In today’s world of rapid advances in artificial intelligence (AI), there is an urgent need for frameworks that incorporate humans “in the loop” with AI systems to ensure moral and ethical decision making. This Diverse Intelligences project, directed by Rajesh Rao, proposes a novel way to achieve this goal using a paradigm that involves “raising” AI agents in different human cultures. This approach ensures deep human feedback throughout the AI’s learning process, emulating the process of raising a moral and ethical human child.

Just as a child raised in a particular culture learns the cultural values and norms of that society, Dr. Rao’s team hypothesizes that AI agents that go through a period of childlike learning in a human society will acquire a “value system” aligned with that society’s moral, ethical, and cultural norms. Rather than arguing that the machine is a moral agent, the project team believes that today’s AI, when steeped in human culture, is capable of learning cultural rules, norms, and values. 

To test their idea, Dr. Rao and his team propose a new experimental paradigm in which AI systems are “raised” in online virtual worlds by large numbers of humans from different cultures across the world. They will implement AI systems based on probabilistic reasoning and model-based reinforcement learning using an online virtual platform and allow the AI to learn behaviors and cultural mores by observing and interacting with humans in online virtual worlds. By deploying such an experimental paradigm across the world using different languages and custom environments, the team will explore the impact of cultural and social learning on the development of uniquely human attributes such as morality and altruism toward others.