Extending Socio-Technological Reality for Ethics in Artificial Intelligent Systems

Hybrid Cognitive-Affective Strategies for AI Safety

“…Fourthly, due to its complexity, it might require a cognitive-affective extension of society (e.g. using targeted virtual reality studies [8]) facilitating a high-quality ethical self-assessment and ethical debiasing which constitutes a scientific and technological challenge. Fifthly, while the case might seem to correspond to a rather narrow domain, it has implications that extend beyond it and will need a supportive context which can be characterized as an institutional, legal and societal challenge.…”

“…However, interpretability at the sensor level remains an important outstanding challenge.) The necessary ethical self-assessment and ethical debiasing to craft these utility functions can be assisted by experts from the legislative and be supported by technology such as virtual or augmented reality [8] (see Chapter 7) providing a rich counterfactual experiential testbed for a responsible human-centered decision-making. To do justice to the time-dependency of human ethical conceptions, one would also need to update these augmented utility functions.…”

“…Depending on the type of constellation and the focus of different recent virtual realitybased experiments [8], the moral judgements or moral actions of participants (denoted as perceivers in the following) were heterogeneous and partly contradictory overall. In these experiments elements that were decisive included for instance: the perceived nature and transparency of the agent, the legal liability of the agent, whether the accident happened by action or by inaction, whether the action involves a self-sacrifice, the number of patients, the age of patients, the personality traits of the perceiver, the culture of the perceiver and the amount of time the perceiver had for a decision [8]. This is not surprising, since moral judgements are related to a perceiver-dependent dyadic cognitive template encoding a continuum along which an intentional agent is perceived to cause harm to a vulnerable patient [373].…”

“…Third, an education of the general public might be important, since many people exhibit ethical biases based on incorrect assumptions. In the AV case, this could for instance include anthropomorphism, presumed level of intentionality and agency or misconceptions on the functioning of AVs [8]. These epistemic gaps can be addressed via a more in-depth education leading to a more informed experience and ethical debiasing which respects the manifestation of moral pluralism known from psychology [373].…”

“…In order to achieve that, the set of SDGs has to be formulated in a machine understandable version to facilitate goal-oriented AI-based solutions. In order to identify for AI value alignment purposes what a society wants (ethical self-assessment) and in a second step what a society should want (ethical debiasing), it has been suggested to combine a scientifically grounded assessment of human ethics with technological methods such as virtual reality studies for experiences from a first-person perspective [8] (see Chapter 7). Thereby, we believe that the SDGs could serve as a heuristic able to supplement ethical self-assessment by qualitatively specifying candidate human values.…”