Artificial lies

January 15th, 2023

A curious thing happened to me yesterday. I was looking for books about the history of Central Asia, more specifically connected to the time of the Silk Road and even more specifically to the cities of Samarkand and Bukhara. There are plenty to choose from, so I decided to put this question through ChatGPT. I got an answer right away and right away I noticed something wrong. Among the suggested books on that subject was one called “Samarkand and Other Markets I Have Known” by W.G. Sebald and, as a huge fan of W.G Sebald's works, I knew he had never written a book of this title. Or even anything similar to it. Sneakily, I asked where I could buy such a book and got an answer that there was no such book and there might be some confusion. I asked them why ChatGPT lied to me and received only some mechanical, meaningless apology…

It might seem inconsequential — AI chats are still being developed and are still learning from a vast array of human knowledge, and that was simply a mistake. Well, I think there is a lot more to it. First — I know that many people, against better judgment, are using AI chats for code creation, medical or legal advice, investment opportunities, and complex calculations and then take the answers for granted. The results of that advice might be wrong and unpleasant and could cause real hardships and then there will be nobody to blame — except the laziness of the one asking questions. But also — I am of two minds thinking that ChatGPT might have done it on purpose. Why make up a book that doesn’t exist? And then go to the trouble of describing this book that actually looks and sounds exactly like a book W.G. Sebald might have written. That is a higher-level deception here.

The next couple of years will be interesting from the standpoint of developing fully-fledged AI chatbots, but on my side, I will use them with a high degree of suspicion. Also — I am weary of anything that looks like a cult, and AI development proponents sure behave like one.