Flowers for Algerithm

Since antiquity, philosophers have reflected upon a question whether machines can truly think. In his famous article "Computing Machinery and Intelligence" Alan Turing introduced a new way to look at this question. He looked at the machine as a black box and invented his famous test, the one that now bears his name, Turing Test. We stumble upon the most common form of this test, CAPTCHA, every day. CAPTCHA stands for Completely Automated Public Turing test to tell Computers and Humans Apart, but in his paper Turing didn't mention pictures with distorted letters. His idea was about chatting: a human interrogator engaging in a conversation, attempting to determine if they are interacting with a computer program or another human. In a recent paper titled "People cannot distinguish a large language model from a human in a Turing test", it was demonstrated for the first time that humans could not reliably distinguish between a human and a large language model under the research settings and specific prompts used in the study. In "Flowers for Algerithm," I explore the connection between the prompts used in this study and Daniel Keyes' famous story. This artwork suggests that we have reached a point where we might need to make large language models less intelligent to convincingly emulate human behavior, rather than vice-versa.