1 of 47

The Ethics of

Implications for Policy & Practice

AI Value Chains

Blair Attard-Frost (they/them)�PhD Candidate

University of Toronto

Faculty of Information�

AI Ethics Speaker Series: What is an AI System?

Research & Innovation Lab Directorate, Canada Revenue Agency�December 6, 2022

@blairasaservice

blairaf.com

linkedin.com/in/blairaf/

2 of 47

About Me

2

2015: Research Intern @ Institute for Citizen-Centred Service

2016: Information Management Intern @ Legislative Assembly of Ontario

2016: FOI Analyst @ Ontario Ministry of Environment, Conservation, and Parks

2017: Digital Transformation Consultant @ University of Toronto

2018: Research & Strategy Associate @ Ontario Centre for Workforce Innovation

2019: National Director & Strategic Advisor @ PigeonLine - ResearchAI

2019: PhD @ University of Toronto

3 of 47

Overview

What is an AI system?�
What is AI ethics?�
What is an AI value chain?

How can ethics be built into AI value chains?

3

4 of 47

What is an AI system?

01

5 of 47

5

Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press.

“Each way of defining artificial intelligence is doing work, setting a frame for how it will be understood, measured, valued, and governed.”

6 of 47

6

From Mishra, S., Clark, J., & Perrault, C. R. (2020). Measurement in AI Policy: Opportunities and Challenges. ArXiv:2009.09071 [Cs]. http://arxiv.org/abs/2009.09071

7 of 47

7

How we interpret “AI” has become an important site of political contestation

We can see this in conversations around recent AI legislation like AIDA here in Canada and the EU’s AI Act: in Canada about differences between an AI system and automated decision system and the nature of a technological system, and in the EU’s framing of AI as software and more recent re-framing as a more general goal-oriented intelligent system

Won’t go into the specifics of those conversations here, but what’s important to note is that definition is used in this legislation as an approach to setting an “interpretive frame” for AI: a different set of considerations result from defining AI as “software” rather than a “technological system” or just a “system.”

These definitions serve to set boundaries on an interpretive space and resolve some of the ambiguity of the term, which is particularly important for law and policy, where there’s usually some space afforded for interpretation, but definition serves to narrow that interpretive frame down as specifically as possible.

8 of 47

3 Interpretive Frames for AI

8

1) AI as a computational system

2) AI as a socio-technical system

3) AI as an ecosystem

9 of 47

Interpretive Frame #1

9

“AI” is a set of cognitive processes performed within a computational structure.
AI is made of data, algorithms, software, computer vision, machine learning models, neural networks, etc.

AI as a computational system

AI “seeks to make computers do the sorts of things that minds can do.” �(Margaret Boden, 2016)

10 of 47

Interpretive Frame #2

10

AI as a socio-technical system

“AI” is a set of cognitive processes and computational structures that emerge from social activities and values.
AI is made of data, algorithms, models, AND developers, users, data subjects, organizations, policies, etc.

“Intelligence manifests itself only relative to specific social and cultural contexts.” (Joseph Weizenbaum, 1976)

11 of 47

Interpretive Frame #3

11

AI as an ecosystem

“AI” is an interdependent set of computational, social, technical, and material components that integrate across multiple scales and contexts.
AI is made of data, algorithms, models, developers, users, data subjects, AND computing infrastructure, bodily labor, energy, water, minerals, etc.

“Artificial intelligence is both embodied and material, made from natural resources, fuel, human labor, infrastructures, logistics, histories, and classifications.” (Kate Crawford, 2021)

12 of 47

12

Computational components

Socio-technical components

Ecological components

��

Socio-technical components

�

�Computational components

Any of these bubbles can be interpreted as the boundary of an “AI system”
These bubbles are porous, phenomena often span multiple bubbles (e.g., a data collection process can cross all three bubbles)

13 of 47

13

Crawford, K. & Joler, V. (2018). Anatomy of an AI system: The Amazon Echo as a map of labor, data, and planetary resources. https://www.anatomyof.ai

The porous nature of AI systems and interpretive frames for AI is beautifully illustrated by Kate Crawford & Vladen Joler’s “Anatomy of an AI system” project

This project maps out all the computational, socio-technical, and ecological components involved in creating and operating the Amazon Echo. Five areas of the map highlighted in red boxes: AI training, manufacturing and ethical concerns associated with it, disposal of e-waste and environmental impacts, AWS infrastructure and the environmental impacts of its energy and water consumption in order to train the model and use the system, hardware assembly and ethical issues associated with that like right to repair, open-source design, privacy, security

This map demands us to ask: where does the “AI system” begin and end here? Or is the entire diagram the AI system? The implication of this for our interpretive framing and scoping of AI systems is quite radical: components like infrastructure, human labor, and material resources are more than just the context of the AI system, they are inextricable components of the system itself.

These components are all part of the system, they give the computational “brain” the material and energy it needs to operate - the AI system would cease to function without the bodily labor of humans and the material resources being input into it

14 of 47

14

AI systems represent “an extended circle of relationships that includes the non-human kin—from network daemons to robot dogs to artificial intelligences (AI) weak and, eventually, strong—that increasingly populate our computational biosphere.” (Lewis et al., 2018)

Indigenous Protocol and Artificial Intelligence Working Group (2019). Indigenous AI. https://www.indigenous-ai.net/

Lewis, J. E., Arista, N., Pechawis, A., & Kite, S. (2018). Making kin with the machines. Journal of Design Science. https://jods.mitpress.mit.edu/pub/lewis-arista-pechawis-kite/release/1

15 of 47

What is AI ethics?

02

16 of 47

16

AI ethics is a field of study seeking “a set of values, principles, and techniques that employ widely accepted standards of right and wrong to guide moral conduct in the development and use of AI technologies” (p. 3).

Leslie, D. (2020). Understanding artificial intelligence ethics and safety: A guide for the responsible design and implementation of AI systems in the public sector. The Alan Turing Institute. https://www.turing.ac.uk/sites/default/files/2019-06/understanding_artificial_intelligence_ethics_and_safety.pdf

17 of 47

17

Attard-Frost, B., De los Ríos, A., & Walters, D. R. (2022). The ethics of AI business practices: A review of 47 AI ethics guidelines. AI and Ethics. https://doi.org/10.1007/s43681-022-00156-6

Hagendorff, T. (2020). The ethics of AI ethics: An evaluation of guidelines. Minds and Machines (30), 99-120, p. 103. https://link.springer.com/content/pdf/10.1007/s11023-020-09517-8.pdf

In the domain of AI ethics, we see quite clearly how some interpretive frames of AI favor the analysis of particular system components and ethical concerns while pushing other concerns to the margin

Recent reviews/criticisms of AI ethics guidelines have highlighted this point

Thilo Hagendorff’s paper points out the limitations of approaching AI ethics with a computational framing than than a more comprehensive ecological framing

Similarly, a review of AI ethics guidelines I recently co-authored also points out that approaching AI ethics with a computational framing (putting algorithmic decision-making the center of ethical reasoning) limits us from engaging in a broader conversation about the business practices, business decision-making, and business ethics involved in developing and using AI systems

18 of 47

Ethical Concerns

18

AI as a computational system

	Computational system	Socio-technical system	Ecosystem
Example�components	Data, algorithms, software, computer vision, machine learning models, neural networks
Ethical concerns	Data bias, model bias, algorithmic transparency, explainability

19 of 47

Ethical Concerns

19

AI as a socio-technical system

	Computational system	Socio-technical system	Ecosystem
Example�components	Data, algorithms, software, computer vision, machine learning models, neural networks	All computational components AND developers, users, data subjects, organizations, policies
Ethical concerns	Data bias, model bias, algorithmic transparency, explainability	All computational concerns AND human bias, institutional bias, privacy, inclusivity, accountability, liability for harms, legal enforceability, accessibility, justice

20 of 47

Ethical Concerns

20

AI as an ecosystem

	Computational system	Socio-technical system	Ecosystem
Example�components	Data, algorithms, software, computer vision, machine learning models, neural networks	All computational components AND developers, users, data subjects, organizations, policies	All computational and socio-technical components AND computing infrastructure, bodily labor, energy, water, minerals
Ethical concerns	Data bias, model bias, algorithmic transparency, explainability	All computational concerns AND human bias, institutional bias, privacy, inclusivity, accountability, liability for harms, legal enforceability, accessibility, justice	All computational and socio-technical concerns AND material harms, environmental harms, sustainability, labor conditions, physical maintenance

21 of 47

Practical Implications

21

	Computational system	Socio-technical system	Ecosystem
Example�components	Data, algorithms, software, computer vision, machine learning models, neural networks	All computational components AND developers, users, data subjects, organizations, policies	All computational and socio-technical components AND computing infrastructure, bodily labor, energy, water, minerals
Ethical concerns	Data bias, model bias, algorithmic transparency, explainability	All computational concerns AND human bias, institutional bias, privacy, inclusivity, accountability, liability for harms, legal enforceability, accessibility, justice	All computational and socio-technical concerns AND material harms, environmental harms, sustainability, labor conditions, physical maintenance
Practical implications	Debias datasets Validate models Use explainable algorithms	Design inclusive/accountable AI AI standardization & auditing Develop regulatory frameworks	Develop stewardship models Mitigate risks to workers & environment Assess ethics of business practices

22 of 47

Policy Areas

22

	Computational system	Socio-technical system	Ecosystem
Example�components	Data, algorithms, software, computer vision, machine learning models, neural networks	All computational components AND developers, users, data subjects, organizations, policies	All computational and socio-technical components AND computing infrastructure, bodily labor, energy, water, minerals
Ethical concerns	Data bias, model bias, algorithmic transparency, explainability	All computational concerns AND human bias, institutional bias, privacy, inclusivity, accountability, liability for harms, legal enforceability, accessibility, justice	All computational and socio-technical concerns AND material harms, environmental harms, sustainability, labor conditions, physical maintenance
Practical implications	Debias datasets Validate models Use explainable algorithms	Design inclusive/accountable AI AI standardization & auditing Develop regulatory frameworks	Develop stewardship models Mitigate risks to workers & environment Assess ethics of business practices
Relevant�policy areas	Technology policy	Technology policy, innovation policy, privacy policy	Technology policy, innovation policy, privacy policy, social policy, labor/hiring policy, environmental policy

23 of 47

The Integration Challenge

23

AI ethics guidelines usually center computational concerns and a small handful of socio-technical concerns

Many social, political, economic, and ecological concerns and marginalized groups are often treated as externalities�
How can we take a more integrated view of ethical concerns across multiple scales, contexts, and groups?

�Computational components & concerns

Socio-technical�components &�concerns

��

Ecological�components &�concerns

��

24 of 47

What is an AI value chain?

03

25 of 47

Value Chains

25

Value chains: Sequential economic structures through which multiple actors integrate tangible and intangible resources with the goal of co-creating value.

Three structural features of value chains:

(1) Situated: Resourcing activities that occur within value chains are situated within social and environmental contexts.

(2) Patterned: Resourcing activities that occur within value chains are spatially and temporally patterned and so can recur with some degree of regularity.

(3) Differently Perceived: Resourcing activities that occur within value chains are differently perceived and evaluated from various actor perspectives.

26 of 47

AI Value Chains

26

API access: One actor writes the code and trains the system, then sells access to other actors through an API

System contracting: One actor develops an AI system for another actor

Fine-tuning: Initial development by one actor and fine-tuning by another

Model integration: One actor integrates different AI systems into a new one

Engler, A. & Renda, A. (2022). Reconciling the AI value chain with the EU’s Artificial Intelligence Act. Centre for European Policy Studies. https://www.ceps.eu/ceps-publications/reconciling-the-ai-value-chain-with-the-eus-artificial-intelligence-act/

A recent paper published by Centre for European Policy Studies outlines 7 types of AI value chains in which computational resources such as AI models and APIs can be exchanged to co-create value

I’ve outlined 4 examples from that paper here: system contracting; model integration; fine-tuning; API access

Each of these types of value chains has unique ethical and policy considerations associated with it. But what I want to focus in on here is the interpretive frame for AI systems that these 4 value chains have tacitly adopted: it’s a computational frame, where AI systems are mainly computational components

I want to especially focus in on this final value chain type in the bottom-right: the API access value chain - in which one actor writes code and trains a proprietary system, then sells access to another actor through an API - in order to explore the questions we’ve discussed here in more detail

27 of 47

27

Is this really all there is to an “AI system?”

28 of 47

28

Expanding the scope of the AI system & its value chains causes new ethical concerns to emerge:

Who gets to access high quality training data? Financial barriers to access? Open access?

Data hoarding and industry consolidation? Data cartels? Barriers to fair markets? (see Sarah Lambdan, 2022)

The model developer might also need a large volume of high-quality training/testing data that they might have procure from another actor - a data provider that creates and curates datasets such as Google, ImageNet, or Waymo

If it’s not a public dataset, if it’s one that only provided for payment - might there be ethical concerns regarding financial barriers to access? Who gets to access and use the data, what costs are associated with accessing it? Or even if it is a public dataset, might there be ethical concerns regarding data hoarding or consolidation of high-quality datasets into a small handful of organizations? Might that encourage anti-competitive practices or pose barriers to fair markets?

I’m not going to go into much analysis of these particular ethical concerns - Sarah Lambdan’s Data Cartels and many others have done great work on issues related to data hoarding and consolidation - the point here is that if we bring data provider value chains into the more simple developer-client value chains, there’s a whole new set of ethical concerns that emerge that weren’t present before

29 of 47

29

Is this really all there is to an AI system?

30 of 47

30

Harmful work conditions associated with data work? Socio-economic harms? Psychological harms?

Are data workers being compensated fairly by the data provider and the platform operator? Are they being exposed to disturbing media?

Williams, A., Miceli, M., & Gebru, T. (2022, October 13). The exploited labor behind artificial intelligence. Noema. https://www.noemamag.com/the-exploited-labor-behind-artificial-intelligence/

We can keep expanding these value chains out into an even broader network if we think about labor practices involved in AI development and use

As a hypothetical example here in blue in this diagram, I’ve illustrated some of the resources that might be required if that training dataset received from the data provider contains data that was collected or annotated by data workers or microworkers

Recently been some very insightful research on this topic: article in the top right is a great overview of some of that research.

Some of the key ethical concerns here center around the use of data workers to collect, annotate, or verify data/automated decisions, particularly through the use of crowdsourcing platforms like Amazon MTurk

Especially in the case of outsourced data work, many concerns regarding unfair compensation to these workers as well as concerns regarding psychological harms that many of these workers experience after manually reviewing disturbing or graphic content as part of a data preparation or content moderation workflow

31 of 47

31

Accountability and responsibility? Should the model developer bear any responsibility if the client uses the model output for a malicious purpose? Does the cloud computing provider bear any responsibility?

32 of 47

32

Environmental impacts of the AI system: should energy/water use of the model be optimized or limited?

See Global Partnership on AI (2021), �Gupta (2021)

https://gpai.ai/projects/climate-change-and-ai.pdf

https://thegradient.pub/sustainable-ai/

33 of 47

33

Bias in the STEM workforce? Bias in datasets and models? Might bias reproduce through value chains? (see West, Whittaker, & Crawford, 2019; �Widder & Nafus, 2022)

Enforceability of AI development & use requirements: how can regulators, auditors, and other AI governance service providers ensure compliance with their requirements?

If we include Education & Training providers, we might also think about how value chains enable bias in AI education to then reproduce in the AI workforce and then ultimately reproduce in the creation of datasets and models

We might also think about the value chains involved in AI governance and what kinds of resources are required for AI development & use requirements to be enforceable: how can regulators, auditors, and other AI governance service providers ensure compliance with their requirements? What kinds of laws, policy instruments, audit frameworks, standards, or political or regulatory institutions might be needed?

Lots of issues we could talk about here, we could keep going and going and expanding this diagram in many directions, zooming in deeper to the various actors, but the main takeaway from these diagrams: value chains ground our analysis of ethical concerns in the specific practices, experiences, abilities, and relations that link people together

This entails a different approach to ethical reasoning than we typically see in AI ethics

34 of 47

AI Ethics as a Value Chain Ethics?

34

Value chain ethics: An approach to ethical reasoning based on three premises:

Value chains are resource integration structures that connect activity across multiple scales and contexts.
Value chains reflect underlying values, experiences, and abilities of multiple actors.
Value chains enable value co-creation, from which value conflicts and ethical concerns can arise.

35 of 47

How can ethics be built into �AI value chains?

04

36 of 47

Value Chain Governance for AI

36

Model for governing AI systems by intervening in the value chains that enable their development and use

Integrates computational, socio-technical, and ecological components & concerns

�Governance of computational components & concerns

�Governance of�socio-technical components &�concerns

��

Governance of �ecological components &�concerns�

��

Some tools for governing AI value chains:

Value network analysis
Service systems analysis
Expanded AI ethics guidelines
AI value chain standards
Policies & regulations for AI value chains

37 of 47

Analysis Methods

37

Value Network Analysis

Originally created as a business analysis method by Verna Allee (2008), but can be used to analyze any ecosystem

Scalability & flexibility: Analysis can be qualitative, quantitative, top-down, bottom-up, macro-scale, or micro-scale

Scope ambiguity: Value networks are challenging to scope - analyst must represent many activities, resources, actors, contexts, scales of activity, levels of aggregation

Requires constant reflexivity and participatory analysis to apply effectively

38 of 47

Analysis Methods

38

Service System Analysis

Service systems are economic structures through which actors with differing values and abilities attempt to create mutual benefit for one another

Service system analysis accounts for service interactions, intangible resources, and institutional factors that often make AI impacts difficult to evaluate

Frost, R. B., Cheng, M., & Lyons, K. (2019). A multilayer framework for service system analysis. Handbook of Service Science, Volume II, 285-306.

39 of 47

Expanded Ethics Guidelines

39

Economic context & distributive justice

AI ethics guidelines that address a greater variety of AI system components and ethical concerns, such as:

Ecological context & sustainability

Business context & �Labor rights

O’Keefe et al. (2020) The Windfall Clause: Distributing the benefits of AI for the common good. Future of Humanity Institute, University of Oxford. https://www.fhi.ox.ac.uk/windfallclause/

de Peuter, G., de Verteuil, G., & Machaka, S. (2022). Co-operatives, work, and the digital economy: A knowledge synthesis report. Cultural Workers Organize. https://culturalworkersorganize.org/co-operatives-work-and-the-digital-economy-a-knowledge-synthesis-report/��Gupta, A. (2021). The imperative for sustainable AI systems. The Gradient. https://thegradient.pub/sustainable-ai/

We might also think about expanding the scope of AI ethics guidelines to address a greater variety of system components and ethical concerns: economic components and principles/concerns of distributive justice, business and labor concerns and principles of labor rights and human rights, ecological concerns and principles of sustainability.

These kinds of concerns and different types of practical and policy solutions to them have been covered in many research papers and reports, but AI ethics guidelines typically do not address these concerns. Instead, AI ethics guidelines remain focused primarily on computational concerns and a small set of socio-technical concerns. Integrating a greater variety of ethical principles and practices into future AI ethics guidelines would expand their applicability to cover a much wider variety of AI system components and ethical concerns.

40 of 47

Value Chain Standards

40

Fairwork (2022). Cloudwork (online platform work) principles. https://fair.work/en/fw/principles/cloudwork-principles/

Responsible AI Institute (2022). https://www.responsible.ai/how-we-help

SQF Institute (2017). SQFI Ethical Sourcing. Edition 2.1. https://www.sqfi.com/wp-content/uploads/2019/02/Ethical-Sourcing-Code.pdf

Integrate more social & environmental principles into emerging AI standardization & certification programs

There’s a lot of interest in AI standardization at the moment: for example, the Responsible AI Institute is in the process of building and testing a conformity assessment & certification program. It focuses mainly on dimensions such as data quality, accountability, bias/fairness, explainability/interpretability, robustness - these could be expanded into a greater variety of social and ecological dimensions to account for more of the value chains of ethical concern we covered today.

As another example, Fairwork is a UK-based initiative that develops principles and assessments of fair work practices for digital platform workers. These principles and assessments don’t deal exclusively with AI, but because they deal with other digital ecosystems, similar principles and assessment methods could be integrated into AI standards to better account for ethical concerns around labor practices.

Finally, as an example of a more robust certification program that takes a value chain approach in completely separate industry, the Safe Quality Food Institute’s Ethical Sourcing Certification provides an end-to-end assessment framework that covers virtually the entirety of food supply value chains. A value chain based standard/certification program like this could serve as a model for more comprehensive standards/certification programs in the AI/tech industry

41 of 47

AI Policy: A Value Chain Approach

41

A value chain approach to AI policy would identify and target specific value chains of ethical concern �rather than targeting general-purpose AI systems or broad application areas.

Participatory policy design: Public engagement throughout the policymaking process to ensure values & interests of vulnerable individuals and groups are represented in laws, regulations, and standards

Mitigation of input risks: Risk management of the activities involved in the resource inputs of AI systems, rather than current output-centric risk models

Regulatory coordination & co-enforcement: AI regulation is co-developed and co-enforced by technology oversight agencies alongside other relevant regulatory agencies and civil society stakeholders, based on ethical/policy concerns of relevance to other regulatory agencies (e.g., human/Charter rights concerns, labor rights concerns, environmental concerns)

https://www.telefonica.com/en/communication-room/blog/a-fit-for-purpose-and-borderless-european-artificial-intelligence-regulation/

42 of 47

Thank you!

42

43 of 47

43

This presentation draws on research supported by the Social Sciences and Humanities Research Council of Canada�and the University of Toronto Faculty of Information

44 of 47

Further Reading

44

Boden, M. (2016). AI: Its nature and future. Oxford University Press.

Bratton, B. H. (2021). Synthetic gardens: Another model for AI and design. In B. Vickers & K. Allado-McDowell (Eds.), Atlas of Anomalous AI (pp. 91–105). Ignota.

Bryson, J. (2022, March 2). Europe is in danger of using the wrong definition of AI. Wired. https://www.wired.com/story/artificial-intelligence-regulation-european-union/

Council of the European Union (2021). Artificial Intelligence Act - Presidency Compromise Text. https://data.consilium.europa.eu/doc/document/ST-14278-2021-INIT/en/pdf

Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary Costs of Artificial Intelligence. Yale University Press.

Crawford, K. & Joler, V. (2018). Anatomy of an AI system: The Amazon Echo as a map of labor, data, and planetary resources. https://www.anatomyof.ai

Cristianini, N., Scantamburlo, T., Ladyman, J. (2021). The social turn of artificial intelligence. AI & Society. https://link.springer.com/content/pdf/10.1007/s00146-021-01289-8.pdf

Halpern, O. (2021). Planetary intelligence. In J. Roberge & M. Castelle (Eds.), The Cultural Life of Machine Learning: An Incursion into Critical AI Studies (227-256) Palgrave Macmillan.

Haugeland, J. (1989). Artificial intelligence: The very idea. Bradford Books.

Indigenous Protocol and Artificial Intelligence Working Group (2019). Indigenous AI. https://www.indigenous-ai.net/

Lewis, J. E., Arista, N., Pechawis, A., & Kite, S. (2018). Making kin with the machines. Journal of Design Science. https://jods.mitpress.mit.edu/pub/lewis-arista-pechawis-kite/release/1

Mishra, S., Clark, J., & Perrault, C. R. (2020). Measurement in AI policy: Opportunities and challenges. http://arxiv.org/abs/2009.09071

Parliament of Canada (2022). Bill C-27. https://www.parl.ca/DocumentViewer/en/44-1/bill/C-27/first-reading

Weizenbaum, J. (1976). Computer Power and Human Reason: From Judgement to Calculation. W. H. Freeman & Co.

What is an AI system?

45 of 47

Further Reading

45

Attard-Frost, B., De los Ríos, A., & Walters, D. R. (2022). The ethics of AI business practices: A review of 47 AI ethics guidelines. AI and Ethics. https://doi.org/10.1007/s43681-022-00156-6

Birhane, A. (2021). Algorithmic injustice: A relational ethics approach. Patterns, 2(2).

Gray, J. & Witt, A. (2021). A feminist data ethics of care for machine learning: The what, why, who and how. First Monday, 26(12).

Greene, D., Hoffman, A. L., & Stark, L. (2019). Better, nicer, clearer, fairer: A critical assessment of the movement for ethical artificial intelligence and machine learning. Proceedings of the 52nd Hawaii International Conference on System Sciences, 2122-2131. https://scholarspace.manoa.hawaii.edu/bitstream/10125/59651/0211.pdf

Gwagwa, A., Kazim, E., & Hilliard, A. (2022). The role of the African value of Ubuntu in global AI inclusion discourse: A normative ethics perspective. Patterns, 3(4).

Hagendorff, T. (2020). The ethics of AI ethics: An evaluation of guidelines. Minds and Machines, 30, 99-120. https://link.springer.com/content/pdf/10.1007/s11023-020-09517-8.pdf

Häußermann, J. J. & Lütge, C. (2021). Community-in-the-loop: Towards pluralistic value creation in AI, or—why AI needs business ethics. AI and Ethics, 2021.

Lauer, D. (2021). You cannot have AI ethics without ethics. AI and Ethics. https://doi.org/10.1007/s43681-020-00013-4

Leslie, D. (2020). Understanding artificial intelligence ethics and safety: A guide for the responsible design and implementation of AI systems in the public sector. The Alan Turing Institute. https://www.turing.ac.uk/sites/default/files/2019-06/understanding_artificial_intelligence_ethics_and_safety.pdf

Rességuier, A. & Rodrigues, R. (2020). AI ethics should not remain toothless! A call to bring back the teeth of ethics. Big Data & Society, July 2020, 1-5.

van Wynsberghe, A. (2021). Sustainable AI: AI for sustainability and the sustainability of AI. AI and Ethics. https://link.springer.com/article/10.1007/s43681-021-00043-6

What is AI ethics?

46 of 47

Further Reading

46

Global Partnership on AI (2021). Climate change & AI: Recommendations for government. https://gpai.ai/projects/climate-change-and-ai.pdf

Gupta, A. (2021). The imperative for sustainable AI systems. The Gradient. https://thegradient.pub/sustainable-ai/

Howson, K., Johnston, H. Cole, M., Ferrari, F., Ustek-Spilda, F., & Graham, M. (2022). Unpaid labour and territorial extraction in digital value networks. Global Networks: A Journal of Transnational Affairs. https://onlinelibrary.wiley.com/doi/full/10.1111/glob.12407

Lambdan, S. (2022). Data cartels: The companies that control and monopolize our information. Stanford University Press.

Ligozat, A.-L., Lefevre, J. Bugeau, A., & Combaz, J. (2022). Unraveling the hidden environmental impacts of AI solutions for environment life cycle assessment of AI solutions. Sustainability, 14(9), 5172. https://www.mdpi.com/2071-1050/14/9/5172

Miceli, M. & Posada, J. (2022). The data-production dispositif. Retrieved from https://arxiv.org/pdf/2205.11963.pdf

Sambasivan, N., Kapania, S., Highfill, H., Akrong, D., Paritosh, P., & Aroyo, L. (2021). Everyone wants to do the model work, not the data work: Data cascades in high-stakes AI. CHI 21, May 8-13, 2021.

Strubell, E., Ganesh, A., McCallum, A. (2019). Energy and policy considerations for deep learning in NLP. https://arxiv.org/abs/1906.02243

West, S. M., Whittaker, M., & Crawford, K. (2019). Discriminating systems: Gender, race, and power in AI. AI Now Institute. https://ainowinstitute.org/discriminatingsystems.pdf

Widder, D. G. & Nafus, D. (2022). Dislocated accountabilities in the AI supply chain: Modularity and developers’ notions of responsibility. https://arxiv.org/pdf/2209.09780.pdf

Williams, A., Miceli, M., & Gebru, T. (2022, October 13). The exploited labor behind artificial intelligence. Noema. Retrieved from https://www.noemamag.com/the-exploited-labor-behind-artificial-intelligence/

What is an AI value chain?

47 of 47

Further Reading

47

Allee, V. (2008). Value network analysis and value conversion of tangible and intangible assets. Journal of Intellectual Capital, 9(1), 5-24.

Gupta, A. (2021). The imperative for sustainable AI systems. The Gradient. https://thegradient.pub/sustainable-ai/

Fairwork (2022). Cloudwork (online platform work) principles. https://fair.work/en/fw/principles/cloudwork-principles/

Frost, R. B., Cheng, M., & Lyons, K. (2019). A Multilayer Framework for Service System Analysis. In P. P. Maglio, C. A. Kieliszewski, J. C. Spohrer, K. Lyons, L. Patrício, & Y. Sawatani (Eds.), Handbook of Service Science, Volume II (pp. 285–306). Springer International Publishing.

Lim, C., Kim, K.-H., Kim, M.-J., Heo, J.-Y., Kim, K.-J., & Maglio, P. P. (2018). From data to value: A nine-factor framework for data-based value creation in information-intensive services. International Journal of Information Management, 39, 121-135.

Maglio, P. P., Vargo, S. L., Caswell, N., & Spohrer, J. (2009). The service system is the basic abstraction of service science. Information Systems and e-Business Management, 7, 395-406.

O’Keefe, C., Cihon, P., Garfinkel, B., Flynn, C., Leung, J., & Dafoe, A. (2020). The Windfall Clause: Distributing the benefits of AI for the common good. Future of Humanity Institute, University of Oxford. https://www.fhi.ox.ac.uk/windfallclause/

Responsible AI Institute (2021, April). Responsible Artificial Intelligence (RAI) Certification Beta. https://assets.ctfassets.net/rz1q59puyoaw/1myaH22mA16Y0eIXQND3qv/7974df6bd0973e65f100d327b93129a2/Whitepaper.pdf

How can ethics be built into AI value chains?