Transition Towards AI for Sustainability - Possibility and Challenges of Implementing Transition Design Framework in the AI Industry

(1)

TRANSITION TOWARDS

AI FOR SUSTAINABILITY

Possibility and Challenges of

Implementing Transition Design Framework in the AI Industry

Kevalin Saksiamkul

Paolo Nardi Fernandez

Main field of study – Leadership and Organisation

Degree of Master of Arts (60 credits) with a Major in Leadership and Organisation

Master Thesis with a focus on Leadership and Organisation for Sustainability (OL646E), 15 credits Spring Semester, 2019

(2)

ABSTRACT

Today humanity sits at the intersection of two major drivers of change: sustainable development and artificial intelligence. The former aimed at promoting the creation of a sustainable world where the economy, environment and society are balanced in such a way as to meet the sustainable development goals set out by the United Nations. The latter being a powerful technological tool that acts as a double-edged sword. One no hand, AI can create unprecedented innovations, economic growth, and serve as a game changer for the pursuit of sustainable development. On the other hand, AI solutions can unintentionally harm the very people they are supposed to help, increase inequality and inhibit sustainable development. To ensure that the latter doesn’t occur, it is crucial that AI practitioners and decision-makers come together to envision a sustainable future enabled by AI. To do this, a deep understanding of the current state of the industry, the problems it faces, multidisciplinary collaborations, and the usage of new tools and ways of designing to solve wicked sustainability problems are needed. As suggested by a group of design professionals and academicians from Carnegie Mellon University, Transition Design can be useful when working to solve such a complex problem. However, the practical implication aspect of Transition Design has never been explored in connection to AI in relation to sustainability. Thus, this study provides a basic understanding of the factors that affect AI practitioners and how those factors align and conflict with the Transition design framework for sustainability. Some opportunities and barriers to implement the framework were discovered by looking at their relationship. Most importantly, the study served as a starting point to work towards practical implication of the Transition Design framework in the AI industry and other business areas, to bring about transitions towards sustainable development.

(3)

4.2.1.1 Automation and optimization through AI in the working process 32 4.2.1.2 Integration of AI across all domains could lead to changes in the society 32 4.2.1.3 New skills needed for the AI driven future are identified 32 4.2.1.4 See the potential to reduce negative environmental impact through AI 33

4.2.2 Big Companies’ Perspective 3₃

4.2.2.1 Economic changes 33

4.3 Regulations that Affect the Development Processes of AI 3₃

4.3.1 SME perspective 3₃

4.3.1.1 Regulations depend on the organizational values 3₃

4.3.2 Big companies perspective 34

4.3.2.1 Direct regulations 34

4.3.2.2 Indirect regulations 34

4.3.3 Shared perspective 35

4.3.3.1 Explainability discourse 35

4.3.3.2 Data management discourse 35

4.3.3.3 Regulation discourse 35

4.4 Challenges and improvements in the industry 36

(i) Identified challenges in the industry 36

4.4.1 Challenges from SME perspective 36

4.4.1.1 Infrastructure problems 37

4.4.1.2 Working process problems 37

4.4.1.3 Distribution of resources 37

4.4.2 Challenges from big companies perspective 37

4.4.2.1 Sustainability challenges 38

4.4.2.2 Development bottlenecks 38

4.4.2.3 Societal and value conflicts 38

4.4.2.4 Inclusivity challenges 38

4.4.3 Challenges from shared perspective 39

4.4.3.1 Public understanding 39

4.4.3.2 Personnel challenges 39

4.4.3.3 AI governance 39

4.4.3.4 Lack of transparency, openness and effort to collaborate 39

(ii) Identified improvements in the industry 40

4.4.4 Improvements from SME perspective 40

4.4.4.1 Infrastructure improvement 40

4.4.4.2 Need of more transparent auditing for the impact of the technology 40 4.4.4.3 Reducing resource inequality within the industry 40

4.4.5 Improvements from big companies perspective 41

4.4.5.1 Creation of a compatible and accessible infrastructure 41 4.4.5.2 Need for eco-friendly, smart data infrastructure driven by

the AI for good movement 41

(5)

4.4.6.1 Improve public understanding behind technology 4₂ 4.4.6.2 Make models more transparent, open and explainable by educating

practitioners on the creation of useful, scalable and accessible solutions 42 4.4.6.3 Need for collaboration within team and industry 4₂ 4.4.6.4 Need to understand how tech can fundamentally change

human kind, to be intentional of where investments go to 43 4.4.6.5 Be more clear on the ethical guideline and privacy aspect of the

technology, as well as the decentralization of the tech giants 43

4.5 Design and Development Procedures 43

4.5.1 Discovery Phase 44

4.5.2 Development Cycle 44

4.5.3 Production Cycle 44

4.6 Collaboration During the Development Process 45

4.6.1.1 Collaborate mainly with the clients who are not familiar with AI. On the other hand, collaboration with third-parties happens only when

those functions are needed. 45

4.6.2 Big Companies’ Perspective 45

4.6.2.1 Working to push for industry's standard to understand the

long-term vision of the technology 45

4.6.3 Shared Perspective 46

4.6.3.1 Not enough collaboration with outsiders, due to conflicting

values with data access. Collaboration is mainly between partners and clients. 46 4.6.3.2 If the project needs the input from domain expertises,

the practitioners are willing to collaborate to get the insight and

cross-check biases. 46

4.6.3.3 Practitioners mainly collaborate with economic stakeholders 46

5. ANALYSIS 47

5.1 Synergy 48

5.2 Contradiction 51

5.3 Polarization 53

6. DISCUSSION AND CONCLUSION 55

REFERENCES 57

APPENDICES 61

Appendix 1: AI capability in relation to the sustainable development goals 61 Appendix 2: Risk profile of the different social impact domains 62 Appendix 3: The mitigation and adaptation driven climate change solution domains 63 Appendix 4: Assessment of the impact of AI on the SDGs 65

Appendix 5: Holistic VS Dominant mindset 66

Appendix 6: Max-Neef’s theory of needs for Transition Design 68 Appendix 7: The Multi-level perspective (MLP) as a tool to understand

socio-economic context and create transition to sustainability 69 Appendix 8: The Winterhouse matrix as a tool to design solution for

multi-level changes 70

Appendix 9: Analysis for motivation and values 71

(6)

Appendix 11: Analysis for regulations that affect the development processes 73 Appendix 12: Analysis for challenges and improvements in the industry 74 Appendix 13: Analysis for design and development procedures 75 Appendix 14: Analysis for collaboration during the development processes 76

Appendix 15: Interviewee A's summary 77

Appendix 16: Interviewee B's summary 78

Appendix 17: Interviewee C's summary 79

Appendix 18: Interviewee D's summary 80

Appendix 19: Interviewee E's summary 81

Appendix 20: Interviewee F's summary 82

Appendix 21: Interviewee G's summary 83

Appendix 22: Interviewee H's summary 84

(7)

1. INTRODUCTION

1.1 Background

1.1.1 Artificial Intelligence and Sustainability Dilemma

Figure 1: Industrial Revolution Landscape (Khoshafian, 2017)

In the past 200 years, a series of industrial revolutions have radically transformed the living conditions for human beings, where each revolution has borrowed from the future to pay for the present by achieving economic growth through the degradation of our planet’s health (PricewaterhouseCoopers, 2019). This in turn, has created numerous problems such as ecological crisis, economical inequalities, polarized political views and much more. The way we live nowadays impacts both human and natural systems in the way we couldn’t imagine. In other words, we are in the so-called Anthropocene age (PricewaterhouseCoopers, 2019) where human activity is the dominant influence on the planet. In response to these conflicts, the concern about planetary boundaries arose. Agenda 2030 has been created during the UN summit in 2015 to emphasize the need for _{sustainable development} (sustainable development.un.org, 2015) in which we care about the well-being of the future generation as much as ours.

While the effects of the latest industrial revolution are still perceivable, the world’s economy is now in the middle of the fourth industrial revolution (Kravchenko & Kyzymenko, 2019). This time to the data-driven economy, in which the world is fueled with technology and data, which have become the new wealth-drivers of humanity (Cavanillas & Wahlster, 2016). Amongst the vast technological innovations, Artificial intelligence, AI for short, stands out, as it’s predicted to generate the biggest disruption to our current socio-economic system. As pointed by (PricewaterhouseCoopers, 2019), the potential contribution of AI to the global economy by 2030 could be as much as US$15.7 trillion, impacting millions of jobs and making it the biggest commercial opportunity in today's fast-changing economy. On the other hand, the importance of AI may result in increased inequalities due to the unevenly distributed educational and computing resources, as well as the creation of computational propaganda based on big data–big nudging (Vinesa et al., 2020). In addition, as mentioned 1

(8)

by (Michael Chui et al., 2018), AI solutions can unintentionally harm the very people they are supposed to help, where the malicious use of AI can harm organizations and society at large. Lastly, despite the fact that sustainable AI or ethical AI are encouraged by practitioners and academicians around the globe (Kim et al., 2018; Li et al., 2016; Goralski, 2020), there are chances that AI will be used in unsustainable ways.

Looking at the passages above, today we sit at the intersection of the AI-driven and the anthropocene age (PricewaterhouseCoopers, 2019). The contradiction between the need to undo negative effects of human’s activity on nature and the need to grow economically is, therefore, a huge dilemma. The failure to merge these worlds has raised a host of complex questions and broad concerns about how technology will affect our society and environment. Thus, the necessity for humans to transform industries, markets, and behaviors to change the course of ecological crisis and to lay the foundations for a positive, safe and responsible digital future is needed (PricewaterhouseCoopers, 2019). However, not enough has been done to bring these two paradigms together, where a huge opportunity is foregone if leaders and decision-makers do not help enable AI innovations for sustainable development (PricewaterhouseCoopers, 2019)

Figure 2: Visualization of the current stage of the world (Own creation)

In conclusion, our innate human ability to innovate, our bounded rationality, lack of knowledge and differences have brought us great wealth and innovations at a heavy price to the environment and society. With that being said, today we sit in a time where we can learn from the past and utilize our newly acquired knowledge and tools to enable us to craft a sustainable future. Thus, now more than ever, it is of crucial importance that we face our current problems and come together to understand what’s needed for us to transition to such a future. All of this brings new challenges to our current mental model of how we operate.

(9)

Therefore, exploring tools and frameworks that might be useful for future-making is necessary, but is there such a framework?

1.1.2 Envisioning the Future Through the Lenses of Transition Design

When it comes to merging AI potential with sustainable development goals, envisioning the narratives of the future can be helpful to get us to where we want (Lippitt, 1998). Through the system thinking process, futuring helps businesses to predict the problems and make strategic decisions to avoid the potential harms ( _{Millett, 2006). Future study is not a new} thing. The concept has been around since 1950s, initiated by a social scientist Ron Lippitt (Lippitt, 1998), and has been discussed throughout various fields of study: strategy and leadership (_{Millett, 2006), governmental policy making (Misuraca, 2009), psychology (Sools} and Hein Mooren, 2012), as well as in the design field (Fry, 2009; Hyysalo et al., 2014). The design school of thought has been evolving and developing continuously within the changing societal context. From serving the complete commercial purposes–industrial product design (Meier, Roy, and Seliger, 2010) and service design (Zomerdijk and Voss, 2010), to addressing the social needs–design for social innovation (Chick, 2012). In addition, design methods have been widely adapted and used as a solution development framework in various fields and disciplines (Tschimmel, 2012; Wrigley and Straker, 2017; Lindberg,

Meinel, and Wagner, 2011). This proves the practicality and flexibility of design as a

framework to create solutions. Several design methodology has been developed in response to the trend of future study: Speculative design ( Auger, 2013) and Design futuring (Fry, 2009), for instance. However, there is a recently emerged approach that emphasizes envisioning the future and focuses on creating transitions to that vision, called _Transition Design.

Transition Design was provoked by a group of designers from Carnegie Mellon Design School in the US in 2015 (Irwin, 2015), with an aim to bring the discussion to a more long-term, holistic, and sustainable design solution. Developed from the transition movement led by Rob Hopkins (Hopkins, 2014), Transition Design brings together new knowledge and skill sets aimed at seeding and catalyzing systems-level change (Irwin, 2018), which are presented through a framework with four pillars: (i) Vision for Transition, (ii) Theories of Change, (iii) Mindset and Postures, and (iv) New Ways of Designing. Transition Design encourages designers and change makers to create a clear vision of futures before thinking how to get there. The idea of creating and facilitating the preferable future has been adopted by many groups and projects around the world. The most famous one is the Transition Network, a community-based movement that encourages and empowers people around the world to reimagine and rebuild their areas together towards a more sustainable future (transitionnetwork.org, 2020).

In addition, one of the most outstanding aspects of Transition Design is embracing a long-term vision, as mentioned in (Irwin, 2015). The framework includes _{The Long Now} philosophy, which challenges people to rethink how they perceive _time (longnow.org, 1996). How people frame their _Now (how long the present is) might lead to a different way of doing business. For example, including a broader range of stakeholders, both current and future generations and taking into consideration complexity of the triple bottom lines of sustainability. The long-term perspective can be a game-changer when it comes to bringing about sustainable changes.

(10)

Transitions can open the gate to sustainable development and allow humans to implement proper interventions in order to achieve such futures (Sondeijker et al., 2006). Therefore, the concept of Transition Design can be beneficial for businesses in shifting toward sustainability, especially in the tech industry, as the framework focuses on foresight study, stakeholders relations, and the design of systems. There is a potential that the framework could be used as a tool to merge AI-driven development with sustainable development. However, Transition Design might be contradicting many of the practices in the tech industry. For example, success is evaluated based on the fiscal year basis, which is apparently not long enough to create sustainable changes in some senses. Hence, it is interesting to look into these conflicting values between triple bottom lines of sustainability in the business sphere and identify the barriers or possibilities to implement the concept of Transition Design in practical circumstances.

1.2 Research Problem

Humanity is reaching the point where the sustainability aspect can not be traded off any longer. This is evident as the climate crisis is getting closer to the tipping point each day (European Environment Agency, 2019), while the majority of people on the planet are still living in poverty (hdr.undp.org, 2018). As pointed by (Vinesa et al., 2020), on one hand AI can be used to empower society to reach sustainable development goals, on the other hand, it can also lead to polarized society and increase inequality when utilized in unethical ways or against sustainable manners (Shelby Fan, 2019). Therefore, a high-impact technology like AI should be an engine to drive a development in which triple bottom lines of sustainability, including social and environmental aspects, are prioritized. Moreover, despite needs of sustainable AI solutions, researchers have mainly focused on the technology’s ethical implications, potential use cases, and the current limitations and barriers that the industry experiences. Thus, conversations and research regarding the clear vision of sustainable futures that AI can enable, as well as ways of transitioning there, are needed.

Through the implementation of Transition Design into the AI solutions development processes, there is a possibility to highlight the missing aspects of triple-bottom lines to the eyes of AI practitioners and decision makers. However, as Transition Design is a new initiative, its practicality has barely been explored. The most cited papers are either the proposal of the concept or the reflection papers from the design educators’ perspective. Only a few action research have been conducted, and only a handful of documentation from the workshop could be found. According to conversation via email with Terry Irwin, the initiator of the concept, there is no so-called ‘case study’ or ‘ success story’ so far. Therefore, it is difficult to say if the concept is applicable to the business sphere.

To investigate whether Transition Design is a viable option towards sustainable AI development, it is important to first understand the current state of the industry and compare the findings with the four core pillars of Transition Design framework, which are (i) Vision for Transition, (ii) Theories of Change, (iii) Mindset and Postures, and (iv) New Ways of Designing. To do this, the researchers looked into the current business ecosystem of the AI industry to draw an overall picture of how the practitioners operate in the live environment and what are the factors that they take into consideration when developing solutions. From there, the findings were used to see the alignments and contradictions between the Transition Design framework and reality.

(11)

1.3 Aim of the Study

Through exploring the current conversation regarding AI and sustainability, and hearing from the practitioners, this paper aimed to investigate the AI solution development processes: perspective and mindset of key decision-makers in the field of AI, as well as what factors and stakeholders are taken into consideration, from both organizational and network perspective. Then, by analysing those findings through the lenses of the four pillars of the Transition Design framework, the study revealed the synergies and the missing links between the theories and practices. Figure 3 below presents the working hypothesis of the paper: by gathering data regarding the development processes from the practitioners working in the AI industry (individual, organizational, and network perspective) and put them through the Transition Design filter (with four pillars of Transition Design framework), the process should reveal aligned and conflicted values between the framework and those factors. The results should lead us to a better understanding on the practicality of the Transition Design framework, as well as the opportunities and challenges to implement the framework in the live business environment of the AI industry.

Figure 3: Working hypothesis (Own creation)

1.4 Research Questions

1.4.1 What are the factors that affect the developing processes of AI solutions, from individual, organizational, and network perspective?

1.4.2 What are the alignments and conflicts between those factors and the four pillars of the Transition Design framework?

(12)

1.5 Disposition

To clarify the objectives of each section of the paper, the following roadmap is made to present the overall outline and how this paper is structured. Readers can use this as a guideline to navigate throughout the paper.

Figure 4: Roadmap of the research paper (Own creation)

(13)

2. RESEARCH DESIGN

This section presents the methodology, research methods, data collecting and analysis approaches used throughout the study. This research is divided into two main parts using different approaches and methods to answer the research questions.

2.1 PART 1: Understanding Where We Are

To get an overview of the current situation in the AI driven industry; how do practitioners work, what influences them and their business activities, who are the key stakeholders that impact the working processes and so forth, researchers used an inductive exploratory approach.

2.1.1 Data Collecting Methods

To explore present discourse of the industry, theoretically and practically, the researchers combined two main methods: (i) document analysis and (ii) semi-structured interviews. Document analysis covered research papers, white papers, peer-review articles, as well as public speaking videos (e.g. TED Talk) for the purpose of exploring the current conversations on AI solutions in relation to sustainability, as well as the factors that affect the development processes of AI in the eyes of researchers, therefore to help this paper formulate the interview questions.

Semi-structured interviews were conducted with practitioners who work in the field of AI solution development, to investigate the processes, factors and the perspectives of people who are key actors in the development of AI solutions with the specific focus on relationships to decision-making.

To assure the credibility of the research results, the interview questions were formed attentively to get the relevant responses that answer the research questions, while keeping in mind the ethical aspects and avoiding leading-the-witness questions. In addition, the interviews were recorded with the consent of the participants. However, the interviews were conducted through a teleconference method due to the limitation caused by the Covid-19 pandemic.

2.1.2 Document Selection Method (Secondary Data)

AI is a fast-moving industry, the same goes to the conversations around it. Hence, it is important to carefully select the documents to ensure that the data is relevant and up-to-date. The researchers used Google Scholar as a main search engine to look for the documents. The search keywords were sustainable AI, machine learning and sustainability, and AI and sustainability. Then, to ensure the validity of the data across the documents, the author/speaker profile, the publisher/organizer profile, country of origin, as well as the published date were taken into consideration. As AI is a new field of study, not many peer-reviewed documents were found. Therefore the researchers chose wildly cited literature from reliable sources as substitutes. In addition, researchers are aware of the possible biases that could affect the credibility of white papers published or funded by companies in the industry. Hence, cross-checking across a number of papers was implemented to reduce such effects on the result of the study.

(14)

2.1.3 Interviewees Selection Method (Primary Data)

To gain the insight data of how AI solutions are developed, the semi-structured interviews were conducted with practitioners whose careers involve the AI solution development process; tech developers; system designers; project managers; fundraisers; founders of tech startups, for instance. Interviewees were required to have been involved with the AI solutions development project within the past 1-3 years, so their experiences were still up-to-date. The researchers are aware of various factors that possibly affect the worldview of the interviewees, such as academic and career background, age range and genders, as well as the nationalities, which could disturb the findings of the study. Therefore researchers attempted to reach out to a diverse group of people. Eventually interviewees with specific qualities were interviewed, which can be seen as one of the limitations of the study.

Apart from personel of the interviewees, the size of organization might affect the type of the projects, the purposes they work towards, as well as the complexity and the number of stakeholders involved. Having that in mind, the researchers selected the interviewees from various kinds of organization, in size and organizational structure. This ensured that the data acquired from the interviews represented the industry as it is. In addition, since the main drivers for the AI industry are based in the US and Europe, to ensure the reliability of the study, the interviewees were selected from AI driven companies based in both regions.

The profiles of research participants are detailed in the table 1 below. Each of them participated in a 30-90-minutes semi-structured interview conducted via an online meeting program. The summaries for each interview can be found in the appendix 15-23.

Table 1: Research participants’ (interviewees’) profile Interviewee

Code Gender Profile Current position Size of the company Location of the company A Male 4 years working in robotic area

(mainly in pharmaceutical industry)

AI developer SME Sweden

B Male Engineering background, more than 10 years experience in tech industry

Research director

Global Sweden

C Male Engineering background, worked for his own tech startup for 3 years before joining the current organization

Head of innovation

SME Sweden

D Male Software engineering background, worked as head of AI in a small company before starting his own AI consultancy and research company

Founder SME Germany

E Male Responsibility for technical

development, engineering and data science teams and IT strategy

CTO SME Sweden

F Male Background in mathematics, working as software engineer in an aircraft company before joining AI consultancy focus on social domain

Technical Lead

Global USA

(15)

G Male Background in software

engineering, more than 15 years of working experience in tech industry, currently in a decision-making position in global cloud service provider

Project

Manager Global USA

H Male Background in engineering physics and mathematics, currently designing curriculum and teaching Applied data science, as well as giving consultation to businesses to use data science as problem solving tool

Founding partner

Global Sweden

I Female Background in Economics and Data science, working as a business advisor in innovation and a consultant in fin-tech

Consultant SME Sweden

2.1.4 Data Analysis Method

By using the _{Gioia methodology (Gioia et al, 2012) as a coding and interpreting framework to} bring qualitative rigor and transparency to conducting and presenting inductive research (cees.leeds.ac.uk, 2016), it is likely to find the common factors that affect the processes, as well as the behaviors, and attitudes behind the solution development (Gioia et al, 2012). Once the common theme of processes, perspectives, and the factors that influence AI solutions development process were formulated, based on the grounded theory, the researchers used this information to answer the research question no.1: What are the factors that affect the developing processes of AI solutions, from individual, organizational, and network perspective? The findings were categorized into 6 sections, in which 48 sub-concepts that reflect the commonalities among interviewees were included.

2.2 PART 2: Looking Through The Lenses of Transition Design

This part of the study answwered the second research question: what are the alignments and conflicts between the development practices of AI solutions and the four pillars of the Transition Design framework? Hence, the descriptive deductive approach was used to analyze the findings from part 1 and respond to the research question.

(16)

2.2.1 Data Analysis Method

Based on the findings from section 4 of the paper and research question 1, researchers analyzed 48 themes from the six categories through the four pillars of Transition Design framework which are (i) Vision, (ii) Theories of Change, (iii) Mindset and Posture, and (iv) New Ways of Designing. The analysis focused on the synergies between the practices (findings) and the theories (framework), and identified alignments and conflicts between them.

To avoid the bias of the researchers, it is important to make the analyzing process as transparent as possible. Therefore an analytical framework was developed to use as an analyzing tool. However, it is also possible that some aspects that are considered during the solution development process were not mentioned by the interviewees, which need to be considered as a limitation to this study.

(17)

3. THEORETICAL BACKGROUND

This section divided into 2 main sections:

- Section 3.1 Artificial Intelligence and Sustainability presents overall discourse of the AI industry. It includes secondary data gathered from conducting document analysis. This is a ground where researchers build a basic understanding in the current situation of the AI industry. Knowledge obtained from the literature has been developed into a framework to design the interview questions, therefore to ensure the coverage of the interviews. In addition, it is used to reflect on the findings in the discussion and conclusion section.

- Section 3.2 _{Transition Design presents the core idea of the design framework and} how it is developed into an analytical framework to analyze the empirical findings (interview results) of this research.

3.1 Artificial Intelligence and Sustainability

Over the last 20 years, people have become more aware of the persistent threats to, and dysfunctions in modern life. All of this has led to the word sustainability to be used as a means of discussing healthier and more fulfilling ways of living. For the business community, particularly for multinational corporations, the U.N. has established sustainability as a code of word for a variety of emerging standards and expectations by which society judges the performance of corporations and by which corporations can judge themselves (Senge et al., 2006). Thus, in our digital and globalized world where information is freely flowing, consumers are leaning towards products and services by socially responsible organizations (Henry Thacker, 2019).

Artificial intelligence, AI for short, is not just a theoretical technology anymore, but one that is influencing our daily life and shifting the job market. By looking into the Future Jobs Reports (World Economic Forum, 2018), as found in (Perisic, 2018) by 2025 an estimated amount of work done by machines will jump from 29% to more than 50%, which will accompany a rapid shift of new labor-market demands and new challenges to address. Furthermore, as AI methodologies are fairly new and maturing, there is a disconnect between the scientific and tech communities from the social sector which leads to the public not understanding ways AI can be used for “good” (Malliaraki, 2019). To combat this, organizations such as AI4AII, The AI for good, DataKind and Data Science for Social Good, are pushing the so-called “AI for social good” movement and have so far proved that partnerships between AI practitioners and social change organizations are possible and can address problems faced in sustainable development (Kush, Ramazon & Mojsilovic, Aleksandra, 2019). All of this has led to a discourse about the ethical implications of AI, where most people agree that an ethical development of AI that benefits humanity as a whole is critical for the technology to take a quantum leap (Fan, 2019).

According to (Vinesa et al., 2020), AI may act as an enabler of 134 targets (79%) across all SDGs, generally through a technological improvement, which may allow overcoming certain present limitations, while potentially impacting negatively 59 targets (35%) across all SDGs. Furthermore, as AI applications for sustainability are at their infancy, only a small percentage of companies mention AI at all in their CSR disclosure (Riffle, 2017). This can be partly due to companies and individuals being eager to embrace opportunities presented by AI, which leads them to appear remarkably unconcerned about the inherent risks that the technology 11

(18)

might possess. (Riffle, 2017) Moreover, one of the greatest barriers to the development of sustainable AI is that neither individuals nor governments seem to be able to follow the pace of technological development. This fact is illustrated by the lack of legislation and regulations regarding the development of AI (Vinesa et al., 2020). Suggestions for changing this trend range from taking a step to establish adequate policy and legislation frameworks to help direct the vast potential of AI towards the highest benefit for individuals and the environment, as well as towards the achievement of the SDGs, to the adaption of accountable, transparent and explainable AI. Lastly, as argued by (Malliaraki, 2019)_{we should aim to be more precise} and keep the identified AI challenges and their scientific and socio-technical basis in close proximity through deep collaboration, dialogue, and serious engagement, to avoid negative impacts from using AI, and to create robust systems that enable the SDGs.

To understand what the general consensus of AI is in regards on how the technology should be developed, its ethical implications, and to further explore how AI can enable and hinder a sustainable future, the remaining of this section will be divided as followed:

Table 2: Key topics of AI and sustainability discourse

Section Aim

3.1.1 Ethical principles and guidelines behind AI development

Help readers understand what are the identified factors needed to develop AI in a way that is beneficial and safe for humanity 3.1.2 How AI can contribute and

hinder the triple bottom line of sustainable development

Provide a summary of how AI can support and hinder sustainable development

3.1.3 Reflection Merge the ethical principle and guidelines in relation to the identified _{benefits and problems AI can has on sustainability}

3.1.1 Ethical principles and guidelines behind AI development

Fears that AI might jeopardize jobs for human workers, be misused by malevolent actors, elude accountability or inadvertently disseminate bias and thereby undermine fairness have been at the forefront of the recent scientific literature and media coverage (Jobin et al., 2019). Because of this, ample debate behind the need to think beyond the technology itself to address the wider implications AI has on society and our environment has grown (Larsson et al., 2019). Thus, as mentioned in (Larsson et al., 2019) there is a growing need for technologists, industry and governments alike to adopt strong principles around fairness, accountability, transparency and ethics. All of this has resulted in national and international organizations having to respond by developing ad hoc expert committees on AI, often mandated to draft policy documents (Jobin et al., 2019). These committees include: the High-Level Expert Group on Artificial Intelligence appointed by the European Commission, the expert group on AI in Society of the Organisation for Economic Co-operation and Development (OECD), the Advisory Council on the Ethical Use of Artificial Intelligence and Data in Singapore, and the Select Committee on Artificial Intelligence of the UK House of Lords (Jobin et al., 2019). As part of their institutional appointments, these committees have produced or are reportedly producing reports and guidance documents for ethical AI, which are instances of what is termed non-legislative policy instruments or soft law (Jobin et al., 2019). Unlike so-called hard law–that is, legally binding regulations passed by the legislatures to define permitted or prohibited conduct–ethics guidelines are not legally binding but persuasive in nature (Jobin et al., 2019).

(19)

The intense efforts of such a diverse set of stakeholders in issuing AI principles and policies is noteworthy, because they demonstrate not only the agreed need for ethical guidance and principles, but also the strong interest of these stakeholders to shape the ethics of AI in ways that meet their respective priorities (Jobin et al., 2019). With that being, challenges regarding the translation of principles into practice and seeking harmonization between AI ethics codes (soft-law) and legislation (hard law) are important next steps for the global community (Jobin et al., 2019) This is highlighted by both (Jobin et al., 2019 and Larson et al., 2019), which mention that no general standards or commonly shared guidelines for ethical AI has been adopted yet. Although this remains true, by looking at (Larson et al., 2019, Jobin et al., 2019, & Michael Chui et al., 2018 ) there is an emerging convergence around the principles of transparent, explainable, accountable, responsable, unbiased, just, fair, private, secure and non-maleficent AI.

3.1.1.1 Transparent and explainable AI

The push for transparent and explainable AI is often referenced as comprising efforts to increase the explainability/interpretability of AI algorithms as a safeguard for accountability and fairness in decision-making (Larsson et al., 2019). This includes acts of communication and disclosure in regards to the development and deployment of AI solutions (Jobin et al., 2019). Furthermore, as pointed by (Larsson et al, 2019) there are seven major challenges associated with transparency in relation to AI, them being:

(i) Proprietorship: _{software and data are proprietary works, which may not be in a} company’s best interest to divulge how they address a particular problem, as may be the case when a product is commercialised and scaled up for commercial purposes. (ii) Preventing abuse (gaming): transparency can be abused to counteract the intended objective and enable abuse or manipulation to gain advantages of the algorithm.

(iii) Competence and literacy: _{the ability to understand and assess algorithms, how} they are applied to data, and their consequences in everyday situations requires competence, sometimes referred to as _{data literacy} or _{algorithmic literacy}.

(iv) Concepts, metaphors and terminologies: the language, metaphors and symbols used to explain AI processes have a direct impact on how we conceptualize and understand such explanations, which, in turn, is related to acceptance and trust. (v) Market complexity: a combination of proprietary arrangements and data-driven markets that can be seen as complex “ecosystems” in which data is brokered and transferred to a number of actors.

(vi) Distributed, personalised outcomes: the outcome of consumer-profiling services that attempt to “personalise” their services, their prices or marketing campaigns–and pose a challenge not least to regulatory oversight to see whether the problem has been solved or not.

(vii) Algorithmic complexity: self-learning algorithms are endowed with a level of independent autonomy that prevents actual oversight of how algorithms solve problems, often referred to as the black box of AI.

3.1.1.2 Accountable, responsible and unbiased AI

Algorithm accountability deals with the delegation of responsibility for damages incurred as a result of algorithmically-based decisions producing discriminatory or unfair consequences (Larsson et al., 2019). Consequences that can arise from data and algorithms that may covertly and structurally discriminate against certain groups, due to inherent biases which are often discovered upon extensive scrutiny, and which can become reinforced by AI systems 13

(20)

(Larsson et al., 2019). Furthermore, it is important to note that to a certain extent, systematic bias may not arise only as a result of the data used to train systems, but also as a result of value-based preferences held by system developers and users of the system (Larsson et al., 2019). Additionally, when talking about responsible, accountable and unbiased AI, specific recommendation includes acting with “integrity” and clarifying the attribution of responsibility and legal liability, if possible upfront, in contracts or, alternatively, by centring on remedies (Jobin et al., 2019). Thus, it is crucial to improve knowledge and understanding of social bias and the relationship between explainability, transparency and accountability with regards to trust and social acceptance of AI (Jobin et al., 2019).

3.1.1.3 Just, fair, secure, private and non-maleficent AI

Malicious uses of AI can threaten not only national security and economic, political & labour market stability, but also individual’s physical, emotional and financial safety (Michael Chui et al., 2018 ). Because of this, a need for just, fair, secure and data private AI is growing, to help monitor or mitigate unwanted biases, discriminations and harms created by AI systems (Jobin et al., 2019). As explained by (Michael Chui et al., 2018) the ability to assuage worries regarding privacy and safety concerns of AI can help the public accept the implementation of AI solutions. To achieve this, one must ensure AI applications have safety mechanisms that comply with existing laws and regulations, and that are safe and responsible for human use (Michael Chui et al., 2018). Thus, it can be said that an active cooperation across disciplines and stakeholders when developing AI solutions is needed, to achieve compliance with existing or new legislation, and to establish oversight processes and practices of the development of AI–notably tests, monitoring, audits and assessments by internal units, customers, users, independent third parties or governmental entities, often geared towards standards for AI implementation and outcome assessment (Jobin et al., 2019).

3.1.1.4 What ethical AI needs

As mentioned previously, no general standard nor regulations for the development of ethical AI has been adopted yet (Jobin et al., 2019). This is a problem which needs to be addressed, as AI solutions become more prevalent in our society, as they carry the potential of being misused by authorities and those who have access to them (Michael Chui et al., 2018). Furthermore, many parts of the globe are underrepresented in the discussion of ethical AI, and the solutions proposed to meet the ethical challenges diverge significantly (Fan, 2019). Suggestions for overcoming these challenges have been proposed, such as; the creation of technical solutions (data privacy mechanisms, data minimization and access control); calls for more research and awareness on ethical aspects of AI; and regulatory approaches with sources referring to legal compliance more broadly and the creation of certifications, and the adaptation of laws and regulations to accommodate the specificities of AI (Jobin et al., 2019). Thus, as mentioned in (Jobin et al., 2019, Michael Chui et al., 2018 & Larsson et al., 2019) more multidisciplinary and interdisciplinary research on applied AI is needed to gain a greater understanding of the challenges posed by AI, and to define what is- and is not- acceptable when developing the technology. This in turn will allow the positive benefits that the technology can offer to become a reality,(Jobin et al., 2019) and ensure that AI is inclusive, safe, ethical, accountable, responsible, unbiased, explainable, transparent and just.

Note: It is important to highlight that the IEEE has initiated a program to establish a certification system of ethical approaches to autonomous and intelligent systems (Jobin et al., 2019)

(21)

3.1.2 AI in relation to the triple bottom lines of sustainable development

This section provides a summary behind 4 papers that discuss the role artificial intelligence has on sustainable development.

3.1.2.1 Applying AI for social good by the McKinsey Global institute, 2018 summary

In their paper “Applying AI for social good”, the McKinsey Global Institute set out to assess the AI capabilities that are currently most applicable to solve sustainability and social challenges, by identifying domains where their deployment would be most powerful. Additionally, limiting factors and risks to be addressed and mitigated if the social impact of AI is to be realized were identified, alongside bottlenecks to overcome.

To do this, they first built a library of use cases of AI for societal good using both social-first and tech-first approaches. Using this approach, they identified 10 impact domains and 160 possible use cases where meaningful problems can be solved by an AI capability, in relation to the sustainable development goals. (Appendix 1) For each use case presented, they identified at least one existing case study, and for the cases where no case study was found, they worked iteratively with experts to identify gaps. Furthermore, a risk profile of the different social impact domains is presented, (Appendix 2) alongside several bottlenecks limiting the use of AI for societal good grouped into 4 categories as seen below:

Figure 5: Identified bottlenecks limiting the use of AI for societal good (McKinsey Global Institute, 2018) To summarize, on one hand AI has many possible use cases which can support the SDG’s and be of benefit for society. On the other hand, the technology brings barriers to overcome, as well as risks to mitigated if positive social impact is to be created. Lastly, as pointed in the paper: _{“While the potential to do good with AI is impressive, turning that potential into reality} on the scale it deserves will require focus, collaboration, goodwill, funding, and a determination among many stakeholders to work for the benefit of society. Many gaps remain. Some are technological and can be overcome if recent rapid scientific breakthroughs continue. Others relate to talent and the shortage of humans who can develop and train these systems and make them work on the ground. We are only just setting out on this journey. Reaching the destination will be a step-by-step process of confronting barriers and obstacles. We can see the moon–but getting there will require more work and a solid conviction that the goal is worth all the effort, for the sake of everyone.”

(22)

3.1.2.2 Tackling climate change with machine learning by Rolnick et al., 2019 In their 2019 paper titled “Tackling climate change with machine learning”, David Rolnick and his fellow researchers, set out to provide an overview of where machine learning - a type of artificial intelligence solution - can be applied to impact the fight against climate change. Throughout the study, the researchers present 13 climate change solution domains, which either focus on addressing climate change through mitigation strategies aimed at reducing emissions, or adaptation strategies aimed at preparing for unavoidable consequences. Furthermore, individual solutions presented in the paper are categorized into the following 3 sections:

● High leverage solutions: Denotes bottlenecks that domain experts have identified in climate change mitigation or adaptation and that are believed to be particularly suited to tools for machine learning.

● Long term solutions:_{Denotes solution that will have their primary impact after 2040.} ● Uncertain solutions: _{Denotes solutions that are risky in one of the following ways: (i)} the technology involved is uncertain and may ultimately not succeed, (ii) there is uncertainty as to the impact on GHG emissions, or (iii) there is the potential of unwanted side effects

Appendix 3 provides an overview of the 13 identified solutions domains, their impact area and an example for each.

In summary, this paper provided a detailed overview behind the areas in which machine learning can be used to tackle climate change. Moreover, as mentioned by the authors “AI is only one part of the solution; it is a tool that enables other tools across fields”. “The sustainable solutions we envision require dialogues with fields outside and within computer science, which will lead not only to novel application domains but also new methodological insights applicable across AI.”

3.1.2.3 How AI can enable a sustainable future by PricewaterhouseCoopers, in collaboration with Microsoft, 2019

In their papel titled “How AI can enable a sustainable future”, PwC and Microsoft set out to answer the question of how we can assess the economic and environmental gains that the AI era can help to harness, and to furthermore, understand better how this new and powerful tool can help to shape our economy and environment against the backdrop of the Anthropocene age. To do this, the researchers made a very preliminary assessment of some of the opportunities that AI can offer for the 4 subsets of sectors that are critical to the economy, environment and natural systems, namely agriculture, water, energy and transport. From their preliminary assessment, they concluded that using AI for environmental applications has the potential to boost the global GDP by 3.1-4.4%, while also reducing global greenhouse emissions by around 1.5-4.0% by 2030 relative to business as usual. Furthermore, AI applications in energy (up to 2.2%) and transport (up to 1.7%) have the largest impact on GHG emissions reduction for their sectors covered. Moreover, as mentioned in the paper, the usage of AI solutions for the environment face many challenges, such as: Lack of awareness, engagement and prioritization amongst people; Changing labor demands; Relative lack of investment; Insufficient focus on delivering responsible AI; Accountability, transparency and bias; Access to citizen develop tools and much more. Lastly, the researches provided recommendations and enablers that can help unlock the potential of AI for the environment, them being:

(23)

(i) Facilitating awareness, value alignment, collaboration and multidisciplinary partnerships through established cross-industry and international standards, region-specific goal prioritization_, collaborative frameworks and public-private initiatives to develop AI for environmental solutions

(ii) Ensuring that we start with 'Responsible AI' and extend this principle approach to include consideration for societal and environmental impact, by ensuring that sustainability principles are embedded alongside wider considerations of AI safety, ethics, values and governance

(iii) Addressing digital infrastructure, data and technology access, and wider complementary technologies by building appropriate cloud-providing infrastructure, facilitating fit-for-purpose data access and annotation, as well as supporting other emerging technology deployment to maximize AI's potential

(iv) Providing opportunities and training for upskilling and reskilling to adapt to sectoral transformations to not only unlock new innovations and scale applications, but to manage and govern AI-based systems to best serve people and the planet for markets and the workforce of the future

(v) Encourage R&D from research to scalable commercial deployment with a focus on connecting stakeholders

To summarize, this paper highlighted some of the positive areas of impact AI can have to enable a sustainable future, alongside some challenges that the technology faces. As the authors put it, “Ultimately, AI will only reach its full potential for society and the planet, if each stakeholder group participates with a shared responsibility to shape the future of AI and of the future systems and business models it underpins.” Thus, it can be said that there is a huge opportunity foregone if leaders, decision-makers, governments, tech developers, companies, academia and NGO’s do not help enable AI innovations aimed towards sustainable development.

3.1.2.4 The role of AI in achieving the sustainable development goals by Vinesa et al., 2020

Ricardo Vinuesa and his fellow researchers, in their 2020 paper titled “The role of AI in achieving the sustainable development goals” looked at published evidence of AI acting as an enabler or inhibitor for an SDG target, by using a consensus-based expert elicitation process informed by previous studies on mapping SDG’s interlinkages. Their results highlight that AI may act as an enabler on 134 targets (79%) across all the SDGs, while potentially impacting negatively on 59 targets (35%). When looking at the social pillar of the triple bottom line of sustainable development, they identified 67 targets from SDG goals 1,4,6,7 & 11 that can benefit from AI-based solutions, and 31 targets which can be impacted negatively, whose consideration must be taken seriously. Furthermore, when looking at the economic pillar they identified 41 targets that can benefit from AI-based solutions and 20 targets in relation to negative impacts. Lastly, when looking at the environmental pillar, they identified 25 targets for which AI may act as an enabler, but which can be undermined by the high energy needs for AI applications, especially if non carbon-neutral energy sources are used. Reference to Appendix 4 for a detailed assessment of the impact of AI on the SDGs in regards to the triple bottom line.

Besides identifying ways in which AI can enable or inhibit the SDGs, the authors also identified research gaps on the role of AI in sustainable development. Some of the identified gaps include: Discovering detrimental aspects of AI, which may require long-term studies; 17

(24)

Encouraging researchers to focus on designing and developing AI solutions, which respond to localized problems in less wealthy nations and regions; Extrapality of evaluating real-world effects of AI solutions and much more.

To summarize, this article described ways in which AI can help enable the SDGs, but also how it can potentially hinder them. Overall the biggest observed trend is the fact that AI at the moment is being developed by wealthy nations without an agreed upon ethical standard, which can lead to a lot of misuses. This is highlighted by the authors, when they said that “One of the greatest barriers towards the development of sustainable AI is that neither individuals nor governments seem to be able to follow the pace of technological development. This fact is illustrated by the lack of legislation and which needs to be reversed. The first step in the direction to reverse this trend is to establish adequate policy and legislation frameworks, to help direct the gast potential of AI towards the highest benefit for individuals and the environment, as well as towards the achievement of the SDGs” , Lastly, as mentioned by the authors,_{“The great wealth that AI-powered technology has the potential} to create may go to those already well-off and educated, while job displacement may leave others worse off”. Thus is it of crucial importance that AI powered solutions are inclusive, transparent, unbiased, interpretable, accountable and guided upon ethical standards.

3.1.3 Reflection

The creation of AI solutions for sustainability are on their infancy stage. Ideally, AI creates sustainable systems that process data sustainably and whose insights remain valid over time, by being designed, deployed and managed with care to increase its energy efficiency and minimize its ecological footprint (Jobin et al., 2019). This in turn, calls for the development and deployment of AI to consider protecting the environment, improving the planet’s ecosystem and biodiversity, contributing to fairer and more equal societies and promoting peace (Jobin et al., 2019 ).

To achieve the development of AI in a sustainable way, we need to be clear about the policy and market reforms needed to make new solutions scale over incumbent practices and systems (PricewaterhouseCoopers, 2019). This is also about managing second order implications and unintended consequences on society and our environment which the technology might bring (PricewaterhouseCoopers, 2019). Thus, we need now more than ever, technologists, industry and governments alike to adopt strong principles around fairness, accountability, transparency and ethics, which need to include and embed, consideration of environmental, societal and economic impacts (PricewaterhouseCoopers, 2019). This has proven to be quite challenging, as no general standard, framework or regulations around ethical AI practices have yet been agreed upon. This can partly be due to long-standing structural socioeconomic and political conditions, a disconnect between the scientific and tech communities from the social sector, (Malliaraki, 2019) or simply because of the complexity of combining different lenses to have a systematic view of the deployment of sustainable AI. Although this remains true, we should aim to be more precise and keep the challenges we identify and their scientific and socio technical basis in close proximity through deep collaboration, dialogue and serious engagement (Malliaraki, 2019).

In brief, the ethical principles of AI and the type of society we would like the technology to enable–whether it’d be a sustainable society, a dystopian society, etc.–remains unclear. With that being said, AI is a technology that has a vast potential to revolutionize our society and help us enable a sustainable future, if developed carefully. Thus, it is important that we start 18

(25)

to design the world we want to live in, in order to understand what are the needed changes we need to make the transition.

3.1.4 Interview Questions Design

After conducting document analysis on AI and sustainability, 5 key aspects are identified as the factors that affect the development processes of AI solutions, listed below:

(i) A need to assess the socio-technical and environmental impacts when developing the solutions, therefore ensuring the positive impact and reducing the negative impact (ii) A need to address challenges and improvement required for the industry to shift towards sustainability

(iii) A need for regulations and working standards for ethical, explainable, and sustainable technology

(iv) A need for collaboration within the tech industry and outside the field

(v) A need for ‘better’ design and development processes to ensure sustainability Based on a tri-fold perspective, these 5 aspects can be categorized into 2 main perspectives, which are organizational and network. However, to answer the first research question: what are the factors that affect the developing processes of AI solutions, from individual, organizational, and network perspective?, the individual perspective is missing from the literature. Therefore when designed interview questions, the researchers add some questions to explore this missing aspect, as detailed in the table below:

Table 3: Interview questions and structure

Individual Perspective

1. What led you to be involved in the field of AI?

2. What do you consider important when developing an AI solution?

Organization Perspective

1. What are the developing processes of the solution? 2. How do you measure success of the project? 3. What do you find challenging about the industry? 4. What do you think can be improved in the industry? 5. What effects do you think AI has on our world?

Network Perspective

1. Do you collaborate during the development process of AI solutions?

2. Who are the stakeholders you consider during the development process of AI solutions? 3. Are you aware of any regulations/rules that affect the development of AI solutions?

However, these questions were evolved and changed during the study as the patterns among the interviewees’ responses were discovered. On the other hand, some additional questions were added to find out more about specific aspects that occured from specific interviews.

(26)

3.2 Transition Design

Sustainability is highlighted as one of the most challenging problems for humanity in the present time. As the world is one interconnected and interdependent system (Capra, 1996), sustainability is a complex problem in itself. To cope with this wicked problem, design methodologies are being adopted throughout various fields and industries. Similarly, the role of designers as a problem-solver becomes more and more visible (Irwin, 2015). Several design approaches emerged in response to the need for a more sustainable future (Sumter, 2020; Irwin, 2015). Some remain their focuses in the production-based economy but try to minimize the amount of resources extracted from nature, for example, circular economy (MacArthur, 2013), ecodesign (_{Karlsson and Luttropp, 2006), nature-inspired design (de} Pauw et al., 2010), and design for low resource settings ( Jagtap, 2013). Some move away from physical products or solutions to designing human experience/interaction: service design (Zomerdijk and Voss, 2010), participatory design (Muller, 1993;Schuler and Namioka, 1993; Spinuzz_{, 2005;), and design for social innovation (Chick, 2012), for instance. While} most design approaches still operate in the consumer market domain, design for social innovation shifts the problem context into socio-economic and cultural domains (Irwin, 2015).

3.2.1 The Origin

In addition to what’s mentioned above, (Irwin, 2015) suggests a new design approach to deal with complex sustainability challenges. Inspired by the_{Transition Town Movement started by} an English activist, author, and environmentalist Rob Hopkins, _{Transition Design} advocates ‘design-led societal transition toward more sustainable futures’ (Irwin, 2015). In _The Transition Handbook: from oil dependency to local resilience (Hopkins, 2014) uses the peak oil and climate change dilemma to emphasize the need for a transition to a more sustainable alternative economy. By embracing _{Permaculture philosophy he encourages a grassroots} movement to stand up against the collapsing ‘business as usual’ and seek for local, resilient solutions. (Hopkins, 2014) recommends _{Six Principle of Transition Model} , simplified and developed from Permaculture:

(i) Visioning: Transition, which means to move toward something, can only happen when the end goal is clearly defined. Therefore to imagine realistic and desirable situations/outcomes is crucial.

(ii) Inclusion: The Transition approach aims to facilitate inclusive dialogues and collaboratively bring about the solution.

(iii) Awareness-Raising: Clear communication regarding the issues/problem situations is necessary. The Transition Movement provides a clear, accessible source of information hence people have enough information to join the discussion and co-create the solution.

(iv) Resilience: Is one of the most highlighted qualities throughout the Transition Handbook. There are three ingredients needed to rebuild the resilient system in the modern society (Levin, 1999 as found in Hopkins, 2014); (a) embracing diversity and realizing that there is no ‘one-size fits all’ solution (self-dependency); (b) enhancing modularity to allow each part of the system to self-organize in the event of shock (decentralization); and (c) emphasizing the necessity of feedback loop in order to bring the consequences of action to be seen clearly (localization).

(v) Psychological Insights: The transition approach provides a safe space for people to speak up, creates a sense of belonging and power to encourage people to act against the problems, instead of leaving them in powerlessness, isolation, and overwhelmedness which are threads to generate action.

(27)

(vi) _{Credible and Appropriate Solutions: Transition approach encourages people to} work on the community level, which provides achievable solutions of a credible scale. Holding these principles as a base, Transition Design has been developed into a design framework that wraps itself with long-term visioning, lifestyle-oriented, and place-based approach, and always considers the natural world as a key stakeholder while seeking solutions within the socio-economic and political domains (Irwin, 2015).

3.2.2 Transition Design Framework

To bring the idea into practice, (Irwin, 2015) develops the _{Transition Design framework}and suggests a number of thinking tools to be used throughout the design process. The framework contains 4 key elements: (i) Vision for transition, (ii)Theories of Change, (iii) Mindset and Postures, and (iv) The New Way of Designing, which will be demonstrated further in the following part.

Figure 6: Four pillars of Transition Design framework (Irwin, 2015) (i) Vision for transition

“Visioning is crucial; it creates spaces for discussion and debate about alternative futures and ways of being and it requires us to suspend disbelief and forget how things are now and wonder about how things could be.”

(The value of future casting and envisioning alternative futures is underscored by Boaventura de Sousa Santos in The Rise of the Global Left: The World Social Forum and Beyond, 2006)

Before the provocation of Transition Design, there were roughly four types of design (Richard Buchanan): (i) Communication design–symbolic and visual communication, (ii) Industrial product design–material objects, (iii) Service design–activities and services, and (iv) System design–complex systems and environments for living, working, playing, and learning. All these design approaches are articulated as the process of going from a problem state to a solution state (e.g. Doblin 1987; Munari 1981 as found in Peter Scupelli, 2015). In other words, they focus on solving the current problems, without imagining the end goal of the 21