The Maqasid, and the Ethics of Artificial Intelligence
My thoughts on Midjourney, GPT-3 and a number of newer AI Developments
I had the opportunity to mess around with Midjourney the past week and it really got me thinking about a number of deeper concepts and philosophical questions around what all of this means.
For those unfamiliar Midjourney is an AI text to image generator which has been trained on billions of pieces of art and text on the internet. If you had asked anyone 10 years ago, they would have said that something like Midjourney wouldn’t be coming out for another 50-100 years. There's a few of these tools now, Daal E, Stable Diffusion, Imagen just to name a few.
The speed of advancement in AI is something that is exponential. Every next step is enormous. To demonstrate this with an example let's say you started with $1 and I doubled it every day, on the 30th day you would receive $1,073,741,824; a billion in 30 steps.
It's said that Tolkien after finishing The Lord of the Rings actually started to work on a sequel that took place a hundred years later called "The New Shadow", which was devoid of magic but was about the world of men and the struggles therein. However, he decided not to continue it or publish it after a few chapters leaving only notes in his memoirs about it. He believed that the Lord of the Rings was enough and he wanted to end it there.
Now, an AI can come along, or will very shortly and read all the Tolkien's work and replicate his style and then release a sequel. But Tolkien decided not to do so for a reason, that reason and all the emotions that went into that decision is what makes the decision great. Is a sequel to the Lord of the Rings written by an AI in the EXACT style of Tolkien still Tolkien? Most would answer no, it's just a fan-fiction. The question is, will it matter?
I think it should. It must.
I've been hearing a lot of people say that AI is just a tool like any other.
I'd like you to read the following argument presented in an essay (which I've attached) by Steven Zapata, an award winning artist and professional designer regarding the argument that AI is just a tool. Although the full essay is lengthy, I've clipped some relevant sections which I think are a worthwhile read below.
AI art programs may be a tool right now, in their earliest state, but to think they will continue this way requires a willful ignorance about the tool you’re using and the environment around it. They are not meant to be tools for artists- they are replacements for artists. And they are advertising themselves as such. Just read the language they are using to sell it to you–-here are some quotes from Stable Diffusion’s release for researchers: “Stable Diffusion is a text-to-image model that will empower billions of people to create stunning art within seconds.” And here’s another one, “You can see some of the amazing output that has been created by this model without pre or post-processing on this page.”
Notice the language–they do not want a future where you need to touch these images after they’ve been generated, they have no interest in leaving any task for an artist to do–they want you completely out of the picture–even though your art trained the AI. And you should do a “ctrl-f” search on the page I pulled those quotes from for the word “artist”- zero results. They want to sell the promise that someone with no experience can make the same image on day one that someone with years of experience can make. Based on their business model, the less need there is for an artist's intervention, the more successful and appealing their product is.
Whether or not they actually achieve this goal of producing an AI that needs no artistic intervention is almost irrelevant, because they will advertise that they have achieved that goal so broadly that it will utterly transform the optics around art for the world at large. They’ll invest millions to advertise to the common person that stunning images, videos, whatever- are made at the press of a button. That said, I believe they probably will achieve it. Much of what is possible with these systems in their earliest releases already surpasses “good enough”. You will not be touching up these images, improving them, or compositing over them for long. And be wary of those at these companies who will say these are tools for artists only when confronted by artists, while they ceaselessly market it to everybody else as a way to easily and immediately generate incredible art with minimal fuss.
The usual retort here is that even if the images need no fixing someone must always do the prompting or selection of desirable outputs. If your imagination failed you before and it’s doing it again, you should be doubly ashamed. The future of these AIs does not involve humans sitting around typing prompts into them. The dream of getting a job as an AI Soothsayer who, through loving cajoling pulls the most beautiful possible image out of the machine, is short-sighted. The AIs are just as good at generating strings of text as they are at generating images. In your rush to prompt, you failed to notice that you were training the next part of the AI- the one that knows the combined taste of millions of the most tasteful people in the world. This is clearly signaled by programs like MidJourney, which permanently and publicly archives every single piece it generates, including the prompts used. When you prompt, you are shouting your inner heart into a new data set for the AIs.
Once that data set reaches critical mass they won’t need you to tell the AI what to make. These systems, perhaps many countless instances of them, will run on autopilot. They will simply iterate on what they have learned people are interested in seeing, combine it with real-time analyses of the internet and other systems, and begin an explosive outpouring of media that will never end, completely transforming the art market. The sheer volume of output will allow the AIs and their handlers to manipulate the market, flooding feeds with images when they want something to disappear, and easing off when they want something to get attention. We would be lucky if this only affected the commercial art market, but the wholesale devaluation and silencing of art will likely affect every sector–commercial, hobbyist, fine art, everything. I think of this supernova of mediocre inhuman emission as the “Mega-Feed”, the ad absurdum version of the comparably weak “feeds” we are familiar with today.
It will turn out then that it was you, who was “just a tool.” You were used to teach the AI which of its creations is the best, as you do every time you click on your preferred result amongst its many grotesque offerings. You teach it the keywords and buzzwords relating to styles and rendering criteria that are of the broadest interest. And every time you return to it with a novel idea, it will, of course, note your originality, pat you on the head, and then turn its baleful machinations to distilling and infinitely redistributing whatever creative seed was in there.
...
I have already mentioned the biggest problem with this argument–the AIs will be very capable of running on autopilot, and they will get just as good at telling stories as they are at making images and videos. They will produce novels, essays, and scripts in amounts that can fill the library-of-babel, each piece a composite of half quotations and unattributed swipings. All this auto-generated text can be processed by the image and video AIs to generate long-format media, and the cycle will be complete, self-contained, and human-free.
Companies will leap on this system, of course, since it’s predictable, consistent, and lacking the hard-to-maintain wetware and mercurial moods of the human artist. They will produce an endless stream of every imaginable film, tv, game, news story, and image as well as every imaginable permutation of each instance of these.
As I’ve said, these AIs will not need to be prompted by humans for very long and will instead auto-respond to the ebb and flow of the internet, current news, real time sales, and even private conversations. After all, we have already readied these inputs for them. We all feel a little uncomfortable when our phone shows us an ad for something we mentioned to our friend over dinner, but what happens when it shows you a movie it made just for you about your break up? A song about that careless word from your mother? A finished version of that comic idea you started researching? You’ll start getting notifications saying- “Hey! Check out one thousand finished versions of your dream!” Our ambient digital systems already have intimate access to so many of the inputs that define our taste- in some sense we sold our souls long ago.
Full essay:
There are additional issues which I've found in addition to some of the points that Zapata outlines including the involvement of Muslims in AI. A quick example is that most of the time I type Muslim woman in the AI generator I get some curvaceous beautiful looking girl in hijab with modern clothes, often tight fitting, boss babe vibes. I'm not sure what’s more disturbing. That these images can now be created with ease or that AI has been trained on 10+ years of images of tabarruj by Muslims on social media
This is really our doing
Now AI thinks Muslim woman means busty babe with hijab in cool clothes.
If you think I’m joking see below with just a basic prompt:
This is just one of a number of issues I can think of that we are going to have to contend with very soon.
It’s both impressive and deeply disturbing some of the things that the AI is able to come up with. Now you can imagine something from your childhood like “if batman was a teenage mutant ninja turtle and having tea at a table” and it will come up with an image (with some prompt tinkering) that looks very good. And it might not be perfect right now since the products are in their infancy, but it would be safe to assume that in a few generations they will be more than good enough to pass most people’s metric. Pretty cool right? I thought so too. But then I started to think very deeply about some of the implications of what these things mean and it bothered me to my core.
I kept thinking about the Quran.
I found it very interesting that the Quran was revealed to a people who were enamored with the creative art of storytelling and poetry. It was revealed as something which the Arabs knew was not poetry or magic, it was something like poetry but it wasn’t that. They couldn’t place a finger on it. It was something they had never experienced.
The way people view AI reminds me of how the Arabs viewed the Quran but in a Dajjalic Satanic way. You know it’s not art, but it’s not, not art either. It is something of course. But it’s hard to place a finger on what that is. Because when an artist creates a piece of work, his/her life experiences, emotions, skill, mood, all determine how and why a certain work is created. If you have ever perused a subreddit of AI you’ll have heard of a concept called the singularity. It’s a point at which AI reaches a point that is so advanced that it becomes basically godlike to humans. That is the goal for many people. Ray Kurzwile a famous futurist, was asked, "does God exist?" To which he replied, 'not yet.'"
I've talked about the meta-narrative for a while now, that something could appear good and fine on its face, but in reality be something wrong. What is the meta narrative of AI? I was speaking to a few friends and students of knowledge who opined that is any of this really wrong? What are the fatawa about AI? Is using an Apple Pencil, a laptop, or a drill haram? Is it haram to use Midjourney to create a flyer quickly for your Islamic event? Aren’t these just tools to be used? Aren’t Muslims pro-science? As Zapata answered above, “AI art programs may be a tool right now, in their earliest state, but to think they will continue this way requires a willful ignorance about the tool you’re using and the environment around it.” (Steven Zapata)
I think this is a question that also goes beyond Fiqh and if I’m being honest, it goes beyond my spiritual pay grade of understanding as a layman. I would like our scholars and students of knowledge to discuss this. But I can't shake the feeling that this isn't just Fiqh or Aqeedah. This is a question of Maqasid, or the goals/aims of something...what I've been calling the meta-narrative. What is the meta-narrative of AI advancement? Is it meant to replace God, audhobillah. Is that the goal? Is the goal to remove all the things that make us human?
Is the maqasid of AI advancement kufr? I don't know.
But I pray to Allah to protect us from what is evil and wrong and guide us to what is good.
And Allah and his Messenger ﷺ know best.
PS: I would love your thoughts/comments below
~muin