Some AI experts are worried about the potential for an AI system to run amok and hurt humans. Anthropic (a startup founded in 2021 by a group of researchers who left OpenAI) seeks to address these concerns through their Claude chatbot, which has a built-in “constitution” that can instill ethical principles and keep systems from going rogue.
The constitution follows principles from the United Nations Universal Declaration of Human Rights and other AI companies, including Google’s DeepMind. It includes expectations such as choosing a response that encourages personal security, freedom, equality, and respect. Anthropic also incorporated Apple’s rules for app developers, which prohibit content that is offensive and in bad taste. This is an advancement over OpenAI’s ChatGPT, which is improved by having humans grade the quality of a model’s answers, but without specifying what values the system should reflect.
Anthropic’s “constitution” provides an engineering solution to the fuzzy challenges posed by more powerful AI systems. The company’s co-founder Jared Kaplan states that their development is a step in the right direction for creating AI programs that are dependable and not prone to act to their creators’ detriment. With their Claude chatbot, Anthropic hopes to provide a powerful machine learning model that is trustworthy and ethical.
