Human Values and AI – A ViC Workshop

AI hatchlings
AI hatchlings in 3DP production. Photo by sf@vic

The ViC team has been invited to deliver a workshop on Human Values & AI  at CIRN, Prato, Italy, October 2018.

This workshop  takes a deep dive into human values, examines how they work, and what structures they may exhibit. It uses some of the new tools developed for the AI & Ethics seminar series just held at Alpbach and covers some of the questions explored in our previous post on AI & Ethics.

Full Workshop Proposal

Abstract

AI has seen a massive and rapid development in the past twenty years. With such accelerating advances, concerns around the undesirable and unpredictable impact that AI may have on society are mounting. In response to such concerns, leading AI thinkers and practitioners have drafted a set of principles – the Asilomar AI Principles – for Beneficial AI, one that would benefit humanity instead of causing it harm.

Continue reading “Human Values and AI – A ViC Workshop”

Measuring Values in SE

Values as mental representations.
FIgure 1: Values as mental representations to be studied on three levels: system (L1), personal (L2), and instantiation (L3) level.

“Measuring Values in Software Engineering”  is our latest peer-reviewed  work. It has been accepted for presentation at the 12th International Symposium on Empirical Software Engineering and Measurement (ESEM2018), 11-12 October 2018, Oulu, Finland.  Accepted on: 13th  August 2018 . Pre-print version.

Abstract

Background: Human values, such as prestige, personal security, social justice, and financial success, influence software production decision-making processes. Whether held by developers, clients or institutions, values are both highly subjective and deeply impactful on software outcomes. While their subjectivity makes some values difficult to measure, their impact on software motivates our research. Aim: To contribute to the scientific understanding and the empirical investigation of human values in Software Engineering (SE). Approach: Drawing from experimental psychology, we consider values as mental representations to be investigated on three levels: at a system (universal, L1), personal (abstract, L2), and instantiation level (concrete, L3). Method: We design and develop a selection of tools for the investigation of values at each level. As an example, we focus on the design, development, and use of a Values Q-Sort built by mapping Schwartz’s universal values model onto the ACM Code of Ethics. Results: Q-statistic sorts work with smaller samples than R-statistic surveys; from our study with 12 software practitioners, it is possible to extract 3 values ‘prototypes’ indicative of an emergent typology of values considerations in SE. Conclusions: The Values Q-Sort combines the extraction of quantitative values prototypes that indicate the connections between values (L1) with rich personal narratives (L2) reflective of specific software practices (L3), and as such, it supports a systematic, empirically-based approach to capturing values in SE.

Values and Workplace

How can we bring to the open and address the personal, institutional and political values tensions manifesting in our workplaces ?

Vic team writes about the values tensions observed in academia. Below is a re-post of the original article posted in  the ACM interaction magazine blog on 25th June 2018

Values Tensions in Academia: an Exploration Within the HCI Community

Figure 1. Wish you were here – by @_JPhelps

February and March 2018 saw the largest ever industrial action in the UK’s higher education sector. Whilst the cause of the strike was changes to the USS pension scheme, the picket lines were sites for conversations about many other issues within academia. Whether it was dissatisfaction with the corporatisation of universities, the precarious working conditions of early career researchers, or over-work, there was a clear sense that the values held by those striking were in sharp contrast with the realities of university life. The ‘depth of feeling’ was often bitter and angry, and the frustration with today’s higher education system palpable.

Whilst many reported a loss of trust in the system and in their own institutions, fresh hope and renewed energy came from activities such as the teach outs, open teaching and discussion sessions outside the campus. These initiatives offered concrete examples of different ways of engaging with learning and research across disciplines and roles; ideas were proliferating like a “thousand butterflies”. Many now feel that very broad bridges are needed to start filling the values gap that has manifested itself so clearly during the strikes; as Prof Stephen Toope, Cambridge University Vice Chancellor, puts it “the focus should be on what values our society expects to see reflected in our universities, not just value for money”.

Figure 2. Schwartz’s values model. Adapted from [(Schwartz 2012)]
From the HCI community standpoint, a similar values tension was captured by a survey carried out as part of the ‘Values in Computing’ (ViC) workshop at CHI2017. With just over 150 respondents, the survey explored views about the values driving HCI research at a personal and institutional level. The survey was designed around Schwartz’s values model (Figure 2) and tried to capture relationships (i.e. lines of friction) within and between the personal and institutional values held by the HCI community. Although the survey was exploratory and the sample may not be representative of the whole HCI community, the numbers did show tensions within the community.

Overall, most respondents felt their values matched their institution’s to some extent (57% of the respondents). However, almost a third reported that their values either did not match (26.5%) or did not match at all (6%) those of their Institution. The survey also asked respondents to rank a list of options according to which were most highly valued in their work; this is where the values tensions became manifest. As the bar charts in Figure 3 show, the top three most highly ranked options were ‘making the world a better place’; ‘competence and intellectual independence’; and ‘relationships with colleagues, students and partners’. These statements were designed to represent Universalism, Self-Direction, and Benevolence in Schwartz’s values model and followed previous research guidelines. Positive societal impact, autonomy of thought and meaningful relationships were thus the things that these computing professionals most valued about their work. By contrast, ‘financial recognition’ (Power) was the least valued.

Figure 3. Personal and organizational values ranking.

They felt that their institution most highly valued ‘financial success’, ‘international prestige’ and ‘league tables/rankings’. All these three options belong to the Power values group. By contrast, the bottom three options were ‘making the world a better place through work, research and teaching’, ‘staff relationships with colleagues, students and research/work partners’, and ‘supporting the well-being of staff, students and partners’. Thus, the things that the respondents most valued – with the exception of intellectual autonomy – were seen to not be highly valued by their institutions.

The implications of this friction between personal and institutional values cannot be ignored and deserve further attention. Even if this tension may be, to a certain extent, ‘perceived’ or ‘inevitable’ or both, the widening of the values gap may have problematic consequences. For example, recent research and extensive media coverage worldwide suggest high levels of stress and mental health problems within academia. However, the emphasis of these studies is often on the temporal and mental burdens created by the demands of the workplace, and the need for raising awareness and promoting self-care (i.e. through apps and physical activity).

Something that isn’t often talked about is whether the values tensions may have health and well-being implications, and the need for digging deep into the root causes of these tensions before defaulting to self-care coping mechanisms. This may be particularly the case in the HCI community, as many of us not only grapple with personal challenges, but also with the challenges of a much deeper and broader “existential crisis”. This is especially important because much of HCI research focuses on designing and developing digital technologies that can change people’s lives rather than examining how digital technologies come to life. We need to look into values tensions not only for the end-users and broader stakeholders, but for us – researchers, educators, designers, and developers.

To this end, we argue that a better understanding of values is needed, especially when it comes to computing technologies. From a research and practice perspective, this means to build on, but also go beyond, the substantial corpus of research in Ethics and the well-established research field of values sensitive design (VSD).

Our question for the HCI and broader computing community is how to bring to the open the personal, institutional, and political values tensions manifesting in our workplaces (i.e. academia, research). In other words, how can we support the next generation of computing professionals with the deliberative, technical, and critical skills necessary to tell the difference between what is worth pursuing from what is potentially harmful to self and society? And how can we create and support institutions where this civic purpose can flourish?

Thank you!

This work is part-funded by the Engineering and Physical Sciences Research Council UK (Grant number: EP/R009600/1). Warm thanks go to our project partners and to the CHI2017 ViC workshop participants, who have jointly shaped the vision and direction of this research. A special mention also goes to the thousands of conversations had with colleagues and students within our School and across campus. More information about ViC and related work can be found at www.valuesincomputing.org