Competing Visions of Ethical AI: A Case Study of OpenAI
This case study of OpenAI reveals that while the organization's public discourse heavily emphasizes safety and risk, it notably avoids integrating academic and advocacy ethics frameworks, raising concerns about potential ethics-washing and the need for improved governance.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the world of Artificial Intelligence (AI) as a massive, bustling construction site. The machines being built here are incredibly powerful, capable of writing stories, solving complex math problems, and even creating art. But because these machines are so new and so fast, the people building them are still figuring out the rulebook. In this field, there's a big debate about how to make sure these smart machines are "good." Some people talk about "ethics," which is like a moral compass—a set of deep questions about right and wrong, fairness, and justice. Others talk about "safety," which is more like a hard hat and a seatbelt—practical steps to stop the machine from crashing or hurting anyone. While everyone agrees we need both, the question is: who is actually talking about what, and are they really following the rules they set?
This paper takes a close look at one of the biggest construction crews in the AI world, a company called OpenAI. The researchers wanted to see how this company talks about these big ideas over time. They didn't just listen to what the company said; they acted like digital detectives, scanning thousands of documents, blog posts, and research papers to count exactly how often the company used words like "ethics," "safety," "risk," and "alignment." They treated the company's public voice like a giant library, checking the shelves to see which books were being read the most and which ones were gathering dust.
The story the data tells is a bit surprising. When the researchers looked at OpenAI's public website—the place where they talk to regular people—they found that the word "ethics" was practically invisible. It showed up in less than 4% of their articles. It was like walking into a house where the owner keeps talking about "fire safety" and "emergency exits" but never once mentions the word "fire" itself, even though they are the ones holding the matches. In fact, the word "safety" appeared nearly 700 times in just one year, while "ethics" appeared only seven times.
Instead of using the deep, philosophical language of ethics, OpenAI's public messages were dominated by technical terms like "safety," "risk," and "alignment." The researchers found that when the company did use the word "ethics," it was often just a quick mention, like a passing comment in a headline or a single sentence in a paragraph, rather than a serious discussion. It was as if they were using the word as a decoration on the wall rather than a tool for building. Interestingly, the researchers noticed that the company's own tags and labels were shifting. Sometimes an article was labeled "Ethics & Safety," but the text inside barely mentioned ethics at all. This suggests that the labels might be changing to signal different things to the public, rather than reflecting a deep change in how the company thinks about moral questions.
However, the story was slightly different when the researchers looked at the company's academic papers—the serious, technical documents written for other scientists. Here, the word "ethics" showed up more often, appearing in about 17% of the papers. But even in these serious documents, the word was often used in the margins, in footnotes, or in short lists of values, rather than being the main focus of the research. The researchers suggest that while the company is very good at talking about technical safety and managing risks, it tends to keep the heavy moral conversations about "ethics" on the sidelines, treating them more like a reputation badge than a strict rulebook for how they build their machines.
In short, the paper suggests that OpenAI has built a very loud megaphone for talking about "safety" and "risk," but has kept its voice very quiet when it comes to "ethics." They aren't ignoring the idea of doing good, but they are framing it in a way that focuses on technical fixes and avoiding danger, rather than on the deeper questions of fairness and justice. The researchers conclude that this way of speaking might make the company look responsible, but it also means that the hard, messy questions about what is truly "right" and "wrong" in AI are being pushed to the background, hidden behind a wall of technical safety talk.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.