Abstract: Traditional deontology and utilitarianism both prove inadequate in addressing ethical challenges brought by artificial intelligence.Bound by abstract moral imperatives,deontology fails to adapt to real-world circumstances,whereas utilitarianism runs into practical hurdles since happiness and value defy quantitative measurement.Overlooking the interplay between individual motivations and contextual specifics,the two doctrines apply a uniform criterion to rank plural values,leaving them ineffective amid intricate ethical realities.Either theory,if coded into algorithms directly,may produce morally objectionable outcomes.Recent efforts including the three-stage Goodall model and relevant risk management frameworks seek to move beyond conventional theories,yet they still overemphasize objectively quantifiable risks.Neglecting moral subjectivity and impartiality,these approaches are prone to ethical paradoxes.Clearly,resolving ethical dilemmas requires more than rigid principles or numerical calculations;it hinges on the dynamic interaction and structuring of diverse factors within given contexts.This paper therefore puts forward the concept of "structural interaction procedure",arguing that ethical norms evolve into stable resolutions through continuous interaction,competition and iteration of contextual elements.For extreme,low-frequency scenarios like the"trolley problem",the lack of sufficient interaction and social structuring makes it impossible to develop a universal moral algorithm.To formulate ethically sound and workable solutions for complex real-world scenarios,AI ethics needs to focus on dissecting situational variables,incorporating multi-dimensional feedback,and advancing evolutionary learning within interactive frameworks.This framework is not intended to deliver definitive moral answers.Instead,it underscores the practical imperative of valuing contextual interaction and evolutionary dynamics when tackling ethical dilemmas in the artificial intelligence era.

Key words: artificial intelligence, autonomous driving, applied ethics, trolley problem, structural interaction procedure, contextualism