Interdisciplinary Science Reviews

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/yisr20

Beyond ‘AI for Social Good’ (AI4SG): social

transformations—not tech-fixes—for health equity

Cheryl Holzmeyer

To cite this article: Cheryl Holzmeyer (2021) Beyond ‘AI for Social Good’ (AI4SG): social
transformations—not tech-fixes—for health equity, Interdisciplinary Science Reviews, 46:1-2,
94-125, DOI: 10.1080/03080188.2020.1840221

To link to this article: https://doi.org/10.1080/03080188.2020.1840221

Published online: 07 Mar 2021.

Submit your article to this journal

Article views: 1165

View related articles

View Crossmark data

Citing articles: 11 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=yisr20
INTERDISCIPLINARY SCIENCE REVIEWS
2021, VOL. 46, NOS. 1–2, 94–125
https://doi.org/10.1080/03080188.2020.1840221

Beyond ‘AI for Social Good’ (AI4SG): social

transformations—not tech-fixes—for health equity
Cheryl Holzmeyer
Institute for Social Transformation, University of California, Santa Cruz, CA, USA

ABSTRACT KEYWORDS
This paper reflects on proliferating AI for Social Good (AI4SG) Artificial Intelligence (AI); AI
initiatives, with an eye to public health and health equity. It for Social Good (AI4SG); AI
notes that many AI4SG initiatives are shaped by the same Ethics; AI Governance;
Precision Medicine; Social
corporate entities that incubate AI technologies, beyond
Justice; Public Health; Health
democratic control, and stand to profit monetarily from Equity
their deployment. Such initiatives often pre-frame systemic
social and environmental problems in tech-centric ways,
while suggesting that addressing such problems hinges on
more or better data. They thereby perpetuate incomplete,
distorted models of social change that claim to be ‘data-
driven’. In the process, AI4SG initiatives may obscure or
‘ethics wash’ all the other uses of big data analytics and AI
that more routinely serve private interests and exacerbate
social inequalities. As a case in point, it discusses the
prominence of health-related applications in AI and big
data fields, alongside the politics of more ‘upstream’ versus
‘downstream’ health interventions.

Introduction
As the Covid-19 pandemic alters everyday landscapes of possibility around the
globe – in so many disparate, fractured ways – another kind of emerging land-
scape is being articulated under the umbrella of ‘Artificial Intelligence (AI) for
Social Good’ (Tomašev, Cornebise, and Hutter 2020). Championed by stake-
holders from Big Tech companies to corporate management consultancies to
the United Nations (UN), AI for Social Good (AI4SG) projects (or sometimes
simply, AI for Good) emphasize computation and the deployment of big data
analytics, including machine learning, to address a wide range of social and
environmental issues. While AI4SG projects are often just another way to
frame the AI activities of conventional profit-oriented business entities, the
formal field of AI4SG is gathering momentum, basking in the glow of an appar-
ent ‘AI summer’. A recent survey of the field found over 1000 published papers
on AI4SG topics (Shi, Wang, and Fang 2020, 1), growing from 18 papers in

CONTACT Cheryl Holzmeyer caholz@berkeley.edu Institute for Social Transformation, University of

California, Santa Cruz, CA, USA
This article has been corrected with minor changes. These changes do not impact the academic content of the
article.
© 2020 Institute of Materials, Minerals and Mining Published by Taylor & Francis on behalf of the Institute
 INTERDISCIPLINARY SCIENCE REVIEWS 95

2008 to 246 papers in 2019 (Shi, Wang, and Fang 2020, 5). AI4SG projects
might identify endangered species in digital image streams to aid conservation
efforts; they might analyze satellite data to monitor manifestations of climate
change, such as sea-level rise or desertification; or they might support health
diagnostics, for instance, by detecting skin cancers based on mobile phone
photos (Chui et al. 2018). They might help with Covid-19 contact tracing or
symptom tracking. Such projects and their claims to ‘do good’ are part of the
broader landscape of efforts to define responsible innovation and ethical AI
(Ulnicane et al., this issue), with the vast majority of formal, non-binding
ethics guidelines articulated in a handful of the world’s wealthiest countries
(Jobin, Ienca, and Vayena 2019, 391). AI4SG is also part of the longer lineage
of ‘Technology for Good’ initiatives, including the field of Information and
Communication Technologies for Development.
This essay reflects on proliferating AI4SG initiatives, with an eye to public
health and health equity in a US context, engaging questions that emerged in
part from the author’s experiences with projects that attempted to leverage
digital technological infrastructures, including big data analytics, on behalf of
public health (Holzmeyer 2018a, 2018b, 2018c). It draws as well on a broader
literature review and consideration of existing political, economic and social
structures shaping ‘innovation’ in relation to health equity, meaning a society
in which ‘everyone has the opportunity to attain their highest level of health’
(American Public Health Association [APHA]; emphasis added). It highlights
ways in which existing metrics and goals of innovation, within and beyond
AI, such as patents and GDP growth, are deeply inadequate (or even regressive)
as indicators of and guides to a flourishing, inclusive, sustainable, democratic
society (Mazzucato 2018). It asserts that, indeed, Another World is Possible
(contra Margaret Thatcher’s ‘TINA’ claim that There Is No Alternative),
beyond the confines of the neoliberal market fundamentalism that for
decades has been integral to policies shredding social safety nets and the under-
pinnings of well-being for so many people and ecosystems around the world
(Beckfield 2018; Coburn 2010; Wilkinson and Pickett 2019), including in a sup-
posedly innovative and materially wealthy society such as the USA. These social
and environmental justice issues should be integral to science and technology
policy-making going forward, with health equity as a touchstone for innovation
and STEM (Science, Technology, Engineering, and Mathematics) education
(Holzmeyer 2017), in order to realize the vision articulated in the World
Health Organization (WHO) Constitution of ‘the highest attainable standard
of health as a fundamental right of every human being’, with health described
in the preamble as ‘a state of complete physical, mental and social well-being
and not merely the absence of disease or infirmity’ (WHO 2017).
Moreover, whether or not they are explicitly engaged with conversations
around emerging technology governance, the social movements that are cur-
rently demonstrating in the USA and beyond for racial, economic, climate
96 CHERYL HOLZMEYER

and other forms of social and environmental justice are true thought leaders in
imagining the new social compacts that should be the baselines and foundations
of democratic AI governance, including of any empirically reasonable claims of
‘AI for Social Good’ (and various leading organizations, such as Data for Black
Lives [https://d4bl.org/], are engaged in both social movement activism and
technology governance conversations; c.f. Milner 2018, 2020). This social
movement thought leadership is echoed in statements by progressive academic
scholars in fields from economics (Sachs et al. 2020) to public health (Bassett
2015; Krieger 2020), if not as yet in any thoroughgoing, institutionalized way
in Big Tech’s AI public engagement initiatives (with ‘solidarity’ being one of
the underrepresented values in AI ethics guidelines to date; c.f. Jobin, Ienca,
and Vayena 2019, 396).
This paper proceeds with an examination of the ambiguous meanings of AI
and AI4SG – and the politics, harmful externalities and alternative social worlds
that may be obscured and foreclosed by the ‘social good’ presumptions of
AI4SG. As a case in point, it discusses the prominence of health-related appli-
cations in AI and big data fields in the USA, including in avowed AI4SG pro-
jects. It examines the politics of different types of health interventions,
particularly the tension between growing investments in ever more individua-
lized treatments in downstream clinical settings, on the one hand, and under-
investment in upstream social determinants of health and the public health
sector, on the other. Given that people’s health is largely shaped by the daily
conditions in which they are born, grow, live, work and age – conditions col-
lectively described as the ‘social determinants of health’ (WHO 2008) – this
skewed distribution of resources is simultaneously irrational and unjust, perpe-
tuating racialized, gendered and classed health inequities. While AI and AI4SG
projects did not initiate these tensions and inequalities, big data analytics and
AI systems are in many ways primed to exacerbate and conceal them. These
points provide an entrée to reflecting on the expanding universe of efforts to
‘do good’ through and with AI. In the process, this paper delves into the con-
tradictions, conflicts of interest and harmful consequences – the silences – of
AI4SG projects; it highlights social and political issues that such tech-centric
projects tend to disregard. In closing, it discusses alternative possibilities to
foster progressive social change and democratic renewal – whether alongside,
or in spite of, existing and emerging AI technologies.

AI4SG silences: private interests, technological harms and social

foreclosures
Both AI and AI4SG are ambiguous terms. Though references to AI and deploy-
ments of computer automation in everyday life are ubiquitous – from respond-
ing to phone calls to processing drive-thru orders to tailoring online marketing
– AI eludes straightforward definition; it’s meanings are multiple and
 INTERDISCIPLINARY SCIENCE REVIEWS 97

historically contingent, depending on, among other things, contemporary tech-

nological affordances, social agendas and interpretations, and corresponding
parameters of machine ‘intelligence’. For instance, according to Colin Garvey
(2018), one crucial shift in AI history took place between the waning of the
‘expert systems’ paradigm (late 1970s‒80s) and the rise of the ‘machine learn-
ing’ paradigm (2010-present). Expert systems AI, focused on programming
specific human expertise and decision-making (like medical diagnostics) into
computers, though more successful than earlier quests to develop general AI,
was limited by ‘knowledge acquisition bottlenecks’ in tapping and translating
the complexities of experts’ knowledge to computers (Garvey 2018).
However, the growth of digital big data streams and increased computational
power enabled a new AI paradigm that sidestepped such bottlenecks:
machine learning AI, through which computers register or ‘learn’ patterns
directly from massive data sets, on which algorithms are said to be trained
(e.g. to identify images of ‘bicycles’ or ‘angry’ facial expressions). Such technical
processes may be the basis for both AI4SG projects as well as AI and big data
operations in society more generally. While a fuller discussion of AI techniques
and their nuances is beyond the scope of this paper, there is evidence that
machine learning capabilities are increasingly prominent in AI4SG efforts
across domains, especially in healthcare (Shi, Wang, and Fang 2020, 7).
That said, people may also invoke AI’s cutting-edge aura to hype up compu-
terized data analysis projects, which, upon closer examination, involve nothing
more than conventional statistical analyses of large data sets; yet such big data
analytics may be included in AI4SG landscape analyses (Chui et al. 2018, 2).
Though these large-scale data analyses may be impressive, such statistical cal-
culations have little in common with the machine-based learning processes
and neural network algorithms that many currently believe define AI. Never-
theless, such stories can exploit the ambiguous, emergent, speculative nature
of AI to build up the AI4SG brand, and the branders themselves, opening up
new funding streams in the process. The lines blur between big data analytics,
AI, AI4SG and supposed breakthroughs, across fields of science and technology
(Cooper and Paneth 2020a, 2020b; Fortun 2008), with some analysts observing,
‘Online tech-hyping articles are now driven by the same dynamics as fake news’
(Funk 2019). These stories also participate in the broader, persistent mythology
of technological solutionism and tech-fixes for social and political problems
(Huesemann and Huesemann 2011), including emerging variants of compu-
tational social science in which researchers claim that new big data streams
‘will make it possible to learn far more about society and to eventually start
solving – actually solving – the major problems that affect the well-being of
human populations’ (Ledford 2020, 329).
The nascent AI4SG field is embedded in these technical and cultural con-
texts, led by many of the same corporate actors that are incubating AI
systems broadly. Stakeholders shaping this landscape include: (1) Big Tech
98 CHERYL HOLZMEYER

companies, including Amazon, Apple, Facebook, Google, IBM and Microsoft,

which have formed a Partnership on AI to Benefit People and Society (https://
www.partnershiponai.org/), beginning in 2016; (2) corporate management con-
sultancies (e.g. McKinsey, Deloitte, Accenture), which frame the field in stra-
tegic, opportunistic terms overlapping with older discourses of e.g. corporate
social responsibility; (3) established and new NGOs that seek to deploy AI tech-
nologies in their work (from groups that may use AI as one of their many activi-
ties, such as Amnesty International, to NGOs founded around the use of AI in a
specific context, such as Talking Points); and (4) the UN, which since 2017 has
hosted an annual AI for Good Global Summit (https://aiforgood.itu.int/),
oriented toward addressing its 17 Sustainable Development Goals (SDGs).
Hence the UN’s SDGs have become one influential reference point for many
AI4SG projects (e.g. Google 2019; Chui et al. 2018). Numerous academic
research centres aspire to contribute to the AI4SG field as well. Government
policy roadmaps also use AI4SG language and frames, highlighting ways that
AI can contribute to ‘the public good’ and to addressing ‘grand challenges’.
Meanwhile scholars have found such reports, touting values such as transpar-
ency, accountability and ‘positive impact’, to be short on details, long-term
strategies and meaningful commitments to achieve such impacts (Cath et al.
2017).
Whatever their intentions, such language and frames are not only vague; they
foreground a presumed ‘social good’ while directing attention away from the
cultural, economic and political power commanded by Big Tech companies
and the harmful externalities inherent in burgeoning AI systems. Indeed, AI
and AI4SG are in many ways constituted by and entrench colonial relations
(Adams, this issue; Mohamed, Png, and Isaac 2020), characterized by hierarchy,
extraction, exploitation and malevolent paternalism (Amrute 2019), within and
across nation‒state borders. For example, in an era of worsening climate
change, researchers have drawn attention to the enormous quantities of
energy required to train algorithms on big data sets, leading to huge carbon
footprints (Strubell, Ganesh, and McCallum 2019), with consequences that
will most severely affect the most vulnerable communities. In addition, scholars
have analyzed how training algorithms entails extensive low-wage work,
including to clean and label the underlying data and to monitor on-going algo-
rithmic operations for accuracy, in conditions that are frequently exploitative,
particularly in the Global South (Gray and Suri 2019). For these reasons, among
others, Jared Moore has proposed reorienting AI4SG toward ‘AI for Not Bad’,
to more accurately characterize the complexities at stake:
‘AI for social good’ speaks to the desire of many of practitioners to share what oppor-
tunities they have. It sounds nice. It imagines a world of lucrative careers optimized to
better humanity. The world is not so simple … AI practitioners, like myself, are part
of the prospecting of science from which we hope for gold, but in which we will likely
find just sand – and perhaps leave in our tailings environmental damage and labor
 INTERDISCIPLINARY SCIENCE REVIEWS 99

displacement. Lest that be so, we must be honest about what we are doing and what
we might do better’ (Moore 2019, 6).

However, these reflections and their underlying empirical grounding are a far
cry from the dominant refrains of AI4SG practitioners and advocates
(Google AI Impact Challenge 2019), even as AI4SG initiatives may increasingly
sound notes of greater skepticism, precaution and reflexivity (Tekisalp 2020) in
response to such critiques (Latonero 2019). More significant are the issues typi-
cally left off the AI4SG table entirely: AI’s own ecological footprints; AI’s racia-
lized, gendered labour inequalities; the broader social costs of AI’s labour
market ‘disruptions’, including not only labour displacement but greater
labour surveillance; and meaningful government regulation and democratic
governance of AI R&D, including Big Tech stakeholders – to name a few. Ulti-
mately these silences foster foreclosed social possibilities, with US policies ‘dis-
torted against labour and in favour of capital’ (Acemoglu, Manera, and Restrep
2020), even as the growth of the AI4SG field accelerates.

Big data, AI and public health

Currently, health and medicine are a leading application sector for big data ana-
lytics, machine learning and AI, with health care ‘the largest market for invest-
ment in the emerging AI/ML business sector’ (Ostherr 2019; the underlying
business analytics data are proprietary, however). This popularity includes
AI4SG projects, which, as noted in the introduction, are often synonymous
with commercial AI activities in general, in settings from Big Tech companies
to start-ups (though perhaps rebranded as ‘social enterprises’). Google’s 2019
report on its first Google AI Impact Challenge, which analyzed the AI4SG land-
scape revealed by the 2,602 project proposals it received from organizations in
119 countries (of which 20 received funding and capacity-building support),
found that, ‘Of the sectors represented, health-related applications were the
most common, representing more than 25% of total submissions’ (Google
2019, 7). A 2018 McKinsey AI4SG white paper also found the greatest
number of actual or potential use cases in its ‘Health and hunger’ domain, com-
prising 28 of 160 use cases (Chui et al. 2018). Meanwhile the AI4SG literature
survey mentioned above found more papers in the ‘Healthcare’ sector (com-
prising both clinical care and public health) than in any of the eight application
domains it identified (Shi, Wang, and Fang 2020, 5–6), with this sector account-
ing for 32% of the 2019 AI4SG literature surveyed, as well as having the highest
rate of growth in recent years, such that ‘the difference between it and other
domains appears to be widening’ (Shi, Wang, and Fang 2020, 5). Articulating
their enthusiasm for leveraging big data for public health, Muin Khoury and
John Ioannidis write, ‘Big Data stands to improve health by providing insights
into the causes and outcomes of disease, better drug targets for precision medi-
cine, and enhanced disease prediction and prevention’ (Khoury and Ioannidis
100 CHERYL HOLZMEYER

2014, 1054). AI and big data work hand in hand in such a project – an interplay
of technology, analysis and mythology (boyd and Crawford 2012) – identifying
patterns across datasets from genomics to electronic health records (EHRs) to
infectious disease outbreak data to social media and financial data.
While on the face of it these trends and new technological affordances may
seem to be good news for those concerned about health as a social good, such
AI4SG and AI interventions tend to miss the mark in multiple ways when it
comes to promoting public health and health equity, rooted in social justice
(Krieger and Birn 1998). Potential issues include: (1) distracting attention
and resources away from root causes of population health inequities and
upstream social determinants of health; (2) enabling new risks and vulnerabil-
ities, especially for already marginalized communities; (3) failing to adequately
grapple with the politics of data, knowledge and expertise in the ‘computational
turn’ to big data analytics and AI; and (4) fostering new data treadmills, rather
than emphasizing existing warrants for action based on current bodies of
research. The points below unpack some of these issues and their implications
for public health, so-called AI4SG projects and AI governance, with an eye to
health equity.

1) Distracting from root causes and upstream social determinants of

health
First, the emphasis on the novel possibilities of big data and AI for public
health, in A4SG projects and more generally, often dovetails with a particular,
narrow, empirically inadequate framing of health: the so-called medical model
(Iton 2010), which has long been in tension with more capacious, social-eco-
logical frameworks oriented toward population health promotion, prevention,
social justice and health equity (Brandt and Gardner 2000; Fairchild et al. 2010).
The medical model, conceptually and practically, emphasizes downstream,
individualized interventions in health care settings, increasingly wedded to
emerging precision medicine treatments, for which big data and AI are enabling
technologies (Ho et al. 2020; Hager et al. 2017; Schork 2019; Sun et al. 2018). As
noted in the AI4SG literature survey, ‘The majority of research efforts in AI for
healthcare are focused on clinical care’, citing the availability of EHRs and
medical images as having ‘directly facilitated the AI research in disease diagno-
sis, clinical treatment, and clinical prediction’ (Shi, Wang, and Fang 2020, 20).
Among these applications, ‘Disease diagnosis is the topic in healthcare that has
seen the most applications of AI’ (again, with this survey including both clinical
care and public health in its ‘healthcare’ category; Shi, Wang, and Fang 2020,
20). For example, there are now AI-based smartphone apps for diagnoses of
myriad health issues, from chronic obstructive pulmonary disease to pneumo-
nia and acute asthma (Stewart 2017) to skin cancer (Comstock 2018; Chui et al.
2018) to inflammatory skin diseases (Wu, Yin, and Chen 2020) and beyond, in
 INTERDISCIPLINARY SCIENCE REVIEWS 101

addition to the many apps for monitoring and managing health issues, such as
heart disease (Harvard Heart Letter 2019). The Nuffield Council on Bioethics
has also highlighted the preponderance of clinical biomedical applications of
AI in health care and research (2018, 3-4). Meanwhile the need for not only
medical diagnosis but also access to care may be referred to in the AI4SG
field as a ‘last mile problem’ (Chui et al. 2018, 20).
What is not explicitly discussed in this AI4SG survey, nor in other discus-
sions that centre AI and big data as means of advancing health, are the politics
of different types of health interventions: in particular, in a US context, the
tension between growing investments in ever more individualized treatments
in downstream clinical settings, on the one hand, and persistent underinvest-
ment in upstream social determinants of health and the public health sector,
on the other (Bradley and Taylor 2013; Interlandi 2020). Yet as one recent
article put it: ‘An undeclared civil war is breaking out in biomedicine. On the
one side is precision medicine, with its emphasis on tailoring treatments to
ever-narrower groups of patients. On the other side is population health,
which emphasizes predominantly preventive interventions that have broad
applications across populations’ (Cooper and Paneth 2020b). For while pre-
cision medicine privileges individual-level variables – especially genetics, in
practice – it deemphasizes the everyday environments and economic and
social resources (e.g. adequate income, affordable housing, access to healthy
food, freedom from discrimination, clean air and water, accessible parks,
high-quality education, health insurance) that are most consequential to
health (Braveman and Gottlieb 2014; Galea 2019; WHO 2008). It also neglects
the political and social structures that shape the distribution of these resources
(Beckfield 2018). In the process, precision medicine decentres structural racism
and exploitative economic relations, intertwined with people’s diverse intersec-
tional identities, as root causes of public health inequities (c.f. Figure 1), as illu-
minated by multi-level ecosocial frameworks in contemporary social
epidemiology (Krieger 2011, 214, 287). As physician and public health leader
Anthony Iton puts it, ‘Common disease roots in the socioecological context
are often ignored [in the medical model]’ (Iton 2010, 512). In addition, the
medical model of health has often been accompanied by unscientific, false con-
ceptions of biological race in clinical research and practice, including by posit-
ing ‘race’ – rather than racism – as a cause of racialized health inequities (Boyd
et al. 2020; Chadha et al. 2020).
While AI and AI4SG projects did not create these issues and inequalities, big
data analytics and AI systems are in many ways primed to exacerbate and
conceal them. For example, a range of scholars have highlighted the problems
with the US’s disproportionate, growing investment in an individually focused,
clinically oriented, AI-enabled precision medicine agenda versus a population
health and health equity agenda, focused on investments in upstream social
determinants of health (Bayer and Galea 2015; Cooper and Paneth 2020a;
102 CHERYL HOLZMEYER

Figure 1. A Public Health Framework for Reducing Health Inequities. Source: Bay Area Regional
Health Inequities Initiative, as cited in ‘Portrait of Promise: The California Statewide Plan to
Promote Health and Mental Health Equity’, California Department of Public Health (CDPH),
Office of Health Equity (2015), 17.

Chowkwanyun, Bayer, and Galea 2018; Ferryman and Pitcan 2018). Ferryman
and Pitcan’s ‘Fairness in Precision Medicine Study’ (2018), assessing equity
issues raised by precision medicine, concludes that not only is there potential
for biased datasets, as big data advocates acknowledge, there is also the rarely
addressed problem of biased outcomes due to diversion of resources away
from structural, social determinants of health (Ferryman and Pitcan 2018,
34). This point speaks to longstanding skews in US health investments and out-
comes, compared with other high-income countries (Kristof 2020; Woolf and
Aron 2013). However, as scholars Bradley and Taylor document in The Amer-
ican Healthcare Paradox (2013), the purported paradox – of high per capita US
healthcare spending, coupled with worse US population health outcomes – dis-
appears once the US’s much lower spending on social services and safety net
programs is taken into account. That is, rather than promoting health for all
through upstream access to resources for flourishing, the USA chooses to med-
icalize the health effects of racialized, gendered, classed deprivation down-
stream, through the health care system (and to provide social services that
are not only inadequate, but often punitive and disempowering; c.f. Hatton
2020). Another recent report (Lown Institute 2019) highlights these same pat-
terns at the state level in California, a hotbed of AI innovation and AI4SG; it
finds that California spends only $0.68 on social services, public health and
environmental protection for each $1.00 spent on health care – and that
these disparities have worsened since 2007. Moreover, ‘with increasing sophis-
tication of medical technology, the American tendency to medicalize patient
concerns is widely recognized as contributing to health care cost escalation’
(Bradley and Taylor 2013, 162), with enormous implications for health
 INTERDISCIPLINARY SCIENCE REVIEWS 103

equity (Woolf et al. 2008), even as algorithms can further embed racial discrimi-
nation in medicine (Obermeyer et al. 2019). Hence, both the financial and the
opportunity costs of AI technologies in health care – often oriented toward
individualized health diagnostics, monitoring and ‘precision’ medical treat-
ments in downstream health care settings – are prone to distract attention
and resources away from root causes of health inequities and upstream social
determinants of health.
That said, public health discourses, emphasizing social determinants of
health, are increasingly being taken up in conjunction with AI-based interven-
tions, particularly in health care settings, sometimes intertwined with precision
medicine (for a fuller critical discussion of ‘precision public health’, c.f. Chowk-
wanyun, Bayer, and Galea 2018, 2019). Advocates point to the potential of AI to
better identify ‘high-risk’ patients, based on social data (e.g. people at risk
because they lack access to basic resources in their daily lives, often long
prior to becoming patients for particular health issues), to incorporate these
social determinants of health data into individual EHRs and clinical
workflows, and to refer patients to social services and monitor the results (Cog-
nizant 2019). Such possibilities, which have opened up in part due to new
‘value-based care’ payment models that emphasize social determinants of
health for the first time (Cognizant 2019, 5; Ostherr 2019), may help to mitigate
inequitable health care outcomes among patients, especially if the inadequacies
of current value-based models vis-à-vis reducing racialized and other health
inequities are addressed (Ojo, Erfani, and Shah 2020).
However, action on social determinants of health at the relatively down-
stream, health care level (e.g. by providers and insurers), including by using
AI to connect people with non-medical social services and to monitor out-
comes, does not address the larger picture of skewed health investments dis-
cussed above (Alberti, Bonham, and Kirch 2013; Maani and Galea 2020a;
Silverstein, Hsu, and Bell 2019). For despite the promise some see in myriad
AI and AI4SG health use cases, ‘Achievement of primary prevention benefits
depends more on social factors than secondary prevention irrespective of the
marginal benefits of artificial intelligence’ (Panch et al. 2019, e13) – upstream
social factors like jobs and shared prosperity that AI may, on balance, be dis-
rupting in ways more harmful than not. As medical and public health scholars
Nason Maani and Sandro Galea write, ‘[T]he notion of addressing social deter-
minants in the context of clinical practice devalues and medicalizes the complex
burden and barriers encountered by those affected by discrimination or
poverty’ (Maani and Galea 2020a, E1). Instead of new data analytics, they
underscore the centrality of distributions of power: ‘Many fundamental deter-
minants of health are far upstream of health care and are deeply rooted in the
distribution of money and power, at local and national levels’ (Maani and Galea
2020a, E1). And as other scholars point out, ‘[F]orce-fitting strategies to address
social determinants of health into traditional models of clinical care risks
104 CHERYL HOLZMEYER

misdirecting limited resources into programs that may ultimately prove ineffi-
cient or ineffective’ (Silverstein, Hsu, and Bell 2019, E2).
In addition, the potential for commercial entities, including private Big Tech
companies, to coopt and attempt to ‘redefine’ – and monetize – public health in
these contexts is deeply troubling, including claims by Facebook representatives
about the promise of using social media and online consumer data to: ‘trans-
form the traditionally held social determinants of health, including education,
income, housing and community, to encompass a more granular tech-
influenced definition, ranging from simple factors, such as numbers of online
friends, to complex social biomarkers, such as timing, frequency, content and
patterns of posts and degree of integration with online communities’
(Abnousi, Rumsfeld, and Krumholz 2019, 247). Again, tech-centrism and
tech-solutionism can distract from seeing the inequalities in power and
resources that drive population health inequities, as well as the relative lack
of accountability of health care, pharmaceutical and technology companies to
the public health field and diverse publics. While this paper focuses on these
issues in a US context, the power of the US tech sector and tech philanthropies
in shaping public health is certainly global in scope (Birn 2014), often perpetu-
ating colonial relations in the process (Amrute 2019).

2) Big data and AI: new vulnerabilities and risks

Second, not only are AI and AI4SG projects prone to misdirect attention and
resources away from root causes of social and environmental justice issues,
including public health; they also enable new forms of discrimination, disad-
vantage and vulnerability resulting from the confluence of emerging big data
streams, machine learning algorithms and increasingly personalized ‘risk’
profiles (Achiume 2020; Figueroa, Frakt, and Jha 2020; Mittelstadt and
Floridi 2016). These new vulnerabilities and sources of discrimination are
primed to multiply in the years ahead, particularly as Big Tech companies
and other profit-oriented entities mine people’s online data to sort and stratify
them (Abnousi, Rumsfeld, and Krumholz 2019), and data are combined and
used in ways people never imagined, much less consented to. Even advocates
of proposed ‘polysocial risk scores’, which would link multiple social data
streams to profile people in health contexts, acknowledge, ‘[T]he risk of false-
positives and false-negatives could be high, especially in the early iterations’
(Figueroa, Frakt, and Jha 2020, E2).
The risks posed by new risk scores and profiling will tend to deepen the chal-
lenges faced by already marginalized communities, potentially worsening racial
and other forms of discrimination through algorithmic decision-making
(Obermeyer et al. 2019; Heaven 2020), not only at the individual level but at
the level of social groups and collectivities, in both the private and public
sectors (Engstrom et al. 2020). As Ferryman and Pitcan quote bioethicist Lisa
 INTERDISCIPLINARY SCIENCE REVIEWS 105

Park as saying, in an interview regarding precision medicine data analytics: ‘[I]f

there are particular health risks that are associated with traditionally underre-
presented, underserved, or discriminated-against populations, there’s an
opportunity or a likelihood that there will be exacerbation of that discrimi-
nation’ (Ferryman and Pitcan 2018, 30). Moreover, Park notes that information
about people’s purported health risks could be used coercively by a range of
public and private actors – for example, to condition people’s access to social
services or insurance coverage. The result will be that less enfranchised
people who are already exposed to extensive surveillance and constraints on
their autonomy will encounter additional vulnerabilities to repressive risk
assessments, controls and sanctions. If automated, through the use of algor-
ithms, these new risk assessments could both magnify existing social systems
of exclusion and oppression while also being more opaque and difficult to chal-
lenge, as ‘Jim Crow practices feed the ‘New Jim Code’ – automated systems that
hide, speed and deepen racial discrimination behind a veneer of technical neu-
trality’ (Benjamin 2019b, 422; also c.f. Benjamin 2019a). These current and
anticipated risks point to the inadequacies of conventional, individually
oriented bioethics frameworks in grappling with emerging technologies (Oba-
sogie and Darnovsky 2018), including as a touchstone for computer and AI
ethics (Floridi et al. 2018). They point to the need for a ‘new biopolitics’ (Oba-
sogie and Darnovsky 2018), or as technology scholar Sareeta Amrute calls for,
‘Radicalize and politicize your ethics – critique, reimagine, practice, repeat’
(Amrute 2019).
Scholars and activists collaborating on the Our Data Bodies Project (https://
www.odbproject.org/) likewise illuminate these politics in their work, analyzing
the new risks and vulnerabilities enabled by big data analytics and AI. Among
the findings of their Reclaiming Our Data report (2018), which draws on inter-
views with residents of Charlotte, Detroit and Los Angeles: (1) ‘Predatory data-
driven systems routinely hold us back and prevent us from meeting basic
needs’, e.g. through the use of credit scores in job and housing applications,
or otherwise subjecting people to extra stigma, scrutiny or punishment by
social institutions – from law enforcement to health care to public services –
as well as to additional uncertainties, depending on how and by whom data
might be used in the future; (2) ‘Data-driven systems unfairly divert us from
resources we are entitled to and need to survive’, e.g. through one-way, asym-
metric data collection by powerful government and corporate actors, even as
people’s basic needs and information that might be helpful to meet them are
neglected, thereby reinforcing existing power relationships; (3) ‘The hurt and
the harms of data-driven systems connect to past systems of discrimination
and exploitation against members of marginalized communities’, e.g. when
people are profiled in terms of race or ethnicity, socio-economic status, age,
gender, sexuality, housing status, criminal record, or some combination; and
(4) ‘Our policies, priorities and visions for the future focus on just models of
106 CHERYL HOLZMEYER

governance, economics and criminal justice, and on guarantees for our human
rights’, e.g. by emphasizing inclusion of all, sharing and empathy, as well as
renewing ‘off the grid’ connections with people and places as a strategy for
self-defense and survival (Petty et al. 2018; also c.f. Lewis et al. 2018).
These insights, informed by deeper recognition and analysis of social inequi-
ties impinging on people’s well-being, including structural racism, bring a vital
corrective to over-hyped prognostications about leveraging big data and AI on
behalf of public health and other social goods. Indeed, they highlight the poten-
tials for automating inequalities and ‘feedback loops of injustice’ (Eubanks
2018), rather than reducing or dismantling them. Such considerations are all
the more important given that many big data and AI applications are developed
in private corporate or start-up settings, beyond democratic governance, and
are deployed for profit (Singer 2017). Companies like Facebook have already
demonstrated negligence in safeguarding people’s personal data, which in
health care contexts could lead to people being denied coverage by an
insurer, employment discrimination or other harms (Singer 2019). These con-
junctures point to the urgency of innovations such as Data for Black Lives’ pro-
posed Public Data Trust (Milner 2018). Moreover, attending to biases in
algorithms and training data is crucial and necessary, but not sufficient, to
achieving social justice, when entire social systems are unjust – and given
that ‘algorithmic fairness’ is never a merely technical, objective matter
(Heaven 2020; Noble 2018).

3) The politics of data and expertise in the computational turn

Third, discussions of big data, AI and public health (among other ‘social good’
domains) often fail to adequately grapple with the politics of data, knowledge
and expertise – in academia and beyond – in the so-called computational
turn. As boyd and Crawford note, ‘Big Data changes the definition of knowl-
edge’, as it refers not only to vast quantities of data and procedures for analyzing
them, but ‘also to a computational turn in thought and research’ (boyd and
Crawford 2012, 665), as in computational social science, wherein new forms
of digital, big data-mediated research (e.g. computerized social network or
social media analysis) are framed as the cutting edge of insights into social
life. This computational bias increasingly shapes the very conceptions of data
and research embedded in nominally interdisciplinary projects and data infra-
structures, perhaps especially those framed around open science, which empha-
sizes digital data and its ‘usability’ (including machine-readability) across fields.
As the International Science Council’s Committee on Data for Science and
Technology (CODATA) describes, it aims to ‘promote global collaboration to
advance Open Science and to improve the availability and usability of data
for all areas of research’, while defining data as, ‘Facts, measurements, record-
ings, records, or observations about the world collected by scientists and others,
 INTERDISCIPLINARY SCIENCE REVIEWS 107

with a minimum of contextual interpretation’ (https://codata.org). That is, even

while ostensibly devoted to data in all areas of research, CODATA’s conception
of data actually presumes a narrow, positivist approach to knowledge and data
that excludes and implicitly devalues interpretive, contextually attuned research
– whether in the humanities, social sciences, community-based participatory
research, or other fields. A synergy among big datasets, machine learning
and open science rhetorics seems likely to further entrench these compu-
tational, positivist biases in research and knowledge. That said, the Research
Data Alliance, organized by international stakeholders with a similar mission
to ‘develop and adopt infrastructure that promotes data-sharing and data-
driven research’ (https://www.rd-alliance.org/), articulates a more reflexive
approach to interdisciplinarity (Lélé and Norgaard 2005), better attuned to dis-
ciplinary specificities and contexts of ‘data’, indicating the wider array of poss-
ible institutions to navigate these issues (e.g. the Platform for Experimental,
Collaborative Ethnography: https://worldpece.org/). Though in many cases,
open sharing of data and research may not be desired by or serve communities
(Albornoz 2018), as a range of activists and scholars, including founders of the
Global Indigenous Data Alliance (https://www.gida-global.org), have been
leaders in establishing.
At the same time, some big data advocates propose that the challenges of tri-
angulating diverse data streams – of intra- and interdisciplinary knowledge
syntheses – might be overcome by algorithms. Regarding big data and public
health, Khoury and Ioannidis assert that, ‘There also must be a means to inte-
grate knowledge that is based on a highly iterative process of interpreting what
we know and don’t know from within and across scientific disciplines’ (Khoury
and Ioannidis 2014, 1054). They then suggest that machine learning algorithms
might be used to such ends, citing the example of the ClinGen project to ‘create
centralized resources of clinically annotated genes to improve interpretation of
genomic variation and optimize the use of genomics in practice’ (Khoury and
Ioannidis 2014). Meanwhile they neglect the broader public health field and the
politics and challenges of interdisciplinarity across fields such as computational
genomics and social epidemiology, or a wider range of social and natural
sciences, from political ecology to social psychology. These fields may be
centred on disparate units and scales of analysis, methodologies, evidentiary
standards, epistemologies, values and more – rendering their respective
bodies of knowledge in tension with one another (or even incommensurable,
from some vantages; c.f. Espeland 1998). These tensions belie efforts to seam-
lessly synthesize and integrate knowledge, as more critical scholarship on inter-
disciplinarity has highlighted (Brown, Morello-Frosch, and Zavestoski 2011;
Lélé and Norgaard 2005; Sarewitz 2004, 2015; Strober 2010). As social epide-
miologists Nancy Krieger and George Davey Smith write, ‘[D]ata never speak
by themselves – either to computer algorithms or to people – and nor do
beliefs about probabilities simply drop from the sky. Active scientific judgment
108 CHERYL HOLZMEYER

is inevitably involved … ’. (Krieger and Smith 2016, 1794). Algorithms claimed

by some to be innovative, neutral and objective as means to ‘integrate knowl-
edge’ and define ‘evidence-based’ practice instead conceal persistent epistemic
politics and hierarchies (Braveman et al. 2011).
More broadly, inclusive and critical knowledge syntheses would not be
limited to the research and data of credentialed experts, from whatever disci-
plinary fields, or to digital data. Instead, they would incorporate street
science (Corburn 2005), popular epidemiology and lay health expertise
(Brown 1992; Epstein 1995), indigenous ecological knowledge (Anderson
2005), popular education (Onuoha 2018), and other forms of community-
based, embodied, grassroots expertise, including to contest and reframe con-
ventional approaches to data, which may otherwise institutionalize racial
inequities and other biases (Hawn Nelson et al. 2020). In the public health
field, such knowledge syntheses are all the more crucial when they intertwine
with the politics of diagnoses and ‘contested illnesses’, or ‘conditions whose
causes are either unexplained by current medical knowledge or whose pur-
ported environmental explanations are in dispute’ (Brown, Morello-Frosch,
and Zavestoski 2011, 18), from asthma to cancer to Gulf War-related illnesses.
Environmental justice and other health social movements may conduct com-
munity-based research, both independently and in collaboration with academic
researchers, to advance new bodies of knowledge that synthesize diverse forms
of expertise in confronting these illnesses (Corburn 2005). In the process, health
social movements aim to transform conventional hierarchies of credibility
while also directing attention to – and advocating for change around – environ-
mental and other dimensions of disease that are often relatively neglected
(Brown, Morello-Frosch, and Zavestoski 2011).
Again, the politics of expertise and science are bound up with the politics of
different kinds of interventions and social transformations (Martin 2006), or
lack thereof. While physicians may be concerned with the status of their exper-
tise vis-à-vis AI technologies in clinical settings, as discussed elsewhere in this
issue (Hanemaayer; Burton et al., this issue), expertise pertaining to (or imping-
ing on) health equity extends far beyond clinics; it encompasses the public
health field as well as broader political, economic and social policy-making,
as recognized in ‘Health in All Policies’ frameworks (Rudolph et al. 2013).
Building trust and investment in AI at one level, such as in clinical care, may
mean further entrenching hierarchies and inequalities that diminish health
equity overall (Pūras 2020), even if some patients receive high-quality care.
This wide spectrum of decision-making, knowledge and values begs the ques-
tion of which and whose ‘social good’ – or conception of ‘optimization’ – AI
and AI4SG projects purport to advance. In any given big data or AI project,
there will be a politics of ‘domain expertise’; some people’s values, priorities,
epistemologies, types of evidence, forms of data and diagnostic or prescriptive
frames will predominate – marginalizing or foreclosing other possibilities,
 INTERDISCIPLINARY SCIENCE REVIEWS 109

especially from systemic perspectives. Machine learning algorithms cannot

escape or transcend these contentious, value-laden issues. Indeed, to date,
widely used search engine algorithms have instead been found to reinforce
racism and sexism in people’s everyday digital lives, through the types of ‘dis-
coverability’ they foster and suppress (Noble 2018).

4) New data treadmills, or new commitments to action on existing

research?
Fourth, AI4SG and AI discussions of public health and other social issues often
neglect warrants for action based on existing research. Advocates for leveraging
big data on behalf of health may not even be familiar with voluminous bodies of
existing public health research, much less the politics and policies impinging on
action on that research (Hofrichter and Bhatia 2010), which their efforts are
further shaping. Yet overweening attention to big data and AI projects could
displace the value and transformative potential of current knowledge – from
public health and social science research to local knowledge and environmental
health documentation (e.g. activist reports, oral histories, journalism, photogra-
phy). Big data and AI could also lead to new cycles of (re)validating that
environmental and social variables matter to health, not only genes and individ-
ual biomarkers – fostering new promissory ‘precision’ data horizons (Kuch,
Kearnes, and Gulson 2020), ‘data treadmills’ (Wylie, Shapiro, and Liboiron
2017, 412), the ‘datafication of injustice’ (Benjamin 2019a, 116) and the contin-
ued rationalization of inaction. As Benjamin writes, ‘Demanding more data on
subjects that we already know much about is, in my estimation, a perversion of
knowledge’ (Benjamin 2019a, 116). Yet big data, AI and AI4SG advocates tend
to highlight only the opportunities presented by new data streams, and the sup-
posed inadequacy of current data (Abnousi, Rumsfeld, and Krumholz 2019;
Figueroa, Frakt, and Jha 2020), rather than the myriad unrealized opportunities
for policy-makers to call for action on current bodies of research.
As one example, what if US scientific, health and business leaders were
equally or more enthusiastic about acting on current research pertaining to
healthy early childhood development, including healthy neurological develop-
ment, as they are about precision medicine and genomics? What if AI4SG advo-
cates who are so concerned about the ‘opportunity costs’ of underutilizing AI
(Floridi et al. 2018) reckoned instead with the opportunity costs of perpetual
underinvestment in social determinants of health? What if, instead of heralding
the promise of new big data streams and AI, they resolved to mobilize resources
to act on the findings of past National Academies research syntheses on these
topics, such as From Neurons to Neighborhoods: The Science of Early Childhood
Development (Shonkoff and Phillips 2000) and Vibrant and Healthy Kids:
Aligning Science, Practice, and Policy to Advance Health Equity (DeVoe,
Geller, and Negussie 2019)? These reports highlight the science of Adverse
110 CHERYL HOLZMEYER

Childhood Experiences (ACEs), toxic stress, and the importance of social deter-
minants of health to children’s healthy development. Yet decision-makers at
multiple levels have by no means prioritized or adequately acted on the
findings – resulting in a chasm between knowledge and practice that more
big data and AI will not resolve. Nor do efforts by some precision medicine
advocates (California Precision Medicine Advisory Committee 2019) to incor-
porate ‘social determinants of health’ and ACEs data in relatively downstream,
individualized clinical care adequately address these issues, as discussed
previously.
Instead, as Wallack and Thornburg indicate in their analysis of intervening
upstream for healthy childhood development (Wallack and Thornburg 2016),
there is a need to recognize the centrality of politics and new political for-
mations to innovation for health equity, rather than merely more data and
research. After reiterating an observation from Thomas Pynchon’s Gravity’s
Rainbow that, ‘If they can get you asking the wrong questions, the answers
don’t matter’, they go on to suggest the following ‘right questions’, emphasizing
the political and social contexts that invariably mediate research translation
into practice and everyday life:
‘If any particular geographic area or region were to become the healthiest place in the
world to be pregnant and have a child, what would it look like? … What kinds of pol-
icies would be required to move toward that vision? How can we create a social move-
ment built on this collective vision to force the necessary political will to demand
change? How can existing partnerships be expanded? How can we develop new part-
nerships with new allies to move ahead? What political barriers must we overcome?’
(Wallack and Thornburg 2016, 938).

After witnessing years of inaction on the science of early childhood develop-

ment and social determinants of health, Wallack and Thornburg do not empha-
size the potential for new research and data streams to suddenly catalyze
breakthrough interventions. Rather, they articulate a broader theory of
change (and call for action, in policy and practice), encompassing social and
political contexts of data and sociotechnical systems. In contrast, many
AI4SG advocates and projects lack such broad visions and theories of
change, rooted in existing and emerging social movements; nor do they often
designate grassroots activists and social movement leaders as AI4SG ‘domain
experts,’ thought partners and white paper audiences (e.g. Rolnick, Donti,
and Kaack 2019), even when disavowing a belief in tech-fixes. AI4SG initiatives
therefore often fail to critically discuss technical and political dimensions of
interventions, perpetuating technocratic decision-making, despite voluminous
scholarship in these areas (Asdal, Brenna, and Moser 2007; Ferguson 1994),
including work to develop greater ‘structural competency’ among health pro-
fessionals, ‘re-centring the political elements of disease distribution’ (Neff
2020, 8; Metzl and Hansen 2014). Such analyses are all the more crucial
 INTERDISCIPLINARY SCIENCE REVIEWS 111

when extensive research and data are available to warrant and guide action, yet
confront deep resistance in some quarters, including denials of structural
racism. As another recent analysis put it, ‘[M]any agency solutions and data
initiatives are largely disconnected from this root cause [of structural
racism], and the ‘hunt for more data is [often] a barrier for acting on what
we already know’ (Benjamin 2019a)’ (Hawn Nelson et al. 2020). While more
investment in the public health sector’s data infrastructures could support
broader public health efforts, including in response to the current Covid-19
pandemic, deeper investments in upstream social determinants of health for
all – not just more data collection – are ultimately called for. Again, while AI
and AI4SG projects did not create these issues and inequalities, AI systems
are poised to exacerbate and conceal them in multiple ways, given the vast
inequalities, distributions of power and social injustices of the current world.
These interrelated issues point to common blindspots and cross-cutting
biases in the larger universe of AI4SG initiatives, as discussed further below.

AI for not bad and social transformations for health equity

In sum, the confluence of big data, AI and public health presents multiple chal-
lenges from a social justice vantage, oriented toward health equity for all. These
challenges include (though are not limited to): (1) distracting attention and
resources away from root causes of health inequities and social determinants
of health; (2) enabling new risks and vulnerabilities due to emerging data
streams, especially for already marginalized communities; (3) failing to ade-
quately grapple with the politics of data, knowledge and expertise in the ‘com-
putational turn’ to big data analytics and AI; and (4) fostering new data
treadmills, rather than emphasizing existing warrants for action based on
current bodies of research. AI4SG initiatives often pre-frame systemic social
and environmental problems in tech-centric ways, while suggesting that
addressing such problems hinges on more or better data. They thereby perpe-
tuate incomplete, distorted models of social change that claim to be ‘data-
driven’. In the process, AI4SG initiatives may obscure or ‘ethics wash’ all the
other uses of big data analytics and AI that more routinely serve private inter-
ests and exacerbate social inequalities – including through AI’s ecological foot-
prints, displacement and exploitation of workers, surveillance of publics,
heightened discrimination against already marginalized groups, and inadequate
social benefits from publicly funded R&D. So, however well-intended, a more
appropriate frame and goal than AI4SG might be ‘AI for Not Bad’, as Moore
(2019) suggested, in keeping with the bioethical principle to ‘first, do no harm’.
While AI4SG proponents acknowledge some of these challenges and risks –
particularly problems caused by biased datasets, inaccurate predictions, algo-
rithmic opacity and illegal or unintentional breaches of privacy (Chui et al.
112 CHERYL HOLZMEYER

2018, 35) – their inattention to or blindspots around larger social, political and
economic systems often hamper consideration of the fuller range of social
justice issues at stake (e.g. such as those discussed in the 2018 Reclaiming
Our Data report). The lack of diversity among AI engineers likely contributes
to these disconnects, as technologists develop tools that disproportionately
reflect their own social backgrounds and interests (Ferryman and Winn
2018). AI4SG advocates, whether AI developers or otherwise, are also often
removed from the everyday political challenges and social contexts of less
enfranchised communities, even when their projects emphasize collaborative
community partnerships; meanwhile the people involved with such projects
on the ground may well be unfamiliar with the phrase AI for Good and its
accompanying influencer circuits, from South by Southwest to Davos. Hierar-
chies of power in workplaces developing AI and AI4SG projects, from academia
to corporate settings, may also invalidate or discourage adequate critique by
those closest to such projects (Hatton 2020). Though this article has focused
especially on public health issues, these concerns are relevant to a wider
gamut of social and environmental challenges toward which AI interventions
may be directed.
In particular, projects that tout the promise of new digital big data streams,
yet lack clear theories of change connecting those data with desired outcomes,
can too easily lead all involved to believe that a project will result in meaningful
change, regardless of the prospects for such change. These include AI4SG pro-
jects centred on new types of monitoring and tracking (e.g. of climate change
indicators, health indicators, educational indicators, or other forms of environ-
mental or social surveillance) that neglect possibilities discussed in this paper –
of obfuscation, distraction, the creation of new vulnerabilities, invalidation of
grassroots knowledge, and delay and disregard for existing warrants for
action. In light of these issues, some key questions for those considering
AI4SG projects include:

1) Who is at the table? Where are the grassroots activists and social move-
ments among the ‘domain experts’? Which stakeholders are part of the con-
versation about a potential project? How are these stakeholders
representative, or not, of the array of people active around or affected by
a particular issue?
2) What is ‘the problem’ and its history? What are the root causes? What are
the social, political, economic and cultural contexts surrounding an issue
and potential AI4SG project? Do all stakeholders agree on how to define
a problem, as well as hypothetical interventions? Do proposed solutions
centre technology and technocratic decision-making? Or are social and pol-
itical change centred? What are the trade-offs or tensions between different
 INTERDISCIPLINARY SCIENCE REVIEWS 113

potential interventions and problem-solving strategies (e.g. more or less

tech-centric strategies)?
3) What is the project’s vision or theory of change? What are the intended out-
comes of a potential AI4SG project, and the steps to get there? What would
it mean for a particular issue to be mitigated or ‘solved’? What would
systems change entail? Who, if anyone, would be responsible for acting
on data from the project? What other forces could impinge on desired out-
comes? What resources and processes are available to respond to situations
indicated by the data? How might the data help with community organizing
and power-building? What other means of power-building are available to
address a given issue?
4) What are the politics of knowledge and expertise? How might data gener-
ated by the project amplify, or undermine, other forms of community
knowledge and data (particularly people’s lived experiences and commu-
nity-based research of many kinds)? How might data be contested by
various parties? How does the project build on existing technological infra-
structures and technological access (or not), including equity in data access,
analysis, interpretation and legibility?
5) What could be the unintended consequences of a project? What other out-
comes are possible? How might an intervention address or obscure root
causes? How might it foster new data treadmills? How might it fit into
the broader landscape of problem-solving around a particular issue, includ-
ing resource allocations? Which resources and processes could help
empower those most affected, already marginalized, or newly vulnerable,
including if data are misused?
6) How does the project intersect with racial equity? How do a project’s data
engagements dovetail with guidelines for centring racial equity and con-
fronting structural racism throughout data lifecycles, including as articu-
lated in ‘A Toolkit for Centring Racial Equity Throughout Data
Integration’ (Hawn Nelson et al. 2020), focused on civic data use?
7) How does the project intersect with health equity? How does a project’s
implicit or explicit theory of change intersect with the APHA’s statement
on the path to health equity? Namely: ‘How do we achieve health equity?
We value all people equally. We optimize the conditions in which people
are born, grow, live, work, learn and age. We work with other sectors to
address the factors that influence health, including employment, housing,
education, health care, public safety and food access. We name racism as
a force in determining how these social determinants are distributed’.
8) Why should or shouldn’t the project be pursued? How could this assess-
ment change? Considering the issues above and any other relevant con-
cerns, what could be evidence of a potential project’s (in)efficacy,
unintended social or political consequences, or other indicators that a
114 CHERYL HOLZMEYER

project should or should not be pursued? How might these indicators be

revisited periodically?

These lines of inquiry, and many more that could be generated by stake-
holders addressing specific issues, are crucial to surfacing potential AI4SG pro-
jects’ values and visions of change (or lack thereof), including their conceptions
of and approaches to whatever ‘good’ they seek to advance. They are intended
to problematize the technological solutionism and tech-fix frameworks that
often underpin such projects, addressing but going beyond the rules of
thumb for technological fixes developed by Sarewitz and Nelson (2008), as
well as lists of AI4SG ‘best practices’ that make ritualistic disclaimers that AI
is not a ‘silver bullet’ yet fail to seriously incorporate that point into their
overall analysis (Floridi et al. 2020, 1773). By shifting attention toward root
causes of social and environmental problems (e.g. in systems of oppression
and exploitation); new risks and vulnerabilities; the politics of data, knowledge
and expertise; and the pitfalls of perpetual data treadmills, these questions seek
to provoke deeper reckoning with not only the risks and benefits of potential
projects, but alternative paradigms of problem-solving entirely, oriented
toward challenging existing systems of power.
Such questions can also help to evaluate and highlight AI4SG projects that
may indeed be quite helpful to address particular issues, if developed in collab-
oration with community partners. Such projects include ActiveRemediation, a
machine learning model designed by university-based researchers to predict the
location and aid in the removal of water service lines containing lead in Flint,
MI, in the absence of adequate public records (Abernethy et al. 2018; Chui et al.
2018, 26). They also include Talking Points, a natural language processing plat-
form developed in a non-profit context that translates multilingual text mess-
ages among teachers, parents and students, to ‘driv[e] student success in low-
income, diverse areas through building strong partnerships across families,
schools and communities’ (https://talkingpts.org/about-us/). These projects,
articulating with environmental and social determinants of health, feature
readily useable data and clear users in public municipal and educational con-
texts, directly helping to address specific community challenges while building
on existing technological infrastructures, without fostering new data treadmills.
That said, in these cases, too, broader systemic interventions remain crucial to
achieving environmental justice and health equity, in Flint and beyond, and to
the realization of Talking Points’ larger stated mission (i.e. to ‘driv[e] student
success in low-income, diverse areas’). By extension, project funders, including
deep-pocketed Big Tech companies and foundations, should not be perceived
as adequately ‘doing good’ simply by assisting with such efforts, however laud-
able. Rather, they should be held accountable by policy-makers at multiple
levels and across sectors (Rudolph et al. 2013) to ensure that all of the political,
 INTERDISCIPLINARY SCIENCE REVIEWS 115

economic and social relations they foster are consistent with advancing healthy
childhood development and social determinants of health for all.
At present, in the context of COVID-19, US underinvestment in the public
health sector, robust social safety nets and upstream social determinants of
health – including workers’ rights and protections – has translated into diagno-
sis and death rates that are high overall as well as deeply racialized and corre-
lated with income, amplifying preexisting health inequities (Maani and Galea
2020b; Serkez 2020). As Anthony Iton writes, underscoring these preventable
vulnerabilities at the intersections of race, class and place:
‘COVID-19 is reminding us that in the United States, when it comes to your health,
your zip code is more important than your genetic code. Our country manufactures
social vulnerability through policy violence. Policy violence is the intentional absence
of protective policy in the face of abject need. Policy violence leaves large segments of
our society experiencing constant daily stress as they try to navigate a healthy life
without health insurance, decent housing, affordable childcare, paid sick leave, or
quality education … . The foundation of American policy violence is racism.
Scratch the surface of virtually every failed effort at creating universal policies in
this country and you’ll find thinly veiled racism at the root’ (Iton 2020).

So from a wider-view lens, instead of foregrounding large-scale data, AI, or

other technologies in discussions of innovation and public health, there is a
need to centre larger-scale social transformations to address the multifaceted
policy violence Iton describes. In addition, Iton and other US social justice
leaders have called for new social compacts (Iton 2017), emphasizing interde-
pendency, collective care and decision-makers’ accountability ‘to serve the
well-being of people, including the right to health care, food, education, and
shelter’ (https://newsocialcompact.org/). Along these lines, contemporary
social movements have articulated incisive, multifaceted policy platforms
centred on racial, economic and climate justice that could frame AI ethics con-
versations (Movement for Black Lives 2020; National Domestic Workers Alli-
ance 2020; Poor People’s Campaign 2020; Sunrise Movement 2020), rather than
discussing AI4SG and the UN SDGs apart from contexts of pervasive neoliber-
alism and structural racism (e.g. as in Hager et al. 2017; Floridi et al.
2018, 2020), in the USA and beyond. Scholars proposing a decolonial critical
approach to AI have likewise called for renewing affective and political commu-
nities as one key tactic (Mohamed, Png, and Isaac 2020).
Centring these social transformations would entail contesting current politi-
cal and economic structures and the forms of ‘innovation’ they produce. In the
USA, such transformations are all the more justified given that the public sector
often takes on high levels of risk in funding the development of new technologies,
yet the public return on these investments tends to be negligible, compared with
the rewards captured by the private sector. This results in a skewed ‘risk-reward
nexus’, according to economists Williams Lazonick and Mariana Mazzucato
(Lazonick and Mazzucato 2013), and, ultimately, a ‘parasitic innovation
116 CHERYL HOLZMEYER

ecosystem’ that undermines both innovation and shared prosperity (Mazzucato

2013, 2018). These dynamics reflect deeply rooted biases of conventional US
R&D agendas and science and technology policy-making, which disproportio-
nately serve the interests of the most privileged while mostly neglecting social
justice and equity issues, in biomedical research and other fields (Bozeman
2020; Woodhouse and Sarewitz 2007). They also reflect tendencies to seek down-
stream, technological solutions to social and environmental problems rather
than preventing or significantly mitigating them upstream; yet in the current
era of apparently accelerating ecological and social crises, there is a need for
transformed paradigms on multiple fronts (e.g. as articulated in the recent
report: Resilience Before Disaster: The Need to Build Equitable, Community-
Driven Social Infrastructure, Lou et al. 2020; also Interlandi 2020).
On that note, rather than trying to extend technological infrastructures into
new domains, computer and data scientists seeking to do good could instead
collaborate with social and environmental justice activists to analyze the
values and biases embedded in existing data infrastructures and algorithms,
in settings from health care (Chen, Szolovits, and Ghassemi 2019; Obermeyer
et al. 2019) to law and policing (Heaven 2020), as some are indeed doing.
Or, taking a step back, AI4SG advocates could lobby companies for greater
transparency and access to algorithms, which are often proprietary, to enable
such scrutiny in the first place. Looking to the future of AI, AI4SG advocates
could support efforts to diversify AI R&D – such as Black in AI (https://
blackinai.github.io/) and Women in Machine Learning (https://
wimlworkshop.org/) – as well as STEM fields broadly. They could support
calls for substantive public accountability and public governance of big data,
AI, biomedicine and the tech sector in general, including through public data
trusts, in solidarity with diverse social movement activists (Bernhardt 2017;
Milner 2018) and organizations such as the Algorithmic Justice League
(https://www.ajlunited.org/), Coalition for Critical Technology (https://
forcriticaltech.github.io/), Data for Black Lives (https://d4bl.org/), Data &
Society (https://datasociety.net/), Design Justice Network (https://
designjustice.org/), Science for the People (https://scienceforthepeople.org/),
Structural Competency (https://structuralcompetency.org), the Tech Workers
Coalition (https://techworkerscoalition.org/) and many others concerned
with social justice and existing and emerging technologies.
Computer and data scientists could also support health equity advocacy and
action on existing public health research, through environmental and social
justice organizations as well as the formal public health field (and organizations
spanning these sectors, such as Public Health Awakened: https://
publichealthawakened.org/ and the Spirit of 1848 caucus of the APHA:
http://spiritof1848.org/). Relatedly, AI4SG advocates could support enforce-
ment of existing laws and public rights to benefit from science and technology,
such as Bayh-Dole ‘march in’ rights to secure public access to publicly funded
 INTERDISCIPLINARY SCIENCE REVIEWS 117

health treatments (UCL 2018) as well as antitrust laws, alongside groups like the
Economic Security Project (https://www.economicsecurityproject.org). They
could advocate for more democratic ownership and governance of ‘sharing
economy’ platforms, such as Airbnb and Uber, with organizations contributing
to the platform cooperativism movement (e.g. https://platform.coop/ and
https://sassafras.coop). They could also advocate for innovative new insti-
tutions, such as investing profits from publicly funded technologies in public
tech dividends, to support publics’ access to upstream social determinants of
health. As sociologists Manuel Pastor, Chris Benner, and colleagues have
written, this would be a way to ‘more directly link returns to the risk the
public sector absorbs in these new innovations … Such a ‘technology dividend’
could support a universal basic income fund, which would mitigate risk for
those working through the vagaries of employment shifts engendered by inno-
vation and technological change’ (Pastor et al. 2018, 34).
Beyond technology dividends, however, sociologist and labour scholar
Annette Bernhardt underscores diverse publics’ right to shape technology
development from the outset, from technologies impinging on work and
labour to algorithms affecting lending, hiring and sentencing decisions (Bern-
hardt 2017). She outlines strategies ranging from the mitigation of technologi-
cal effects, to collective bargaining, to publics having ‘a seat at the table when
decisions are made over which technologies are developed in the first place,
and in pursuit of which goals’, citing Germany’s collective bargaining and
‘social partners system’ in which the government ‘actively collaborat[es] with
employers and labor to make its manufacturing sector a leader in technology
and preserve[s] a role for workers’ (Bernhardt 2017). Such multi-stakeholder
governance possibilities, reinforced by regulatory enforcement, are far
removed from the USA’s current industry-led, self-regulating ‘Partnership on
AI to Benefit People and Society’. They are also not among the international
comparisons most frequently cited in US policy-makers’ discussions of AI.
This paper just begins to scratch the surface of many of these issues. Yet ulti-
mately, abundant data suggest that new arrangements and social transform-
ations are urgently needed, not only for greater public benefit from and
democratic governance of technologies, but for democracy – with digital and
analog worlds oriented toward social justice and health equity for all.

Acknowledgments
The author is grateful to Shunryu Colin Garvey and the anonymous reviewers, who all con-
tributed extremely helpful, generous feedback on this paper.

Disclosure statement
No potential conflict of interest was reported by the author(s).
118 CHERYL HOLZMEYER

Notes on contributor
Cheryl Holzmeyer is a sociologist who completed her Ph.D. at UC-Berkeley. She is currently
a Research Fellow affiliated with the Institute for Social Transformation at UC-Santa Cruz.
Her research focuses on the intersections of science, technology, social justice, and health
equity.

References
Abernethy, J., A. Chojnacki, A. Farahi, E. Schwartz, and J. Webb. 2018. “ActiveRemediation:
The Search for Lead Pipes in Flint, Michigan,” ArXiv [preprint]. https://arxiv.org/abs/
1806.10692.
Abnousi, F., J. Rumsfeld, and H. Krumholz. 2019. “Social Determinants of Health in the
Digital Age: Determining the Source Code for Nurture.” Journal of the American
Medical Association 321 (3): 247–248.
Acemoglu, D., A. Manera, and P. Restrep. 2020. “Does the U.S. Tax Code Favor
Automation?” Brookings Institution Spring Conference Draft. https://www.brookings.
edu/wp-content/uploads/2020/03/Acemoglu-et-al-Conference-Draft.pdf.
Achiume, E. T. 2020. “Racial Discrimination and Emerging Digital Technologies: A Human
Rights Analysis,” Report of the Special Rapporteur on contemporary forms of racism,
racial discrimination, xenophobia and related intolerance. United Nations Office of the
High Commissioner for Human Rights. https://undocs.org/en/A/HRC/44/57.
Adams, R. 2021. “Decolonising Artificial Intelligence: A Theoretical Critique.”
Interdisciplinary Science Reviews.
Alberti, P., A. Bonham, and D. Kirch. 2013. “Making Equity a Value in Value-Based Health
Care.” Academic Medicine 88 (11): 1619–1623.
Albornoz, D. 2018. “Reimagining Open Science Through a Feminist Lens,” Medium.
https://medium.com/@denalbz/reimagining-open-science-through-a-feminist-lens-
546f3d10fa65.
American Public Health Association. “Health Equity”. https://www.apha.org/topics-and-
issues/health-equity.
Amrute, S. 2019. “Tech Colonialism Today,” EPIC2019 Keynote Address. https://points.
datasociety.net/tech-colonialism-today-9633a9cb00ad.
Anderson, M. K. 2005. Tending the Wild: Native American Knowledge and the Management
of California’s Natural Resources. Berkeley, CA: University of California Press.
Asdal, K., B. Brenna, and I. Moser, eds. 2007. Technoscience: The Politics of Interventions.
Oslo, Norway: Oslo Academic Press. Unipub Norway.
Bassett, M. 2015. “#BlackLivesMatter — A Challenge to the Medical and Public Health
Communities.” New England Journal of Medicine 372 (12): 1085–1087.
Bayer, R., and S. Galea. 2015. “Public Health in the Precision-Medicine Era.” The New
England Journal of Medicine 373 (6): 499–501.
Beckfield, J. 2018. Political Sociology and the People’s Health. New York, NY: Oxford
University Press.
Benjamin, R. 2019a. Race After Technology: Abolitionist Tools for the New Jim Code.
Cambridge, UK: Polity Press.
Benjamin, R. 2019b. “Assessing Risk, Automating Racism.” Science 366 (6464): 421–422.
Bernhardt, A. 2017. “Expanding the Goal of Innovation,” in Forum on “Basic Income in a
Just Society,” Boston Review, Spring 2017. http://bostonreview.net/forum/basic-income-
just-society/annette-bernhardt-expanding-goal-innovation.
 INTERDISCIPLINARY SCIENCE REVIEWS 119

Birn, A. 2014. “Philanthrocapitalism, Past and Present: The Rockefeller Foundation, the
Gates Foundation, and the Setting(s) of the International/Global Health Agenda.”
Hypothesis 12 (1): 1–27.
boyd, d., and K. Crawford. 2012. “CRITICAL QUESTIONS FOR BIG DATA: Provocations
for a Cultural, Technological, and Scholarly Phenomenon.” Information, Communication
& Society 15 (5): 662–679.
Boyd, R., E. Lindo, L. Weeks, and M. McLemore. 2020. “On Racism: A New Standard for
Publishing On Racial Health Inequities,” Health Affairs Blog. https://www.healthaffairs.
org/do/10.1377/hblog20200630.939347/full/.
Bozeman, B. 2020. “Public Value Science.” Issues in Science and Technology 36 (4): 34–41.
Bradley, E., and L. Taylor. 2013. The American Healthcare Paradox: Why Spending More is
Getting Us Less. New York: Public Affairs Books.
Brandt, A., and M. Gardner. 2000. “Antagonism and Accommodation: Interpreting the
Relationship Between Public Health and Medicine in the United States During the
20th Century.” American Journal of Public Health 90 (5): 707–715.
Braveman, P., S. Egerter, S. Woolf, and J. Marks. 2011. “When Do We Know Enough to
Recommend Action on the Social Determinants of Health?” American Jounral of
Preventive Medicine 40 (1, Suppl. 1): S58–66.
Braveman, P., and L. Gottlieb. 2014. “The Social Determinants of Health: It’s Time to
Consider the Causes of the Causes.” Public Health Reports 129 (Suppl 2): 19–31.
Brown, P. 1992. “Popular Epidemiology and Toxic Waste Contamination: Lay and
Professional Ways of Knowing.” Journal of Health and Social Behavior 33 (3): 267–81.
Brown, P., R. Morello-Frosch, and S. Zavestoski. 2011. Contested Illnesses: Citizens, Science,
and Health Social Movements. Berkeley, CA: University of California Press.
Burton, S. D., T. Mahfoud, C. Aicardi, and N. Rose. 2021. “Clinical Translation of
Computational Brain Models: Understanding the Salience of Trust in Clinician-
Researcher Relationships.” Interdisciplinary Science Reviews.
California Department of Public Health (CDPH), Office of Health Equity. 2015. “Portrait of
Promise: The California Statewide Plan to Promote Health and Mental Health Equity.”
Sacramento, CA: Office of Health Equity, CDPH. https://www.cdph.ca.gov/Programs/
OHE/CDPH%20Document%20Library/Accessible-CDPH_OHE_Disparity_Report_
Final%20(2).pdf.
California Precision Medicine Advisory Committee. 2019. “Precision Medicine: An Action
Plan for California”. https://opr.ca.gov/docs/20190107-Precision_Medicine_An_Action_
Plan_for_California.pdf.
Cath, C., S. Wachter, B. Mittelstadt, M. Taddeo, and L. Floridi. 2017. “Artificial Intelligence
and the ‘Good Society’: the US, EU, and UK Approach.” Science and Engineering Ethics 24
(2): 505–528.
Chadha, N., B. Lim, M. Kane, and B. Rowland. 2020. Toward the Abolition of Biological Race
in Medicine: Transforming Clinical Education, Research, and Practice, Othering and
Belonging Institute at UC-Berkeley and Institute for Healing and Justice in Medicine.
https://www.instituteforhealingandjustice.org/download-the-report-here.
Chen, I., P. Szolovits, and M. Ghassemi. 2019. “Can AI Help Reduce Disparities in General
Medical and Mental Health Care?” AMA Journal of Ethics 21 (2): E167–179.
Chowkwanyun, M., R. Bayer, and S. Galea. 2018. “‘Precision’ Public Health – Between
Novelty and Hype.” The New England Journal of Medicine 379 (15): 1398–1400.
Chowkwanyun, M., R. Bayer, and S. Galea. 2019. “Precision Public Health: Pitfalls and
Promises.” The Lancet 393 (10183): 1801.
Chui, M., M. Harryson, J. Manyika, R. Roberts, R. Chung, A. van Heteren, and P. Nel. 2018.
“Notes from the AI Frontier: Applying AI for Social Good,” McKinsey Global Institute,
120 CHERYL HOLZMEYER

McKinsey & Company. https://www.mckinsey.com/featured-insights/artificial-

intelligence/applying-artificial-intelligence-for-social-good.
Coburn, D. 2010. “Beyond the Income Inequality Hypothesis: Class, Neo-Liberalism, and
Health Inequalities.” In Tackling Health Inequities Through Public Health Practice:
Theory to Action, 2nd Edition; A Project of the National Association of City and County
Health Officials, edited by R. Hofrichter, and R. Bhatia, 196–224. New York: Oxford
University Press.
Cognizant. 2019. “The Social Determinants of Health: Applying AI & Machine Learning to
Achieve Whole Person Care”. https://www.cognizant.com/whitepapers/the-social-
determinants-of-health-applying-ai-and-machine-learning-to-achieve-whole-person-
care-codex4379.pdf.
Comstock, J. 2018. “SkinVision gets $7.6M to continue expanding skin cancer app,” Mobile
Health News. https://www.mobihealthnews.com/content/skinvision-gets-76m-continue-
expanding-skin-cancer-app.
Cooper, R., and N. Paneth. 2020a. “Will Precision Medicine Lead to a Healthier
Population?” Issues in Science and Technology 36 (2): 64–71.
Cooper, R., and N. Paneth. 2020b. “Precision medicine: course correction urgently needed,”
STAT News, 3 March 2020. https://www.statnews.com/2020/03/03/precision-medicine-
course-correction-urgently-needed/.
Corburn, J. 2005. Street Science: Community Knowledge and Environmental Health Justice.
Cambridge, MA: The MIT Press.
DeVoe, J., A. Geller, and Y. Negussie. 2019. Vibrant and Healthy Kids: Aligning Science,
Practice, and Policy to Advance Health Equity. Washington, D.C.: National Academy
Press.
Engstrom, D., D. Ho, C. Sharkey, and M. Cuéllar. 2020. Government by Algorithm: Artificial
Intelligence in Federal Administrative Agencies, a report submitted to the Administrative
Conference of the United States. https://www-cdn.law.stanford.edu/wp-content/uploads/
2020/02/ACUS-AI-Report.pdf.
Epstein, S. 1995. “The Construction of Lay Expertise: AIDS Activism and the Forging of
Credibility in the Reform of Clinical Trials.” Science, Technology, & Human Values 20
(4): 408–437.
Espeland, W. 1998. The Struggle for Water: Politics, Rationality, and Identity in the American
Southwest. Chicago, IL: University of Chicago Press.
Eubanks, V. 2018. Automating Inequality: How High-Tech Tools Profile, Police, and Punish
the Poor. New York: St. Martin’s Press.
Fairchild, A., D. Rosner, J. Colgrove, R. Bayer, and L. Fried. 2010. “The EXODUS of Public
Health: What History Can Tell Us About the Future.” American Journal of Public Health
100 (1): 54–63.
Ferguson, J. 1994. The Anti-Politics Machine: ‘Development’ and Bureaucratic Power in
Lesotho. Minneapolis, MN: University of Minnesota Press.
Ferryman, K., and M. Pitcan. 2018. Fairness in Precision Medicine. New York: Data &
Society. https://datasociety.net/library/fairness-in-precision-medicine/.
Ferryman, K., and R. Winn. 2018. “Artificial Intelligence Can Entrench Disparities – Here’s
What We Must Do,” The Cancer Letter. https://cancerletter.com/articles/20181116_1/.
Figueroa, J., A. Frakt, and A. Jha. 2020. “Addressing Social Determinants of Health: Time for
a Polysocial Risk Score.” Journal of the American Medical Association 323 (16): 1553–
1554.
Floridi, L., J. Cowls, M. Beltrametti, R. Chatila, P. Chazerand, V. Dignum, C. Luetge, et al.
2018. “AI4People – An Ethical Framework for a Good AI Society: Opportunities, Risks,
Principles, and Recommendations.” Minds and Machines 28 (4): 689–707.
 INTERDISCIPLINARY SCIENCE REVIEWS 121

Floridi, L., J. Cowls, T. King, and M. Taddeo. 2020. “How to Design AI for Social Good:
Seven Essential Factors.” Science and Engineering Ethics 26: 1771–1796.
Fortun, M. 2008. Promising Genomics: Iceland and deCODE Genetics in a World of
Speculation. Berkeley, CA: University of California Press.
Funk, J. 2019. “What’s Behind Technological Hype?” Issues in Science and Technology 36 (1):
36–42.
Galea, S. 2019. Well: What We Need to Talk About When We Talk About Health. Oxford,
UK: Oxford University Press.
Garvey, C. 2018. “Interview with Colin Garvey, Rensselaer Polytechnic Institute. Artificial
Intelligence and Systems Medicine Convergence.” OMICS: A Journal of Integrative
Biology 22 (2): 130–132.
Google AI Impact Challenge. 2019. “Accelerating social good with artificial intelligence:
Insights from the Google AI Impact Challenge”. https://services.google.com/fh/files/
misc/accelerating_social_good_with_artificial_intelligence_google_ai_impact_challenge.
pdf.
Gray, M., and S. Suri. 2019. Ghostwork: How to Stop Silicon Valley from Building a New
Global Underclass. Boston, MA: Houghton Mifflin Harcourt.
Hager, G., A. Drobnis, F. Fang, R. Ghani, A. Greenwald, T. Lyons, D. Parkes, et al. 2017.
Artificial Intelligence for Social Good. Washington, DC: Computing Community
Consortium. https://arxiv.org/ftp/arxiv/papers/1901/1901.05406.pdf.
Hanemaayer, A. 2021. “Don’t Touch My Stuff: Historicizing Resistance to AI and
Algorithmic Computer Technology in Medicine.” Interdisciplinary Science Reviews.
Harvard Heart Letter. 2019. “Smartphone Apps for Managing Heart Disease”. https://www.
health.harvard.edu/heart-health/smartphone-apps-for-managing-heart-disease.
Hatton, E. 2020. Coerced: Work Under Threat of Punishment. Berkeley, CA: University of
California Press.
Hawn Nelson, A., D. Jenkins, S. Zanti, M. Katz, E. Berkowitz, T. Burnett, D. Culhane 2020. A
Toolkit for Centering Racial Equity Throughout Data Integration, Actionable Intelligence
for Social Policy, University of Pennsylvania. https://www.aisp.upenn.edu/equity-toolkit/.
Heaven, W. 2020. “Predictive policing algorithms are racist. They need to be dismantled,”
MIT Technology Review. https://www.technologyreview.com/2020/07/17/1005396/
predictive-policing-algorithms-racist-dismantled-machine-learning-bias-criminal-
justice/.
Ho, D., S. Quake, E. McCabe, W. J. Chng, E. Chow, X. Ding, B. Gelb, et al. 2020. “Enabling
Technologies for Personalized and Precision Medicine.” Trends in Biotechnology 38 (5):
497–518.
Hofrichter, R., and R. Bhatia. 2010. Tackling Health Inequities Through Public Health
Practice: Theory to Action, 2nd Edition; A Project of the National Association of City
and County Health Officials. New York: Oxford University Press.
Holzmeyer, C. 2017. “Investigating STEM – Health Equity as Touchstone for the Future,”
Boom California. https://boomcalifornia.com/2017/09/06/investigating-stem-health-
equity-as-touchstone-for-the-future/.
Holzmeyer, C. 2018a. “Open Science Initiatives: Challenges for Public Health Promotion.”
Health Promotion International 34 (3): 624–633.
Holzmeyer, C. 2018b. “The Crockett-Rodeo Real-Time Health Monitoring Study: Summary
of Findings,” 27 June 2018. https://drive.google.com/file/d/1QCCZlpRVDbCcsDpA__
3LMb1Z_M9SjIY8/view.
Holzmeyer, C. 2018c. “Wider Worlds of Research for Health Equity: Public Health NGOs as
Stakeholders in Open Access Ecosystems.” The Journal of Community Informatics 14 (2–
3): 1–16.
122 CHERYL HOLZMEYER

Huesemann, M., and J. Huesemann. 2011. Techno-Fix: Why Technology Won’t Save Us or
the Environment. BC, Canada: New Society Publishers.
Interlandi, J. 2020. “The U.S. Approach to Public Health: Neglect, Panic, Repeat,” The
New York Times, 9 April 2020. https://www.nytimes.com/2020/04/09/opinion/sunday/
coronavirus-public-health-system-us.html.
Iton, A. 2010. “The Ethics of the Medical Model in Addressing the Root Causes of Health
Disparities in Local Public Health Practice.” In Tackling Health Inequities Through
Public Health Practice: Theory to Action, 2nd Edition; A Project of the National
Association of City and County Health Officials, edited by R. Hofrichter, and R. Bhatia,
509–516. New York: Oxford University Press.
Iton, A. 2017. “Rebuilding Our Social Compact,” Building Healthy Communities Blog, The
California Endowment. https://www.calendow.org/rebuilding-social-compact/.
Iton, A. 2020. “‘Policy violence’ against people of color increases COVID-19’s deadly toll,”
The Sacramento Bee. https://www.sacbee.com/opinion/california-forum/
article242010461.html.
Jobin, A., M. Ienca, and E. Vayena. 2019. “The Global Landscape of AI Ethics Guidelines.”
Nature Machine Intelligence 1 (2019): 389–399.
Khoury, M., and J. Ioannidis. 2014. “Big Data Meets Public Health.” Science 346 (6213):
1054–1055.
Krieger, N. 2011. Epidemiology and the People’s Health: Theory and Context. New York:
Oxford University Press.
Krieger, N. 2020. “ENOUGH: COVID-19, Structural Racism, Police Brutality, Plutocracy,
Climate Change—and Time for Health Justice, Democratic Governance, and an
Equitable, Sustainable Future.” American Journal of Public Health 110 (11): 1620–1623.
Krieger, N., and E. Birn. 1998. “A Vision of Social Justice as the Foundation of Public Health:
Commemorating 150 Years of the Spirit of 1848.” American Journal of Public Health 88
(11): 1603–1606.
Krieger, N., and G. D. Smith. 2016. “The Tale Wagged by the DAG: Broadening the Scope of
Causal Inference and Explanation for Epidemiology.” International Journal of
Epidemiology 45 (6): 1787–1808.
Kristof, N. 2020. “‘We’re No. 28! And Dropping!’” The New York Times, 9 September 2020.
https://www.nytimes.com/2020/09/09/opinion/united-states-social-progress.html.
Kuch, D., M. Kearnes, and K. Gulson. 2020. “The Promise of Precision: Datafication in
Medicine, Agriculture and Education.” Policy Studies 41 (5): 527–546.
Latonero, M. 2019. “AI For Good Is Often Bad,” Wired. https://www.wired.com/story/
opinion-ai-for-good-is-often-bad/.
Lazonick, W., and M. Mazzucato. 2013. “The Risk-Reward Nexus in the Innovation-
Inequality Relationship: Who Takes the Risks? Who Gets the Rewards?.” Industrial
and Corporate Change 22 (4): 1093–1128.
Ledford, H. 2020. “Computing Humanity: How Data from Facebook, Twitter and Other
Sources are Revolutionizing Social Science.” Nature 582: 328–330.
Lewis, T., S. P. Gangadharan, M. Saba, and T. Petty. 2018. Digital Defense Playbook:
Community Power Tools for Reclaiming Data. Detroit, MI: Our Data Bodies Project.
https://www.odbproject.org/tools/.
Lélé, S., and R. Norgaard. 2005. “Practicing Interdisciplinarity.” BioScience 55 (11): 967–975.
Lou, Z., A. Raval, M. Young, and S. Appel. 2020. Resilience Before Disaster: The Need to Build
Equitable, Community-Driven Social Infrastructure, Asian Pacific Environmental
Network (APEN), Service Employees International Union (SEIU) California, and
BlueGreen Alliance. https://apen4ej.org/resilience-before-disaster/.
 INTERDISCIPLINARY SCIENCE REVIEWS 123

Lown Institute. 2019. California’s Health Care Paradox: Too much health care spending may
lead to poor community health. https://lowninstitute.org/reports/californias-health-care-
paradox-2/.
Maani, N., and S. Galea. 2020a. “The Role of Physicians in Addressing Social Determinants
of Health.” Journal of the American Medical Association 323 (16): 1551–1552.
Maani, N., and S. Galea. 2020b. “COVID-19 and Underinvestment in the Health of the U.S.
Population.” Milbank Quarterly 98 (2): 239–249.
Martin, B. 2006. “Strategies for Alternative Science.” In The New Political Sociology of
Science: Institutions, Networks, and Power, edited by S. Frickel, and K. Moore, 272–
298. Madison, WI: University of Wisconsin Press.
Mazzucato, M. 2013. The Entrepreneurial State: Debunking Public vs. Private Sector Myths.
London, UK: Anthem Press.
Mazzucato, M. 2018. The Value of Everything: Making and Taking in the Global Economy.
UK: Allen Lane.
Metzl, J., and H. Hansen. 2014. “Structural Competency: Theorizing a New Medical
Engagement With Stigma and Inequality.” Social Science & Medicine 103 (1): 126–133.
Milner, Y. 2018. “An Open Letter to Facebook from the Data for Black Lives Movement,”
Medium. https://medium.com/@YESHICAN/an-open-letter-to-facebook-from-the-
data-for-black-lives-movement-81e693c6b46c.
Milner, Y. 2020. “We Will Not Allow the Weaponization of COVID-19 Data,” Medium.
https://medium.com/@YESHICAN/we-will-not-allow-the-weaponization-of-covid-19-
data-e775d31991c.
Mittelstadt, B., and L. Floridi. 2016. “The Ethics of Big Data: Current and Foreseeable Issues
in Biomedical Contexts.” Science and Engineering Ethics 22 (2): 303–341.
Mohamed, S., M. Png, and W. Isaac. 2020. “Decolonial AI: Decolonial Theory as
Sociotechnical Foresight in Artificial Intelligence.” Philosophy & Technology 405: 1–28.
Moore, J. 2019. “AI for Not Bad.” Frontiers in Big Data 2 (32): 1–7.
Movement for Black Lives. 2020. “Vision for Black Lives: 2020 Policy Platform”. https://
m4bl.org/policy-platforms/.
National Domestic Workers Alliance. 2020. “Summary of the National Domestic Workers
Bill of Rights”. https://www.domesticworkers.org/sites/all/themes/NDWA2017/images/
LearnMore.pdf.
Neff, J. 2020. “Structural Competency for Healthcare Providers,” unpublished manuscript,
shared with the Structural Competency Working Group (https://www.structcomp.org/).
Noble, S. 2018. Algorithms of Oppression: How Search Engines Reinforce Racism. New York:
New York University Press.
Nuffield Council on Bioethics. 2018. “Bioethics Briefing Note: Artificial intelligence (AI) in
healthcare and research”. https://www.nuffieldbioethics.org/publications/ai-in-
healthcare-and-research.
Obasogie, O., and M. Darnovsky. 2018. Beyond Bioethics: Toward a New Biopolitics.
Berkeley, CA: University of California Press.
Obermeyer, Z., B. Powers, C. Vogeli, and S. Mullainathan. 2019. “Dissecting Racial Bias in
an Algorithm Used to Manage the Health of Patients.” Science 366 (6464): 447–453.
Ojo, A., P. Erfani, and N. Shah. 2020. “Value-Based Health Care Must Value Black Lives,”
Health Affairs Blog. https://www.healthaffairs.org/do/10.1377/hblog20200831.419320/full/.
Onuoha, M. and Mother Cyborg (Diana Nucera). 2018. A People’s Guide to AI. Detroit, MI:
Allied Media Projects. https://www.alliedmedia.org/peoples-ai.
Ostherr, K. 2019. “You Don’t Want Facebook Involved With Your Health Care.” Slate.
https://slate.com/technology/2019/09/social-determinants-health-facebook-google.html.
124 CHERYL HOLZMEYER

Panch, T., J. Pearson-Stuttard, F. Greaves, and R. Atun. 2019. “Artificial Intelligence:

Opportunities and Risks for Public Health.” The Lancet 1 (1): e13–e14.
Pastor, M., C. Benner, A. Sharma, E. Muña, S. Rosenquist, and V. Carter. 2018. From
Resistance to Renewal: A 12-Step Program for Innovation and Inclusion in the
California Economy, Program for Environmental and Regional Equity, University of
Southern California and Institute for Social Transformation, UC-Santa Cruz. https://
dornsife.usc.edu/pere/california-economy.
Petty, T., M. Saba, T. Lewis, S. P. Gangadharan, and V. Eubanks. 2018. “Our Data Bodies:
Reclaiming Our Data (Interim Report).” Detroit, MI: Our Data Bodies Project. https://
www.odbproject.org/tools/.
Poor People’s Campaign. 2020. “A Moral Policy Agenda to Heal and Transform America:
The Poor People’s Jubilee Platform”. https://www.poorpeoplescampaign.org/about/
jubilee-platform/.
Pūras, D. 2020. “Right of Everyone to the Enjoyment of the Highest Attainable Standard of
Physical and Mental Health,” Report of the Special Rapporteur on the right of everyone to
the enjoyment of the highest attainable standard of physical and mental health. United
Nations Office of the High Commissioner for Human Rights. https://undocs.org/en/A/
HRC/44/48.
Rolnick, D., P. Donti, L. Kaack, K. Kochanski, A. Lacoste, K. Sankaran, A. Ross et al. 2019.
“Tackling Climate Change with Machine Learning,” ArXiv [preprint]. https://arxiv.org/
abs/1906.05433.
Rudolph, L., J. Caplan, K. Ben-Moshe, and L. Dillon. 2013. Health in All Policies: A Guide for
State and Local Governments. Washington, DC: American Public Health Association and
Public Health Institute.
Sachs, J., J. Stiglitz, M. Mazzucato, C. Brown, I. Dutta-Gupta, R. Reich, G. Zucman, et al.
2020. “Letter from economists: to rebuild our world, we must end the carbon
economy,” The Guardian. https://www.theguardian.com/commentisfree/2020/aug/04/
economists-letter-carbon-economy-climate-change-rebuild.
Sarewitz, D. 2004. “How Science Makes Environmental Controversies Worse.”
Environmental Science & Policy 7 (5): 385–403.
Sarewitz, D. 2015. “CRISPR – Science Can’t Solve It.” Nature 522 (7557): 413–414.
Sarewitz, D., and R. Nelson. 2008. “Three Rules for Technological Fixes.” Nature 456 (7224):
871–872.
Schork, N. 2019. “Artificial Intelligence and Personalized Medicine.” In Precision Medicine
in Cancer Therapy, edited by D. Von Hoff, and H. Han, 265–283. Berlin, Germany:
Springer.
Serkez, Y. 2020. “Who Is Most Likely to Die From the Coronavirus?” The New York Times, 4
June 2020. https://www.nytimes.com/interactive/2020/06/04/opinion/coronavirus-health-
race-inequality.html.
Shi, Z. R., C. Wang, and F. Fang. 2020. “Artificial Intelligence for Social Good: A Survey,”
ArXiv [preprint]. https://arxiv.org/abs/2001.01818.
Shonkoff, J., and D. Phillips. 2000. From Neurons to Neighborhoods: The Science of Early
Childhood Development. Washington, D.C: National Academy Press.
Silverstein, M., H. Hsu, and A. Bell. 2019. “Addressing Social Determinants to Improve
Population Health: The Balance Between Clinical Care and Public Health.” Journal of
the American Medical Association 322 (24): 2379–2380.
Singer, N. 2017. “How Big Tech is Going After Your Health Care,” The New York Times, 26
December 2017. https://www.nytimes.com/2017/12/26/technology/big-tech-health-care.
html.
 INTERDISCIPLINARY SCIENCE REVIEWS 125

Singer, N. 2019. “When Apps Get Your Medical Data, Your Privacy May Go With It,” The
New York Times, 3 September 2019. https://www.nytimes.com/2019/09/03/technology/
smartphone-medical-records.html.
Stewart, J. 2017. “Smartphone App Shown to Accurately Diagnose Respiratory Illnesses Like
COPD,” COPD News Today. https://copdnewstoday.com/2017/12/21/resapp-
smartphone-app-accurately-diagnoses-copd-other-respiratory-illnesses/.
Strober, M. 2010. Interdisciplinary Conversations: Challenging Habits of Thought. Stanford,
CA: Stanford University Press.
Strubell, E., A. Ganesh, and A. McCallum. 2019. “Energy and Policy Considerations for
Deep Learning in NLP,” ArXiv [preprint]. https://arxiv.org/abs/1906.02243.
Sun, W., J. Lee, S. Zhang, C. Benyshek, M. Dokmeci, and A. Khademhosseini. 2018.
“Engineering Precision Medicine,” Advanced Science 6 (1): 1–19.
Sunrise Movement. 2020. “Green New Deal”; https://www.sunrisemovement.org/green-
new-deal.
Tekisalp, L. 2020. “Beyond Hype and Innovation: AI for Social Good,” Partnership on AI Blog.
https://www.partnershiponai.org/beyond-hype-and-innovation-ai-for-social-good/.
Tomašev, N., J. Cornebise, F. Hutter, S. Mohamed, A. Picciariello, B. Connelly, D. Belgrave,
et al. 2020. “AI for Social Good: Unlocking the Opportunity for Positive Impact.” Nature
Communications 11 (2468): 1–6.
Ulnicane, I., D. O. Eke, W. Knight, G. Ogoh, and B. C. Stahl. 2021. “Good governance as a
Response to Discontents? Déjà vu, or Lessons for AI from other Emerging Technologies.”
Interdisciplinary Science Reviews.
University College London (UCL), Institute for Innovation and Public Purpose. 2018. “The
people’s prescription: Re-imagining health innovation to deliver public value,” IIPP
Policy Report, 2018-10. London, UK: IIPP, Global Justice Now, Just Treatment,
STOPAIDS. https://www.ucl.ac.uk/bartlett/public-purpose/wp2018-10.
Wallack, L., and K. Thornburg. 2016. “Developmental Origins, Epigenetics, and Equity:
Moving Upstream.” Maternal and Child Health Journal 20 (5): 935–940.
Wilkinson, R., and K. Pickett. 2019. The Inner Level: How More Equal Societies Reduce Stress,
Restore Sanity and Improve Everyone’s Well-Being. UK: Penguin Random House.
Woodhouse, E., and D. Sarewitz. 2007. “Science Policies for Reducing Societal Inequities.”
Science and Public Policy 34 (2): 139–150.
Woolf, S., and L. Aron. 2013. U.S. Health in International Perspective: Shorter Lives, Poorer
Health. Washington, D.C: National Academy Press.
Woolf, S., R. Johnson, G. Fryer, Jr., G. Rust, and D. Satcher. 2008. “The Health Impact of
Resolving Racial Disparities: An Analysis of U.S. Mortality Data.” American Journal of
Public Health 98 (Suppl 1): S26–S28.
World Health Organization. 2008. Closing the Gap in a Generation: Health Equity Through
Action on the Social Determinants of Health”; Final Report of the Commission on Social
Determinants of Health. Geneva: World Health Organization.
World Health Organization. 2017. “Human Rights and Health,” WHO Fact Sheet. https://
www.who.int/news-room/fact-sheets/detail/human-rights-and-health.
Wu, H., H. Yin, H. Chen, M. Sun, X. Liu, Y. Yu, Y. Tang, et al. 2020. “A Deep Learning,
Image Based Approach for Automated Diagnosis for Inflammatory Skin Diseases.”
Annals of Translational Medicine 8 (9): 581.
Wylie, S., N. Shapiro, and M. Liboiron. 2017. “Making and Doing Politics Through
Grassroots Scientific Research on the Energy and Petrochemical Industries.” Engaging
Science, Technology, and Society 3: 393–425.