스스로 도덕적 결정을 내리는 로봇을 가리켜 ‘인공적 도덕 행위자(Artificial Moral Agent: AMA)’라고 부르는데, 현재 인공적 도덕 행위자를 위한 윤리를 마련하고자 하는 접근은 크게 세 가지로 구분된다. 우선 전통적인 공리주의나 의무론적 윤리이론에 기반을 둔 하향식(top-down) 접근, 콜버그나 튜링의 방식을 따르는 상향식(bottom-up) 접근, 그리고 이 두 접근을 융합하려는 혼합식(hybrid) 접근이 있다. 인공적 도덕 행위자 설계에 대한 하향식 접근은 어떤 구체적 윤리이론을 선택한 다음, 그 이론을 구현할 수 있는 계산적 알고리즘과 시스템 설계를 이끌어내는 방식이다. 이 때 선택된 구체적 윤리이론은 도덕적 직관이 불확실할 때의 사례들을 인공적 도덕 행위자가 제대로 분류해낼 수 있도록 도와주는 중요한 역할을 한다. 반면 상향식 접근은 인공적인 도덕적 행위자가 다양한 기계-학습을 통해 도덕적인 추론을 배워나가도록 하는 방법을 모색한다. 상향식 접근의 중요한 전제는 아동이 적절한 발달 단계를 거치면서 도덕성과 도덕 추론을 배워나가듯이, 인공적 도덕 행위자도 적절한 교육을 통해 그와 같이 발전할 수 있다는 것이다.
이 글은 공리주의적 결과론에 바탕을 둔 의사결정의 모형과 규칙들을 칸트의 의무론적 윤리 추론 규칙과 조합하는 방식을 통해 인공적 도덕 행위자를 위한 하향식 접근의 윤리를 구상해 보았다. 그런 후 이렇게 마련된 윤리를 10세 수준의 소셜 로봇 시나리오에 적용해 봄으로써, 도덕적 민감성을 갖춘 인공적 도덕 행위자 로봇 개발에 이 윤리가 어떤 방식으로 활용될 수 있는지 제안하였다.
Robots with the ability to make moral judgments on its own is called "Artificial Moral Agent(AMA)", and there are generally three approaches when it comes to establishing an ethics for the AMA. First, the top-down approach makes use of the traditional ethical theories such as utilitarian or deontological theories. Second, the bottom-up approach follows the central ideas proposed by L. Kohlberg or A. M. Turing. Finally, the hybrid approach is an attempt to unite the top-down and the bottom-up approaches. The top-down approach for the AMA begins with selecting an appropriate ethical theory, followed by coming up with computational algorithms and systems which can be said to incorporate the given theory. The role of the ethical theory in this approach is to provide the AMA with rules to follow in morally complex situations. On the other hand, the bottom-up approach focuses on developing various machine-learning methods for the AMA to learn ethical reasoning without giving it any available rules at hand. The basic idea for this approach is to regard the AMA as a human child who has the potential of obtaining moral reasoning as the child grows up according to proper developmental stages of morality. Among the three approaches mentioned above, this paper will adopt the first one to construct a top-down ethics for the AMA. To be precise, it will be a combination of utilitarian rules for decision-making and Kantian deontological rules for ethical reasoning. After this, the proposed top-down ethics for the AMA will be applied to case scenarios involving a health-care robot with the intelligence of a 10-years-old child so as to show the workings of the ethics.
Robots with the ability to make moral judgments on its own is called "Artificial Moral Agent(AMA)", and there are generally three approaches when it comes to establishing an ethics for the AMA. First, the top-down approach makes use of the traditional ethical theories such as utilitarian or deontological theories. Second, the bottom-up approach follows the central ideas proposed by L. Kohlberg or A. M. Turing. Finally, the hybrid approach is an attempt to unite the top-down and the bottom-up approaches. The top-down approach for the AMA begins with selecting an appropriate ethical theory, followed by coming up with computational algorithms and systems which can be said to incorporate the given theory. The role of the ethical theory in this approach is to provide the AMA with rules to follow in morally complex situations. On the other hand, the bottom-up approach focuses on developing various machine-learning methods for the AMA to learn ethical reasoning without giving it any available rules at hand. The basic idea for this approach is to regard the AMA as a human child who has the potential of obtaining moral reasoning as the child grows up according to proper developmental stages of morality. Among the three approaches mentioned above, this paper will adopt the first one to construct a top-down ethics for the AMA. To be precise, it will be a combination of utilitarian rules for decision-making and Kantian deontological rules for ethical reasoning. After this, the proposed top-down ethics for the AMA will be applied to case scenarios involving a health-care robot with the intelligence of a 10-years-old child so as to show the workings of the ethics.