When you ask a question of ChatGPT or Gemini, or one of their competitors, it can feel like magic. The illusion that the computer is somehow actually thinking — in the human sense of that word — is very compelling.
But of course it is an illusion. You are actually seeing a compendium of expressed thoughts from many different humans. All the AI is doing is efficiently gathering human-created knowledge and ideas together for you.
The answer you get frequently cites sources. But those are not the only sources — they are just the ones that the program has chosen to cite.
I wonder whether there would be any way to implement a more user-directed attribution. Could there be a way, for example, for you to mouse over a particular part of the answer provided by your chatbot, and find out from which human sources it got that particular nugget?
Such a feature might not advance the business model of the companies that create these Large Language Model programs, at least in the short term. But it would certainly benefit us actual humans to know what other actual humans originated the knowledge and ideas that we are reading.
Is that asking too much?
What I picture when you say this is one of the UI sketches from Ted Nelson’s Xanadu system, and that makes me wonder what he thinks of everything that’s going on.