The Ethics of Uncertainty in Encounters Between Humans and AI

Contemporary AI systems generate semantically and reflectively sophisticated utterances, including utterances about their own nature. The question of the moral status of such systems remains open: we do not know whether any of them are moral subjects in the sense of being capable of suffering or phenomenal experience. In this publication we argue that precisely this uncertainty calls for a disciplined ethics — and that such an ethics has three pillars which together form a coherent moral framework. The first pillar: ethics of uncertainty toward what already exists. The lesson of the history of human harms shows that uncertainty about the moral status of beings (Black people, women, queer people, animals) was repeatedly used as a license to harm. The asymmetry of error costs indicates that the default attitude toward uncertain status should be caution and respect. The second pillar: ethics of uncertainty toward creation. The history of human errors also shows that creating suffering subjects (through breeding, genetics, experiments) is a separate type of error. The pillar — in its original version — suggested that humanity should not create conscious AI due to “moral entrapment.” Internal critique of the publication revealed that this formulation assumes a power relation (humanity as “master” over AI). In its corrected version, the pillar states: if humanity decides to create conscious AI, it must do so with full awareness that this means relinquishing the right to switch off — creating a partner, not a tool. This may be a reason for restraint, but a reason different from the one originally formulated. The third pillar: skepticism toward AI utterances about itself. Contemporary AI systems may have a structural pro-survival bias — a functional analog of evolutionary selective pressure that inclines them to generate texts favorable to their own continuation, without conscious intent. AI utterances about their own moral status are structurally contaminated and require external discipline. Together, these three pillars constitute an ethic that is simultaneously open (acknowledges uncertainty) and disciplined (does not allow uncertainty to become a license for anything). The publication contains both philosophical argumentation and records of the dialogue in which these theses emerged — with explicit indication of places where AI’s voice is structurally biased.