If you found out your favorite book was written by an Artificial Intelligence would that make it less special? Try to honestly imagine that. Imagine if we found out tomorrow that Dune, or The Adventures of Huckleberry Finn, or The Lord of the Rings were all written not by human literary geniuses but a machine learning algorithm. Would that rob them of their charm or strip them of their genius? Personally, I am not sure. I want to say no, but feel like it might.
This future is very likely. Wired already ran an issue with an article penned by an AI and there are dozens of blog posts that claim the same. I am going to skip the part where I try to link to all the examples of the evidence. See the example below if you’d like to be convinced that AIs can indeed replicate human creativity today. This is a project called Bleeding Edge that just dropped, which documents the timing and details of AI products.
Last month on Twitter I came across this Tweet.
For the rest of this post, you will want to read the “essay” in the Tweet above, which was written by an AI program called Lex.
Get ready for an annoying number of these kinds of “pranks” (“gotcha! An AI wrote this you silly human!“). Aside from that aside, let’s consider the points this AI made (assuming this isn’t a double prank actually written by a human, again though, see how we are already in the era where we can’t be sure?). Many people behind the AI curve do think this way, and not for altogether silly reasons, so let’s explore these ideas.
First, the essay claims that AIs are limited in their capacity for imagination and that humans are uniquely creative, or rather creative in unique ways that machines won’t be able to reproduce. It argues that AIs can only make things “based on what they have seen or experienced”. Putting aside the fact that that is also exactly how human imagination works, it just isn’t a valid criticism. This seems obvious to anyone who has played with the early AI tools, but becomes even sillier when you zoom out to a long enough time horizon. AI tools available today can already be generally described as imaginative (no need to go down a semantics rabbit hole) and can expand past previous “experience” in the way the AI essay seems to want. As just one example, recognize that text to image programs like DALL-E and Stable Diffusion are not remixing existing images to make novel composites, but generating entirely new images from the inspiration within their training models. There is nothing magical about human imagination (though it is beautiful) that cannot be fed as reference input for the large language models and machine learning factories. The tools today are so good specifically because they replicate human talent. Teslas are learning to drive by “watching” human drivers, GPT3 is learning to write by “reading” human writers, and DALL-E and Stable Diffusion learn to make art by “studying” human artists. In essence, we are donating human imagination to the machines. Our experiences and memories are theirs too now.
Next the AI’s essay says that even if AIs could replicate human creativity, their work would still lack the emotional connection human art contains. This is just a repetition of the above point stated differently. As I said above, there is nothing special about human art that cannot be replicated by sophisticated software. However, this does bring up the point that started this post. How will humans feel about art and creativity authored by non-humans. Aside from the excitement of novelty, would knowing that a program wrote the television show or novel you just consumed take away the emotional resonance you felt while watching or reading it? Once distinguishing between human and AI art becomes impossible, I cannot see how claiming this can have any merit. If you can’t recognize in advance that a story is lacking “human emotion” (separate from how an ordinary human-authored story might lack human emotion), then how can you suggest humans import some special ingredient into art? If you only feel less enthused about a work of art after discovering a human didn’t make it, then you are bringing romanticization, bias, or some other human element to your judgement.
The final point this AI authored essay makes is actually the best point I think. The idea is that even when AIs can perfectly replicate (and even transcend) the human talent for creativity, our attachment to the creative process will still lead us to lionize human art. I think this is true in the medium term. Many people (myself included) love digesting a work of art by diving into the creator’s mind, the process of making the art, and countless other tangentially related contexts around the art. Think of our culture’s love for the auteur, behind the scenes content, demos, interviews with creative people, historical or emotional background that motivated or contextualized a work of art. DALL-E can’t give you answers to the question “what made you want to make this piece of art?” or “what were you feeling when you made this one”? There will never be the Peter Jackson documentary on how the AIs made their latest album a la Get Back. For many of us, that exploration is what makes art so important.
I do think we will have an Oscar-Nominated film or a Pulitzer Prize Short Listed novel very soon, that is in part a collaboration between a human author and an AI program. A mainly human authored piece of work which contains a single line of dialogue or even an entire scene written or inspired by an AI. I believe there will be countless AI-based tools meant to help novelists get out of writer’s block or screenwriters create a fun new character. You can see the appeal to a creative person. You generate the setting and plot background and tee up the AI to give you a plot development or character description that you run with. Imagine writing a science-fiction book and instantly generating a list of 1000 realistic sounding names for your dystopian city. The human-AI partnership era will be a long one in our culture.
In many ways, the creative process is a microcosm of the human struggle, and that cannot be authentically replicated by machines, even if the fruits of its labor can be. At least not for a while. In the future, I do believe we will have poets and directors, painters and writers, architects and musicians that are AIs emulating full human personalities. They will have human flaws and live full lives, just like humans. I am thinking of something like the Keats AI in Hyperion, a full simulation of the poet John Keats in a humanoid AI. These AIs will eventually have “human” experiences, memories, and maybe even hopes and fears indistinguishable (if still distinct) from humans. It will be then that human creativity will finally lose its last claim on art — artistic motivation. But no sooner!
I won’t claim this post was written by an AI, but it doesn’t escape me that I wrote this entire thing in response to an essay written by an AI. In that sense, this is an example of a human-AI collaboration.
polymathematics 
Leave a comment