I recently read Empire of AI by Karen Hao. The book is a recount of the beginnings of OpenAI, Sam Altman and how he’s been able to keep himself at the top of the AI industry.
She covers the legal set up of the organization, the drama between Sam and Elon Musk, Sam’s relationship with his family, particularly with his sister, the exploitation of both natural and human resources in developing countries. The overall consensus of the book is how OpenAI became an organization with an unlimited amount of resources who gets to define what AI actually is today.
ChatGPT only employs 15 world langauges above an 80% accurracy in their software. As things become more integerated with AI, these large language models shut out or force a large portion of the population to learn languages and other cultures that are better represented in the software.
One example from Empire of AI showed me how cultural values can shape AI safety decisions. ChatGPT suggested candy as a reward for children completing chores, but U.S. safety staff flagged the suggestion because of its potential association with child grooming. A European employee questioned whether this reflected an American cultural sensitivity rather than a universal concern.
That struck a chord with me. Growing up in New York, I identified strongly as Colombian, but living in Colombia has made me realize how deeply American my worldview is. Some of the issues that feel particularly sensitive to me in Colombia simply don’t carry the same weight here.
This question of language and cultural difference also reminds me of an interview I did with San Andrés reggae artist Prophet Negus. San Andrés, part of a Caribbean archipelago off the coast of Nicaragua, has three official languages: Spanish, English, and Creole. Despite being Colombian territory, the islands have strong historical ties to Britain, and Prophet Negus spoke about how language itself has become a source of division between the islands and mainland Colombia.
