondly, the challenging explainability of algorithms exacerbates accountability concerns and distrust of the energy system, and risks creating hidden outcomes 11 such as new forms o...

show abstract

Section: Discussionmentioning

confidence: 93%

Applying responsible algorithm design to neighbourhood-scale batteries in Australia

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Moreover, a well-known problem is that reliance on algorithms may introduce various kinds of bias, including discrimination based on gender, race, or religion (Buolamwini and Gebru, 2018). Algorithms learn from the data that are fed to them, including the prejudices associated with different kinds of groups (Williams et al, 2018). If the output of the models are taken at face value, decisions are likely to reproduce existing biases in society and reinforce discrimination patterns under the appearance of objectivity.…”

Section: Big Data and Algorithmic Decision-makingmentioning

confidence: 99%

Digital Technology, Politics, and Policy-Making

Gilardi

2022

View full text Add to dashboard Cite

This element shows, based on a review of the literature, how digital technology has affected liberal democracies with a focus on three key aspects of democratic politics: political communication, political participation, and policy-making. The impact of digital technology permeates the entire political process, affecting the flow of information among citizen and political actors, the connection between the mass public and political elites, and the development of policy responses to societal problems. This element discusses how digital technology has shaped these different domains, identifies areas of research consensus as well as unresolved questions, and argues that a key perspective involves issue definition, that is, how the nature of the problems raised by digital technology is subject to political contestation.

show abstract

“…In this specific example, the algorithm might create a health risk factor based on the viewing of television programs! To combat this phenomenon, [Williams et al, 2018] show that the collection and use of protected variables should not be prohibited, but rather used as a means of monitoring non-discrimination instead. But this is not easily done either, as will be shown below.…”

Section: Interpretability Vs Accuracy: the Opaque Effectiveness Of Ne...mentioning

confidence: 99%

The Fairness of Machine Learning in Insurance: New Rags for an Old Man?

Barry¹,

Charpentier²

2022

Preprint

View full text Add to dashboard Cite

Since the beginning of their history, insurers have been known to use data to classify and price risks. As such, they were confronted early on with the problem of fairness and discrimination associated with data. This issue is becoming increasingly important with access to more granular and behavioural data, and is evolving to reflect current technologies and societal concerns. By looking into earlier debates on discrimination, we show that some algorithmic biases are a renewed version of older ones, while others show a reversal of the previous order. Paradoxically, while the insurance practice has not deeply changed nor are most of these biases new, the machine learning era still deeply shakes the conception of insurance fairness.

show abstract

How Algorithms Discriminate Based on Data They Lack: Challenges, Solutions, and Policy Implications

Cited by 21 publications

References 59 publications

Applying responsible algorithm design to neighbourhood-scale batteries in Australia

Applying responsible algorithm design to neighbourhood-scale batteries in Australia

Digital Technology, Politics, and Policy-Making

The Fairness of Machine Learning in Insurance: New Rags for an Old Man?

Contact Info

Product

Resources

About