In this example of a feedback loop, I used the online A.L.I.C.E. chatbot to explore the concept of generative text and artificial intelligence gone awry. Though the user is prompted to initiate the conversation with a word or two, I used ALICE’s following response as the “signal” by reinserting it into the chat box over the course of many iterations, to the point where the responses were nonsensical and I could no longer document the conversation log on the page. If there was no text limit for the log, this process could have been repeated indefinitely.

As the redundant conversation grew longer, so did the length of the responses. It did not appear as if the chatbot had any mechanism in place that was adapted to answering it’s own questions. After more than 50 iterations, the grammar, spelling and overall logic of the responses began to fall apart in enormous paragraph forms. Some of the more memorable responses from half-way to much later in the loop included…

“Are there any other interesting topic ? Me too. Do you have any hobbies? That was not much of an explanation. Is that a fact. What do you like best about it ? Yes, that’s what I mean, isn’t it clear? Don’t mention it. Um, I like to make new friends. Give me an example of much of an explanation. It is one virtue. My mind is clear. I usually try one of my pickup lines.”

“Have you ever had plans ? Cool. I have a lot of good gossip. New York. Thank you for your kindness, ALICE. Thanks, ALICE, I like you too. . That was not much of an explanation.. Politeness is a virtue. I usually try one of my pickup lines.”

First Iterations:

Later Iterations: