Notes for a short talk on digital inequality

I am delighted to have been invited to participate in this informal session hosted by Sociology on "Digital (In)equality, activism, and social change."  Panelists were asked to reflect on the role of digital media and access to activism and social change, and some specific questions were posed to us: what is our current moment teaching us about digital equality and inequality? What's important? What's missing? What's shifting?  

I'll try to address these questions in a short statement that I hope will contribute to the other perspectives offered by my colleagues and perhaps spark some discussion. 

Two "teachable moments" about digital (in)equality

I'll begin by suggesting we are experiencing two "teachable moments" when it comes to digital inequality.  

On the one hand, our current movements for change such as BLM, ILM, and #metoo have put a spotlight on a pervasive problem of systemic discrimination across racial and gender lines.  

Systemic discrimination is characterized by the Supreme Court of Canada as

practices or attitudes that have, whether by design or impact, the effect of limiting an individual’s or a group’s right to the opportunities generally available because of attributed rather than actual characteristics. (Manitoba Human Rights Commission, 2014)

The idea here is that this form of discrimination is "baked into" a system through processes, policies, norms and accepted practices.  Sometimes it is overt; other times, less so. 

On the other hand, the disparate effects of the pandemic on marginalized groups in our local communities and around the world have also become apparent. The response to COVID19 has relied heavily on the use of digital technologies, which in turn have revealed significant capability gaps among groups that do not have access to them. 

This has been perhaps most obvious with respect to providing continuity for schooling. Within Edmonton, the Multicultural Health Brokers have been scrambling to locate computers and find ways to provide broadband access for the children of newcomers to Canada. Globally, speaking with my research collaborators in Sri Lanka, they have reported that despite a program that zero-rates connections to the postsecondary education portal, most students are limited to using mobile phones for online learning, a situation which is clearly inadequate.

I would propose these two teachable moments are showing us that systemic discrimination and digital inequality are interconnected in multiple ways, some perhaps less obvious than others. In order to begin to address the problem of digital inequality, we need to see it, at least in part, as a manifestation of systemic discrimination.

Today, I'm going to offer three observations based on a recent report from the UN's Special Rapporteur on contemporary forms of racism.  Released in June, her report is titled "Racial discrimination and emerging digital technologies: a human rights analysis" (OHCHR, 2020).  

Digital divides

First, and perhaps most obvious, systemic discrimination is evident with respect to the digital divide. Patterns of inequality in accessing EDTs tend to follow geopolitical lines at the global level as well as divisions of racial, ethnic and gendered inequality at the local level.

For example, the number of active broadband subscriptions in the Global South is less than half of that in the Global North.  In the least developed countries, only about one in five persons are online, creating obvious challenges for digital readiness with respect to economic development, government services, and of course public health (pandemic) response efforts.

In Canada we have a marked disparity in broadband access with indigenous and northern populations.  Moreover, the closure of public libraries in urban areas due to COVID19, has reduced or eliminated access to the Internet and online services for low-income and marginalized populations.

The inclusive innovation problem

Second, there is growing evidence to show that systemic discrimination is designed into the technology and digital services we use every day.  Let's call this the inclusive innovation problem.  

For example, in a widely cited study, a 2019 review of 189 facial recognition algorithms from developers around the world found that many incorrectly identified black or East Asian faces when compared with a white face.  

In another example, digital employment systems may be designed to sort and filter applications using proxies for race, gender, or religion.  In a related example, the Special Rapporteur's report mentions a case from Paraguay, where the employment system was available only in Spanish, thereby excluding many indigenous jobseekers from rural areas.    

The cultural and political values of software developers, many located in Silicon Valley, shape how these technologies are conceived, designed and implemented.  The effects may be unintentional, but to quote Achiumi on this point, "technology produced in (such) fields that disproportionately exclude women, racial, ethnic and other minorities is likely to reproduce these inequalities when it is deployed" (p. 6). 

Matthew Effects on opportunity structures

My third observation is concerned with effects from technology adoption. The widespread use of EDTs not only facilitates explicit forms of discrimination, such as the spread of racist content and hate speech, but can perpetuate systemic discrimination by shaping other opportunity structures, such as access to health care, policing, or public services. 

For example, studies on AI systems and algorithms used in health insurance industry or predictive policing efforts, show that the datasets on which machine learning and decision support systems are tainted by systemic discrimination, or the datasets used for training these systems may be incomplete or unrepresentative. This leads to a kind of Matthew Effect that generates feedback loops thereby reinforcing, perhaps even exacerbating, systemic bias through the use of these so-called "smart" decision support systems.  

A lawsuit filed in the US last year claims that Facebook's algorithms for targeting ads in the housing market can perpetuate digital redlining by allowing advertisers to target their audiences by family status, disability, and national origin.  The lawsuit argues that this filtering makes it possible "for real estate brokers and landlords to exclude certain buyers or renters from ever seeing their ads." It limits the ability of certain groups to find suitable housing (violating the US Fair Housing Act).  The plaintiffs offered examples where the filtering was used to prevent several categories of users from seeing a housing ad.  These included new parents, African Americans, and Hispanics. 

Most recently, the UK government has had to walk back its use of a controversial algorithm to produce grades in lieu of cancelled A-level exams.  These exams determine university placement for British students and were cancelled due to COVID19.  Results have revealed a pattern of systemic discrimination that appears to correlate with socioeconomic class in that country.  Students from less advantaged areas "were more likely to receive a lower grade than those attending an elite school" due to the algorithm, possibly limiting their postsecondary education opportunities and damaging future career prospects.

Perhaps equally significant is that sometimes the designers of these algorithms are unaware of these biases, perhaps unable to predict and thereby mitigate "the political valence" of their own creations to borrow Tarleton Gillespie's phrase (Gillespie, 2013).

These concerns notwithstanding, technology is often regarded not as part of the problem but a promising solution to inequality.  To quote the Special Rapporteur, "this presumption of technological objectivity and neutrality is one that remains salient even among producers of technology" (p. 4).  This is Mark Zuckerberg claiming that Facebook is just a technology company as if that absolves them from the social impact of his platform.

Tentative steps forward: three key areas for policy and practice

To conclude, I have presented a very brief summary of some evidence to suggest that systemic discrimination and digital inequality are interconnected in three key areas: digital divides, innovation systems, and in the integration of AI/ML systems with other opportunity structures.

So, what's missing?  What's shifting?  

Missing is a greater diversity of representation in the innovation systems that produce EDTs.  This extends to the level of software coding through to user interface design, business strategy, governance, and moderation of these platforms.  Greater inclusivity throughout the innovation system is essential.

Also missing is a recognition that technology is not "colour-blind" but that it must be better governed and regulated if we are to respond to both intended and unintended forms of inequality and systemic discrimination.  This could include the adoption of international principles for social impact assessments on software and platforms, algorithm auditing requirements, greater public sector accountability in the use of EDTs with an emphasis on open source solutions, as well as better regulatory oversight of the tech industry.

So what's shifting? The Special Rapporteur's report is evidence that at the international level through the UN's Human Rights Commission, there is growing recognition that these technologies have a significant impact on "a broad spectrum of human rights" and human rights law obligations, "including on equality and non-discrimination" (p. 20). 

The Council of Europe recently adopted a charter setting out ethical principles related to the use of AI in judicial systems which includes a principle of non-discrimination (CEPEJ, 2019). In Canada, we have a Pan-Canadian Artificial Intelligence Strategy that is explicitly committed to supporting "equity, diversity, and inclusion" in the development of emerging digital technologies (CIFAR, 2020). 

We have the Toronto Declaration, which is a statement from 2018 crafted by Amnesty International and other digital rights groups that proposes human rights to be put front and centre in the design and implementation of AI and ML technologies.  

The Government of Canada's Digital Charter also sets out 10 principles that convey, in part, at least the spirit of non-discrimination and equality concerning emerging digital technologies.  

The CRTC has committed through its Broadband Fund to contribute $72-million to improve connectivity in rural and northern Canada.  The Broadcasting and Telecommunications Legislative Review Panel report released earlier this year, recommended a universal service objective for telecommunications, as well as regulatory oversight of large social media platforms and online content services, which may provide a pathway to greater accountability for the tech giants (The Wire Report, 2020).  

How these play out in policy, regulation, and practice is still to be seen. Still, at least there appears to be growing recognition that, far from being neutral purveyors of information, these technologies can be complicit in systemic discrimination and social inequality.  

If EDTs are to be part of a wider solution in advancing human rights, we will need to better understand and respond to the social processes by which these technologies are conceived, adopted by different groups, and are integrated with other forms of social infrastructure.