All Tech Is Human Mentorship Profile: What is ethical AI? A multidisciplinary thought exchange
The following blog post is a contribution by All Tech Is Human mentors Leah Junck, Paul Schütze, and Rüya Tuna Toparlak. All views expressed are those of the contributing authors and do not necessarily reflect the views of All Tech Is Human.
We are three emerging scholars from different fields interested in questions pertaining to digitalisation. Throughout the last year, we came together to discuss the ethics and impacts of AI in a broad sense. However, having researched AI and its manifestations at universities in Germany, Switzerland, and South Africa, our attention naturally turned to AI in European and African contexts. During our discussions, two things became particularly clear to us:
1) Envisioning an ‘ethical AI’ is not a straightforward matter and 2) envisioning it must alwaysbe an interdisciplinary effort. With these thoughts consistently resurfacing in our discussions, we decided to bring together our individual perspectives and research backgrounds, namely anthropology, ethics and philosophy, and law, within this commentary. What follows is a glimpse into the main insights and exchanges of our collective thinking.
The first fundamental observation, which we all shared, was that current forms of AI are deeply embedded in social, economic, and material structures. The common AI applications we all know and talk about, such as ChatGPT, social media, or search algorithms, in fact have to be viewed in virtue of their embeddedness within these larger structures. ChatGPT, for example, is not just an immaterial chatbot, but recent public debate has shown that its use consumes massive amounts of energy, it is built on the exploitation of data workers and in its outputs it reproduces discriminatory societal structures. This shows that this seemingly immaterial and singular algorithm is inseparable from the social, economic and material structures which usually remain hidden in the background.
The take-away is clear: AI can never just be grasped as an algorithm or a technology, but it always refers to a variety of elements, to a multidimensional and global network. Kate Crawford, for instance, calls this the “megamachine” of AI. With this, she emphasises that AI is at once an infrastructure and an industrial formation, including capital, labour, politics, and culture (Crawford, 2021). All of this comes together and produces what we currently perceive as AI. Therefore, when talking about an ethical or responsible AI, considering this megamachine is imperative. Responsible AI cannot only refer to well-designed technologies or algorithms, but it has to address the entire machinery of AI. This also means to acknowledge the click and data workers enabling applications such as ChatGPT, to address the environmental costs of AI technologies, or to question the interests of the big tech companies that drive AI’s continuous development. Everything that lies beyond the immediately visible of artificial intelligence must not be pushed aside, but instead needs to be brought to the fore and addressed head on.
From this material and socio-economic perspective, discussing responsible AI requires us to question the very dynamics and conditions current tech developments are built upon. Such a view raises not only questions about biases, costs and deceptions embedded in AI systems. But it puts up for discussion our current economic and societal organisations at large. Do we really want to continue building technologies that rely on the global exploitation of workers?
Do we want to build ever larger models, fuelling ‘better and better’ algorithms, which constantly consume the amount of electricity equivalent to cities or small countries? Do we grant big tech corporations and tech moguls such as Jeff Bezos or Elon Musk the power to shape public discourse about AI and stir the debate in the direction of AI hype and fear? And how do we work towards producing counter-narratives of what our future might look like?
These questions bring us to a second point: When we look at issues of AI from a global north perspective, the conversation is largely shaped by a few very powerful players and occupies a huge space in our zeitgeist. But the issues, questions, and problems that accompany AI are not the same for the world at large. The regions perhaps most exploited by the implementation of AI while also serving as a testing ground for experimental applications might have very different interpretations of whether and when technologies are useful. At least this was the impression Leah had while working on her project on AI in healthcare in Mozambique, where she found the discourse of transnational investors to be discordant with how healthcare workers on the ground were talking about technologies. This is also reflective of the fact that curiosity in technology studies rarely reaches beyond general areas of application and into more layered explorations of meaning-making. Even though there is now a growing scholarship concerned with the implications of technologies, it is not yet a consolidated effort that involves equipping experts with a broader understanding of the social and environmental implications of the current tech hype, or one that strategically connects those schooled in different disciplines (with the exception of some pioneer projects).
How we learn to talk about ethics in a particular scholarly field has a significant impact in how progressive we dare to be in envisioning digital futures and our role as researchers in it. In the discipline of anthropology, for instance, ethics are a big deal! And rightfully so, given that anthropologists typically do research with people in the intimate spaces of their day-to- day lives and over an extended period of time. Given how complex we and our needs as humans are, a lot can go wrong. Relationships with research participants have to constantly be renegotiated, which is why self-reflexivity is praised as an essential skill in the discipline: it has to be showcased when seeking permission to do research or when defending findings to fellow anthropologists looking for ‘weak spots’. More importantly though, ethics in this practice is having a sense as a researcher that the inherent power-imbalances between investigator and subject are navigated in a way that lets us sleep well at night. Where anthropologists can contribute to attempts to grapple with rapid technology changes, then, is with an understanding of ethics that does not just serve to create a protocol of propriety and ‘correctness’. But that instead understands ethics as a negotiation between individuals and powerful structures at play. Translating between them is an important skill in creating a collaborative, responsible scholarship on digital technologies.
For us, these two initial observations around AI and ethics (their materiality and multitude of meaning) are followed by a third point – a challenge if you will: How can we transform these insights into action or into meaningful practice? Here, we particularly turn towards the discipline of law. Both law and ethics try to guide human action. Law is the functional equivalent of morality (Luhmann, 2008) . It generalises and stabilises our behavioural expectations. Legal rules should then ideally reflect our societal priorities. This is where we again find our first observation: the global materiality of AI. In an increasingly globalised world, AI applications and its effects are said to be all-encompassing; but neither ethics nor law is. So how can we come up with a baseline of ethical guidelines for AI, that would be applicable around the world while also respecting the diverging priorities of different societies?
In 2021, UNESCO published the first ever global standard on AI ethics. This document is essentially a recommendation that approaches AI ethics as a multicultural and evolving framework of interdependent values. The document sets four core values to accept or rejectAI technologies: 1) respect and protection of human rights 2) just and interconnected societies; 3) ensuring diversity and inclusiveness; and 4) respecting the environment and the ecosystem. The document is helpful in approaching the issue with values which seemingly can be agreed upon by most. We would like to stress that while these frameworks can be useful in responding to injustice, but that they cannot substitute for more qualitative efforts to find out what AI justice might mean to people in specific contexts. Giving voice to these people through the different scholarly tools at our disposal is an important test of our time.
Another concern is that the nature of the recommendation means that it does not have a binding effect. To make rules binding, we need to turn to law.
Herein the question becomes, how can we come up with a baseline of legal rules for AI, that would be applicable around the world while also respecting the sovereignty of various jurisdictions? This question reminds us of the General Data Protection Regulation of the EU.
Few regulations have impacted global digital economy more. When the EU created the GDPR, they extended its scope not only to their member countries, but to all corporations that are offering goods or services to the people in the EU. With its high fines and extraterritorial application, the GDPR cemented the European Union’s role as the global regulatory power, particularly on digital economy (Bradford, p. 132 ff., 2020). Currently, the EU is drafting a specific AI Act. This shows the need for rules governing how to create, maintain, and operate AI. We need to know whom to direct liability when something goes wrong. We need to be reassured that the humans - especially those who are disproportionately affected by this global exploitation - do not carry the risks inherent in AI applications. At the same time, this need opens a critical conundrum: How ethical is it for the EU to position itself as the ultimate rule maker or harbinger of the best standards in AI? What about the global and diverse needs of different societies and communities?
We want to emphasise once more that answers to the concerns around AI practices are not trivial. Nor should they be seen as a mere matter of protocol. Our discussion shows that perspectives can only be partial: We learn from political and social philosophy that AI is a vast socio-material network which has global effects. We learn from anthropology the different meanings of ethics and the plurality of perspectives and experiences that need to be considered. And we learn from law that while we need tools to put the many insights into action, there is a variety of issues that need to be considered before action can even be taken.
Thus, in a forward-looking discourse all these viewpoints necessarily need to go together in developing a dynamic, kaleidoscopic vision of what the future might look like. The discussion on an ‘ethical AI’ and on turning insights into practice is an ongoing challenge which requires continuous work and will always remain incomplete. This may be the most integral insight we have gained from our joined thinking.
References
Bradford, A. (2020). The Brussels Effect: How the European Union Rules the World. Oxford University Press USA - OSO.
Crawford, K. (2022). Atlas of AI: Power, politics, and the planetary costs of Artificial Intelligence. Yale University Press.
Luhmann, N. (2008). Die Moral der Gesellschaft (Orig.-Ausg.). Suhrkamp.
Blog Post by Leah Junck 1 , Paul Schütze 2 and Rüya Tuna Toparlak 3
1 Dr Leah Junck, independent researcher
2 Paul Schütze, Research Assistant, Ethics and Critical Theories of AI, University of Osnabrück
3 Rüya Tuna Toparlak, LL.M., PhD student and academic assistant at the University of Lucerne
Bio: Rüya Tuna Toparlak is a registered attorney at law with the Bar of Istanbul. She is currently a doctorate student and an academic assistant at the University of Lucerne. She researches the intersection of law and digitalisation. Her dissertation focuses on AI, social robotics, and legal subjectivity questions. Her broader research interests include disinformation, transparency, and platform liability questions, particularly through an intersectional legal gender lens.
Bio: Paul Schütze is a research associate in the Ethics and Critical Theories of AI group at the University of Osnabrück, Germany. In his current research he focuses on the social and material structures of AI and their connection to the climate crisis. In particular, he works on the textures of power and subjectification in the age of Big Data and AI. He has a background in critical social philosophy, affect theory, and philosophy of mind.
Bio: Leah Junck is an anthropologist fascinated with questions of what the integration of computational technologies into peoples’ lives means for our ability to relate to one another. This interest was fueled through her work on Tinder dating applications, the role of social media in neighbourhood surveillance in South Africa, and understandings of artificial intelligence in healthcare in Mozambique. Her work in journalism, policy research and collaborations across disciplines convinced her that storytelling is key in dealing with accelerated change.
