A Value-Driven Framework for the Design of Ethically Aligned Artificial Intelligence Systems

AI systems are operating with increasing autonomy and capability in complex domains in the real world becoming more and more ubiquitous in ways that we do not understand yet. These systems will continue to become more complex and pervasive as they become more intelligent and newer applications are enabled. Increased autonomy in such systems means that they have the potential take actions that can be detrimental to humans within their increased sphere of interaction unless they are designed to be otherwise. Thus systems with AI capabilities urgently require better alignment with human values. This is called the AI Value alignment problem (AI VAP). In this paper, a framework for creating such an AI system is proposed based on the traditional Indian conceptualization of intelligence and values. Human values have been the foundation of Indian scriptures, including the Bhagwad Gita, Vedas, and the Upanishads. Values act as guidelines for our behavior. They have a significant influence on our personalities, attitudes, and perceptions. The traditional Indian philosophy offers a holistic perspective on intelligence and values by considering various aspects of human cognition and consciousness. The proposed framework for creating Value-aligned AI system based on these perspectives provides conceptualization of the different modules required to actually implement Value aligned AI systems. This makes the proposed framework different from those that provide high level ideas without delineating implementation aspects. A rudimentary case of AI value aligned system is also provided to illustrate ideas presented in the paper.