Context matters

3 min read 

There is growing debate in academia and policy circles, particularly with a focus on the Global South, that the pessimistic narratives about AI and digital labour common in Western media do not always capture experiences elsewhere. While poor conditions in outsourced digital work are often condemned, some women in these contexts may experience such work as empowering rather than exploitative. Rather than framing this as a “West versus the rest” divide, I favour a more contextual perspective that moves beyond both geographic and ideological binaries, especially when examining AI’s role in public sector institutions.

I want to argue that context and conditions matter regarding both the perceptions and experiences of/with AI. To make my point I use the situation of Swedish cultural heritage institutions(CHIs) in which my colleague Anna Foka and I, together with several postdocs and PhD students, have been conducting research on bias mitigation and diversity as inclusion. Note here that 'diversity' is the term that has succeeded 'gender' in many public and private institutions as the preferred term for dealing with inequalities and systematic disadvantage. I will say immediately that in research I have conducted in a different arena (https://nordwit.wordpress.om) it became very clear that - at least in some contexts - 'diversity' has become a way of NOT naming specific inequalities, thereby making it difficult to make any claims on behalf of specific inequalities to do with race, gender or divergent abilities, for example. In other words, 'diversity' displaces engagement with specific inequalities, making it easy to ignore these.

To move on to my main point: Sweden is a western country renowned for its equality policies and politics. It is also a country that occupies a large geospatial territory (it is roughly double the size of the UK) with a very small population, about 10.5 million. This means that the population density in the country is low (26 per km² in Sweden, 243 per km² in Germany, 488 per km² in India). Three important points arise from this: a) you often cannot make the economic argument for something in Sweden because of the dispersal of the population - in other words, sociopolitical and cultural arguments are more significant; b) due to the dispersal of the population, digitization and digitalization are very significant as they facilitate connection among the dispersed population (so digital health care is very well developed, for example); c) institutions and data are very limited because of the limited size of the population. Additionally and importantly, Sweden is a very static population where people love to stay in the place where they were born and raised and population mobility is rather limited. The implication of the latter is that once people have a job they are unlikely to leave or change this. In the context of cultural heritage institutions (but also in other institutions) this means that many people may spend 40 years or more in the same workplace.

We are living in an era of technological transition, progressing at varying speeds in different contexts. Whilst industry and the private sector forge ahead at, at times, alarming rates regarding the development of AI, the public sector (the largest employer in Sweden) lags significantly behind, as it is fiscally and politically incapable of moving at the same speed as industry. In addition, people in decision-making positions today across the public sector who are older than 40-45 grew up in an era when there were no or few computers, and AI was just a pipe dream. But these are now the people required to implement AI in their institutions.

This requirement, however, has also been accompanied by the increasing withdrawal of support for continuing professional education, and the need for appropriate training to disseminate knowledge and know-how regarding technology and AI developments. Further, in Sweden there are relatively few cultural heritage institutions and they have specialized collections, for instance on Sweden's indigenous population, the Sami, which may be too limited to allow appropriate AI training. Between the problematic of a lack of relevant AI expertise among heritage institution staff and limited material on which to effect machine learning, CHIs - if they digitize and digitalize collections - may be forced to buy off-the-peg programs that were trained on datasets that differ from what the CHIs contain and may therefore be unsuitable and/or mislabel items in collections. A well-known example of this is the mislabelling of men in 18th century portraits as women because they sported long curly hair (wigs) and wore jewels.

The upshot of all this is that context and conditions (for the implementation) of AI matter, and their differences create diverse environments for the possibilities that AI offers. The 2024 AI Index Report (https://aiindex.stanford.edu/report/) provides sobering statistics in this context. It shows that the number of AI job postings as a % of all job postings remains very small (for 2023, 1.62% in the USA, 1.31 % in Sweden, 0.81% in Germany). The report itself addresses gender very narrowly. And it is not alone in this. The TBI's 2024 Governing in the Age of AI report leaves gender completely unaddressed. This suggests that we have a long way to go in every respect. The rise of gender conservatism as part of the rise of right-wingism, the climate considerations that the use of energy in the development of AI in terms of graphic processing units brings with it, and the changing political orders that we observe (e.g. the merger of technology with politics in the USA in the person of Elon Musk) create new contexts for questions of technology, employment and wellbeing which will have different answers in the diverse contexts in which we live.