Shared Moral Foundations of Embodied Artificial Intelligence

Meaningful human control: actionable properties for AI system development

“…In the following subsections we describe these properties in detail and illustrate their practical implications. To better highlight the properties and the implications, we make use of two example application scenarios: automated vehicles 3 and AI-based hiring 4 . Both cases manifest an urgent need for meaningful human control in non-forgiving scenarios that strongly impact people's Fig.…”

“…Both cases manifest an urgent need for meaningful human control in non-forgiving scenarios that strongly impact people's Fig. 1 The diagram, based on the framework of abductive thinking by Dorst [45], illustrating the positioning of the abductive thinking we performed in search for a solution space that would meet the claimed need for meaningful human control 4 AI-based hiring scenario: Job candidates applying for a vacancy go through an automated video interview where they record their answers to questions formulated ahead of time by the employer. After the interview is completed, an AI agent applies machine learning methods to quantify candidates' suitability for the job by correlating their facial expressions, choice of words, and voice tone to personal traits such as creativity, willingness to learn, and conscientiousness.…”

“…Deploying AI algorithms in human-inhabited environments comes with the risk of inappropriate, undesirable, or unpredictable consequences [1,2]. The misinterpreted skills of AI systems, combined with their rapid impact in public and private spheres of life, can lead to situations with a clear misalignment between human moral values and societal norms [3][4][5][6], and where moral responsibility for such undesired impacts can often not be properly attributed to any person [7].…”

“…Fig 4. Property 3: The relevant humans and AI agents have the ability and authority to control the system so that humans can act upon their responsibility, e.g., if the human recognizes that a given situation might bring the system outside the moral ODD, they can intervene to avoid this…”

Vrline in transhumanistična nadgradnja človeka

“…sistem Compas v ZDA), določitvi višine varščine ali odobritvi bančnega kredita. Ne gre za to, da bi jim v vsakem primeru pripisovali poseben moralni status kot moralnim delovalcem, vsekakor pa se je smiselno vprašati, kakšno raven transparentnosti lahko v teh postopkih od takšnih sistemov pričakujemo (kljub njihovi zanesljivosti oziroma natančnosti, ki morda prekaša človeške odločevalce) in kako je z odgovornostjo v primeru napak oziroma zmot (Cruz 2019).…”