Artificial intelligence (AI) is being explored for its potential to assist in moral decision-making. OpenAI is supporting a research project at Duke University in North Carolina, USA, which aims to develop algorithms that can predict human moral judgments. The project is titled “Research AI Morality” and is led by Professor Walter Sinnott-Armstrong and co-researcher Jana Borg. OpenAI Inc., the nonprofit arm of the company, is funding the project with one million dollars, and the support is set to end in 2025.
The research seeks to create algorithms that can predict human moral judgments, especially in fields like medicine, law, and business where moral values may conflict. The project leaders have previously published a book and several studies on the potential of AI as a “moral GPS” to help people make better decisions. They have developed a “moral” algorithm to assist in decisions like organ donation and explored scenarios where people might prefer AI to make moral decisions.
However, this research faces significant challenges. Details about the project are limited, and Sinnott-Armstrong has not commented on the research’s current status. It remains unclear how or if the results will be practically implemented. A key question is whether current algorithms can capture the complex and nuanced concept of morality. Modern AI systems are essentially statistical machines trained on large datasets from the internet to recognize patterns and make predictions. For example, applications like WhatsApp can suggest how to complete a sentence, and chess algorithms calculate the best strategic move.
Creating a “moral AI” is a challenging task. AI systems often reflect the values of Western societies, as these are the predominant inputs they are trained on. They lack a true understanding of ethical principles, reasoning, or the emotions involved in moral decisions. Even within a single society, views on ethics and morality can vary widely, leading to different answers to seemingly simple questions depending on who is asked.
Philosophers have debated the merits of various ethical theories for millennia, yet a universally accepted approach remains elusive. An algorithm designed to predict human moral judgments must consider these diverse perspectives, which is a significant hurdle. Whether such an algorithm is possible remains an open question.
The exploration of AI in moral decision-making continues to be a complex and evolving field. As research progresses, it will be crucial to address the ethical implications and limitations of AI systems in this context.