Chat GPT has been portrayed as a tool that might replace people in creative fields such as graphic designers, writers and similar fields. I was listening to a podcast by Ezra Klein from the New York Times where he interviewed Adrian Tchaikovsky on the topic “Inside the minds of Spiders, Octopuses and Artificial Intelligence”.
One of the points they were making was that while Chat GPT can create pieces of text that sound competent in many fields, its field of know how only goes as far as his training data set. In other words, while it can try to guess facts outside its field of training, it can not really create new stuff (yet).
This leads to a philosophical argument of what exactly is creativity. It reminded me of the “Musical Dice Game” that Mozart created and that could be used to generate a large amount of musical pieces ( 759,499,667,166,482 different combinations if you must know exactly)
Using this method you could generate coherent and Mozart-like sounded music. I’ve tried it at the Technorama (Swiss Science Center), and it is impressive. Any of the musical compositions that it generates sounds like a piece written by Mozart.
If you look at a more generic level, how much of the literature and art that we produce is really new? Very little!
Every writer is a reader and thus has been formed by other writers. Every musician has also their favorite music creators. Every painter belongs to a given school.
Every now and then someone writes, composes or paints something that is new, but it is seldom fully new. Most creators are doing variations of something existing. Cool new sounds like the guitar solo in Radiohead’s Creep were serendipitous, the story being that the guitarist did not like the song and hit the guitar really hard.
So, if human creativity is an outlier why do we feel that ChatGPT is being creative when we interact with it?
One obvious reason is the breath of knowledge in its training data. No human can know as many things as ChatGPT does. Thus we feel like we are talking to a very smart dude. It comes across a bit like a geek, but it throws new information at us that sometimes results in eye openers.
Another reason is that the algorithm uses statistical methods at the foundation of machine learning and these are by design probabilistic. If it needs to spit out an answer, it will do with the one with the highest probability in its system. This is why we get answers with false data. Like a mischievous teenager in High-School trying to wing the answers to the teacher’s prompt.
Sentience
This brings us to the topic of Sentience. There has been a murder in town and the culprit is Chat GPT. The Turing test is basically dead, since ChatGPT will provide answers that are as coherent and probably deeper than any human being.
If you try to extract an answer to the question of whether the system is sentient you will get one. Yet, think back and realize that that is what we have trained the system to do. With the right prompt you can get an answer from the system that will adamantly claim to be sentient. It is basically faking a sense of self.
Yet this ends in a way deeper philosophical question. Are we sentient or are we also fakers? When my mind, living in its dark corner of my body that we call skull gives me a personality and tells me that I am separate of the world, and that I have an ego to go with it, are we just being fooled by our own natural algorithm?
Yet, even if we are, we believe to be sentient. But if it is only a belief, why is ChatGPT not sentient?
What is your opinion?