The fourth session at Sustainability Frontiers discussed Digitalisation and Sustainability Transitions, and will be the third in the series of blog posts about this conference written by Laila Mendy, PhD student at NRHU, Uppsala University. The first can be read here and the second post can be read here.
Somya Joshi moderated a conversation between Stefan Daume, Maja Essebo and Andrea Owe. Joshi introduced the hope and parallel concern about the relationships between technological development and innovation and the environment. She then invited each speaker to present their approach to this issue.
Stefan Daume began on the idea of digitalisation from the perspective of disruptive technologies and AI for Sustainability science. He pointed out the meta-quality of the conversation and how the online conference structure can facilitate fantastic connections around the globe more quickly and without attributed transport emissions.
Daume’s research spans the mature forms of technologies to the niche and cutting edge. He suggested that the internet acts as a mature tool for sustainability and pointed towards Greta Thunberg’s social movement enabled through the availability of social media. He considered open access research as another form.
To the more cutting edge innovations, he introduces the idea of AI, block chain and crypto. AI, he argues, has been around since the era of Alan Turing. Today’s machine learning, however, has the gigantic accessibility of computing powers available in the cloud. That said, these systems are not just virtual. He explained that there is an embodied and material element as these consume resources to run and are often designed to serve existing dominant interests.
The second intervention was given by Maja Essebo, who came from the point of storytelling and looking at how algorithmic technologies are not only disseminating stories but also storytelling themselves. A side note was given here, though, is that this is more about machine learning than AI.
Looking at the elements of storytelling you have character creation, the plot, events connected by relationships, place and more. Character creation is where machine learning technologies get particularly interesting. These technologies are interested in a group of people to understand what common characteristics occur within certain sets of the population. An example was given that individuals who liked KitKat pages were more likely to join anti-semitic hate groups. Such a connection would not necessarily be imagined by a human, but a machine noticed this trend.
Her conclusion was that algorithmic processes are performing character creation by noticing these trends and connections. But she is not sure how. These technologies are finding these connections in a black box. They are given input instructions such as “go look, go see, go search”, but how they solve that problem is not really clear. What the creator thinks is a fairly clear task can have wider interpretations by the machine.
There are two reactions from Sustainability scientists when they learn about the scale of stories these technologies are creating: interest and disgust. They might want to stay away from these technologies but, she concluded, the algorithms will not stay away from them.
The final presentation was given by Andrea Owe, who explained that the biases of humans are fed into AI. She started from the point that sustainability has complex socio-political associations and is more than a natural or technological issue. Much sustainability science is done from an anthropocentric approach, she claimed, which posits that humans are distinct and more important than the rest of the natural world. With that she argued that AI discourse needs to break away from those systems of thought.
Owe’s concern was that non-human animals and the natural environment is neglected in the field of AI ethics, which as of yet has mainly focused on the social implications of AI structures. The non-human is not considered in concepts of justice, but only appear to some extent as beneficiary of sustainability. This is a significant blind spot when developing AI for sustainability as it is neglecting how the environment is considered in those systems.
The presentation concluded by reiterating the concern in terms of environmental footprint: when selecting which AI to pursue a footprint might indicate that certain forms of AI should not be developed in the first place. This requires full and thorough life-cycle analysis during development stage. She finished by reiterating that while sexism and racism is perpetuated by AI and is, to some extent, being addressed, the abuse of nature must be addressed in research.
After the panellists presented their opening statements they were given the opportunity to respond to a number of questions. Much was said about the idea of agency in story telling: that it is not full agency, but that there are aspects within the black box issue. Daume continued by explaining that inbuilt biases certainly exist and in multiple forms including training data bias, transfer bias, and also interpretation bias by the people using the AI. Owe explained the challenge of interjecting ethics in systems we are still struggling to grapple with. She summarised in ways that echoed the decolonial scholars from the first session: ethics is not a to do list, and ethics washing is widespread. For ecological and environmental ethics in AI, the research is very new and underdeveloped.
A second theme of questions emerged from the audience Q and A which asked about whether – and how – AI can change human-nature relations. Owe pointed to incredible work being done by indigenous communities around the world who are developing their own AI based on their own data.
My own question was raised at this point, which concerned the reconciliation of technological acceleration with the degrowth of the economy. Owe suggested she was pessimistic about decoupling technological innovation from environmental impact. Essebo, however, offered a more optimistic framing, suggesting that these technologies can enable economic transitions at a more rapid rate. Daume reiterated this in terms of broadening access and participation but cautioned that the platforms on which these tools rely are provided for by companies who are not motivated by sustainability. They also, he said, have long been recognised for their emissions but there are now some transitions to more sustainable practices such as submerged data centres.
This was the third in a series of blog posts about the Sustainability Frontiers conference. Click here to read the fourth on Inner Transformation and Imaginaries.