Katha Dornenzweig

This text was written because the public debate on assessing AI consciousness is urgent, but often does not reflect the available scientific tools and insights. Specifically, it is often forgotten that artificial agents are not the first non-humans in whom we have attempted to scientifically assess consciousness; doing so in non-human animals is meanwhile an established and successful field in which substantial progress has been made, significant consensus has been reached, and increasing protections have been possible. Many lessons learned with non-human animals have startling implications for AI, and transfer, with non-human animals making for a far more compelling benchmark than humans when assessing minimal consciousness, consciousness occurring at different scales, or under radically different conditions. Where the novelty AI brings pushes our existing understanding to its limits, the specific differences still allow us to ask precise questions and find a scientific and ethical way forward. This text will break down the relevant research in a way that people from different disciplines and walks of life can follow. This topic matters for people who have jobs that are not being a researcher full time, and who deserve to know what is going on. It will start (1.) by explaining the concept of sentience as the most ethically relevant core of minimal consciousness. It will then (2.) give reasons to take the scenario of sentient AI seriously, showing the role of anthropocentric bias, industry interest, and mystification of consciousness in denying the possibility as obviously unthinkable, and paralleling it with historic underestimation of sentience in other non-humans. More importantly, it will highlight that sentience is a trait that has developed multiple times independently in nature due to its highly efficient, functional enabling of rational responses to novel problems an organism tries to solve, which makes unintentionally recreating it due to the advantages it would give industry not implausible.
 Then (3.), it will indicate the good reasons we still have to be sceptical of AI sentience in current models. Beyond issues of gaming, grounding and embodiment, it will list legitimate functional concerns behind the idea of life mattering, the difference between intrinsic causality and symbolic representation for consciousness, the relevance of dynamic processes and agency, the issues of dispersal, and different implementations of recurrence. It will also point to the alarming risk of anthropocentric bias and the design of current models systematically leading to general anthropomorphic overascriptions of sentience to AI, and to the rise of AI romance, AI psychosis, and AI cults, which are becoming a significant social factor. Whether AIs deserve protection should depend on whether they actually are sentient, which we should assess to the best of our scientific ability, not on whether they are eloquent and attractive enough to have sentience ascribed to them. Dealing with the analysis responsibly should be the mandate of democratic society ethically debating in citizen’s panels, not of powerful industries with no democratic mandate or independent ethical review. Hence, (4) highlights how crucial and urgent interdisciplinary cooperation and translation and dedication of resources will be to recognising AI suffering (and its absence) and actually passing protective measures, and calls for a united front between the public and the scientific community against industry determining the development of technology when there are such high ethical stakes, seeking allies.

This article evaluates experiments seeking to teach human language to various non-human primates and birds, with a focus on the agency, self-expression and resistance to their own predicament that became apparent in the experimental subjects once communication was genuinely attempted with them, and the anthropocentric framing in which it was received and devalued in the general perception. These experiments, the problematic assumptions behind them and the remarkable results deserve far more critical scientific and ethical analysis than they were given; the full scope of what these animals proved capable of is still underestimated.

This transdisciplinary anthology, a project of the "Chimaira - Working Group for Human-Animal Studies", is the first German-language publication to address the diverse questions surrounding animal agency and power. Human-Animal Studies are thus addressing a gap in previous research on the central concept of agency, which is on the agenda in a wide variety of disciplines. Controversial approaches beyond anthropocentrism, such as actor-network theory and new materialism, are explicitly focused on animal actors for the first time.

This is a published German translation of an excerpt of K. Dornenzweig's Master thesis (which was written in English, and is substantially longer). The thesis applies Miranda Fricker's theory of hermeneutical injustice to our growing understanding of stalking as a crime. It analyses the experience of stalking victims making sense of their trauma while encountering societal framings in pop culture which depict stalking as romantic, a justice system that expects a particular victim narrative, and a gender framing in which victims of no gender are taken seriously, while their actual experience leaves them shattered and feeling mad.