It seems fairly obvious that it's meant to be an analogy for ChatGPT only supporting textual outputs until now (disregarding that it had some limited capabilities to show static images).