Semantic satiation is the term for saying a word so often that it doesn’t sound real or meaningful for a while.
An EEG study from the MIT Media Lab found that people using chatbots to do their work displayed the weakest signs of neural activity for semantic processing, memory load, and creative ideation compared to those using traditional search engines or no tools at all. That means the cognitive offloading provided by a language doesn’t meaningfully activate your brain’s mechanisms for semantic coding, or the process of taking abstract ideas and constructing language that conveys the meaning. Your brain doesn’t perform the work of matching a word and its past context to present concepts, or invoke a theory of mind or human communication since you know the output comes from a program. Instead, your brain slips into a passive pattern recognition state that doesn’t commit the information being processed into long-term memory.
In other words, chatbots are really good at semantic satiation, which means they’re really good at killing the meaning of words. What we’re looking at is a Losing Language Model, where humans are trained not to actually engage with the relationship between abstractions and their symbolic representations, but to quickly scan for key symbols that seem relevant to their task.