OpenAI funds research on artificial intelligence morality

OpenAI is investing a million dollars into research at Duke University aimed at creating algorithms capable of predicting human moral judgments. The team led by ethics professor Walter Sinnott-Armstrong will try to teach artificial intelligence to make decisions in ethically complex situations in medicine, law, and business.

Artificial intelligence as a moral compass for humanity

Researchers from Duke University have previously created a morally oriented algorithm for kidney transplant recipient decisions. Their vision now heads towards creating a kind of moral compass to assist people in complex ethical decisions. But is it even possible to teach a machine to understand human morality?

History warns us. In 2021, the Allen Institute developed the Ask Delphi tool, which was supposed to provide ethical recommendations. It could correctly assess that cheating on a test is wrong, but just rephrasing the question made the system begin to approve overtly unethical behavior. The reason? AI systems are essentially just statistical machines without true understanding of ethics.

Western bias in digital morality

Current AI systems are primarily trained on data from the Western world, leading to a one-sided view on morality. Systems often reproduce values of Western, educated, and industrialized society, while other cultural perspectives remain overlooked.

Additionally, each algorithm is trained on different data. This is evident in controversial judgments, such as when Delphi deemed heterosexuality more morally acceptable than homosexuality. Other AI systems also hold different philosophical stances.

Claude, for instance, leans towards Kantian absolutism, while ChatGPT inclines towards utilitarianism. It seems truly challenging to create a universal moral algorithm when even philosophers have been debating ethical theories for thousands of years.