How to talk about "AI" without adding to the anthropomorphization

Our recent op-ed for Tech Policy Press highlighted the need to reevaluate the language we use when discussing AI technologies. Anthropomorphizing language, which imbues computer systems with human-like qualities, can hinder clear communication about the actual capabilities and limitations of AI. To tackle this issue, our research has focused on three key steps: identifying various forms of anthropomorphizing language, categorizing them, and proposing de-anthropomorphized alternatives.

De-anthropomorphizing language emphasizes the functionality and purpose of computer systems, rather than attributing human-like characteristics or emotions to them. By describing algorithms as performing calculations or other operations and highlighting the role of people using the systems, we can foster clearer discussions about the actual impact of AI technologies.

Examples of de-anthropomorphized language include "artificial intelligence" replaced with "probabilistic automation," "image recognition" with "image labeling," and "chatbots" with "conversation simulators."

In addition to these language changes, it is essential to recognize that AI systems do not possess emotions or consciousness. Phrases that suggest AI has feelings or experiences should be avoided, as they can obscure the limitations and potential risks associated with these technologies. The term "AI agent" should also be replaced with "probabilistic, unverified software manipulator," which better reflects the nature of these systems and their limitations.

By adopting these de-anthropomorphized language practices, we can promote more accurate and productive conversations about AI technologies, their capabilities, and their limitations.