Are hybrid-human-AI-systems a more humane and responsible version of AI? Empirical inquiries into the conundrums of value-by-design in language technology
TL;DR
The development of a humane and responsible version of AI requires a fundamental realignment of AI technology and research paradigms, challenging the concept of superior intelligence and power for the purpose of providing a rational and reflective evaluation of technology that goes beyond Turing's black boxing.
Abstract
The development of hybrid human–artificial intelligence approaches offers the prospect of creating a more humane and responsible form of AI. However, technological anthropologists question whether these approaches represent a significant shift. The idea that good and bad, humane and inhumane, or responsible and irresponsible versions of AI can be achieved through hybrid human–AI systems oversimplifies the complexity of technological design and implementation. Focusing on 'intelligence performance' is too narrow, distracting from the conundrum of value-oriented design and the crucial dimensions of socio-technical systems. These systems intertwine the various socio-technical layers of AI technology design with specific stakeholders and practices in social reality. Value orientation requires practices at all layers involved in technology design and implementation, and is therefore a continuous technological activity rather than emerging from a particular AI system architecture. Furthermore, the development of a humane and responsible version of AI requires a fundamental realignment of AI technology and research paradigms, challenging the concept of superior intelligence and power for the purpose of providing a rational and reflective evaluation of technology that goes beyond Turing's black boxing.