Last week, I decided to get into my thoughts about AI, and what it means for artists. Before I really lean into that though, I want to get into a slightly more technical approach to what all this AI stuff is really doing. After having quite a few discussions with folks about what exactly AI is, and what it’s doing, I’ve come to realize that while we all see some images and text being generated, there are a myriad of applications and mechanics most folks are not really fully aware of. We’re fixated, but on what exactly?
Heads up, this is going to be a long one, so it’s probably best to view this in a browser, or in the Substack app. Either way, grab a coffee or something, because this gets a little thick.
So, in this second part of this three-part article, I want to lightly touch on how AI works, non-artistic applications of AI, and how I think we are just scratching the surface of what is going to shape the world around us in coming years. I want to preemptively clarify that this article is taking a very high-level approach, and while I will stay objective on how things work, the views I express here are entirely my own and should not be considered any sort of advice.
Full disclosure: I used chatGPT to generate some answers on some questions I had. I have been studying AI for over a year now, and in some instances, I learned more through OpenAI’s chatbot in 30 mins, than I did in a year. I have not copy-pasted any text here, but have certainly paraphrased some bits here and there.
Unrelated: I wonder if the “are you a human” check at chat.openai.com is AI generated 🤔
What & Why
First thing’s first. I consider AI a tool. It is not a sentient entity. “It” is not taking anybody’s jobs, the same way the advent of database software did not kill analog bureaucracy; people still use folders and filing cabinets for fucks sake. Sheesh. As with all tools, there needs to be a person behind it, telling it what to do, giving it a purpose to exist. In and of itself, an AI is an inert, functionless mechanism.
So who are the folks behind the various tools we have suddenly become aware of? Where did DALL-E, VQGAN+CLIP, Midjourney, ChatGPT, and all the other bazillion text and media generators come from? Well you can google the shit out of it, and there are tons of articles out there that explain it all. For the purposes of creating context in this article, I’m going to quickly break down three organizations that are responsible for the bulk of AI generated content we are seeing today: OpenAI, DeepMind, and Stability AI.
I consider AI a tool. In and of itself, an AI is an inert, functionless mechanism
OpenAI
OpenAI is a for-profit company that was founded in 2015. It is primarily a research organization with a very broad focus on AI systems, such as fields like finance and robotics. Their three primary products are DALL-E, GPT models, and Whisper. DALL-E is a deep learning model that generates images, GPT-1/2/3 is a neural network learning model that is capable of generating virtually any type of text, and Whisper is an open-source speech recognition model.
Stability AI
A private company founded in 2019, Stability AI is one of the newer organizations developing advanced AI algorithms and tech, and is primarily focused on improving reliability and stability in related systems. They are behind Stable Diffusion, which has proliferated in variants very rapidly, thanks to the open-source nature of the codebase. It can be used for all manner of text, image, and even audio, and there are a great many forked projects that are being trained across a multitude of different datasets.
DeepMind
Unlike most AI research organizations, DeepMind is a subsidiary of Alphabet - yes, the same parent company that owns Google. Their focus is almost entirely on developing highly advanced AI algorithms and technologies. They don’t have any products like Wombo to generate a weird festive penis or anything like that, but their research is unparalleled in the field. They have literally published thousands of papers on AI, including some that explored AIs playing video games, winning at chess, solving archeological questions involving dead languages, and even molecular biology, to list a few. Fucking. Wild.
A Bridge
Now a crucial component of a lot of this is something called natural language processing (NLP). It is essentially a technology that makes it possible for computers to communicate with us sentient humanoid carbon units in a more “natural” way. The term intuitive comes to mind, but it’s difficult for me to accept that a non-sentient chunk of software can have such a thing; it’s more of a predictive system built on tons of data. For a fun (terrifying) sidenote, check out this article on Lemoine running a Turing test on an AI and being convinced it’s sentient. (apologies in advance for the WaPo link).
NLP stands for Natural Language Processing
The irony of folks developing algorithms to better commune with machines, is that it has resulted in a higher understanding of the complexities of human language. From applications like virtual assistants and sentiment analysis, to language translation and even grammatical correction for writing, it would appear that we have actually learned more about ourselves by developing AI technologies. Noam Chomsky probably has some thoughts on this. Check out how many research papers Stanford has published currently, solely under the topic of NLP.
Complexity
Everyone is so worried about how easy it is getting for anyone to simply use an AI to generate their content, but I think what many fail to understand is that the more complex the datasets get, the more difficult it gets to actually train an AI. OpenAI’s GPT-1/2/3 language models are a great example of this
GPT-1, the first generation model had a parameter size of ~1.5 billion. GPT-2 was around 175 billion. And the current, latest third generation model, GPT-3 has a parameter size of 175 trillion. Let that sink in for a minute. That is not linear growth; this is an exponential spike quantitatively in the amount of data a model is being built on. As you can imagine, the more data a query has to run through, the more computing power & resources it will require to be completed.
As more AI products flood and saturate the market, we will see a decrease in the types of tools we see in that industry. Partially it will be due to a decrease in novelty - how many Midjourney fantasy portraits do you think we actually care to see after a while? But it will likely also be because training your own AI will take an insane number of processor cycles, and the affordability of running a major rig is unreasonable unless you are well funded for your project - probably one of thousands trying to achieve similar goals. At this stage, it will likely be more about knowing how to train and use an AI, than it will be about developing your own personalized one.
Industry
So what exactly are these AIs being trained to do? Is there some evil cabal of people that hate artists that are feeding datasets to AIs with the label “I hate artists and want them to stop making art?” Who fucking cares? There are so many other things we should be worried about being upended. From education to healthcare industries, we should be talking about what future jobs will look like, and why we need to begin thinking about things like copyright, privacy, and training - training actual humans to use these new technologies.
Chatbots are replacing therapists, and customer service folks. I’m guessing most of us have experienced the futility of yelling “connect me to a human you fucking piece of shit!” on some call with an ISP, utilities company, or even the IRS. Next up, trying to figure out if you’re talking to an actual human or not when all you need is a refund for the busted ass toaster hotbread.wtf sent you. Caps may be cruise control, but are just as ineffective as yelling into a cell phone when all you need is to be heard about your cause for depression.
Folks are using AIs to streamline and optimize marketing and advertising campaigns. Finance AIs are becoming a standard for preventing fraud, checking credit scores, and even for predictive market forecasting. In retail, an AI just sent you an email about what you are actually interested in shopping for. And in healthcare? Well let’s just say that AIs analyze medical imaging, and diagnose diseases way the fuck faster than some 70 year old dingbat that perceives you as a fucking nuisance while all you want to do is be treated like a human.
Nobody is taking your paintbrush away, but if you don’t want to be a starving artist, maybe start paying attention to what companies are handling the rest of your existence with new technologies. The scariest part of this to me, is potentially not knowing who trained and developed the AIs that are in use, when handling your private information, and data that may impact your life directly. Does a human decide who gets the next transplant, or an AI trained by a company that prioritizes profit over equity?
This is why I believe it is important to de-mystify the mechanics behind AI. I wanted to get deeper into GANs and the importance of noise in creating more robust outputs, but you can easily scour the web for data on this. Or, you can just ask chatGPT about it.
Copyright
I’m not even going to attempt to get into this here in great depth, but I think it has to be mentioned. We talk about copyright infringement in relation to our rights as artists, but it goes way, way, way beyond that. Looking at U.S. copyright laws, one will realize how outdated, and problematic they are. I’ll write an entire article about this eventually, but I want to point out that copyright laws here don’t generally serve public interest at all; they exist primarily to protect larger entities (corporations mostly).
In fact, copyright can be so restrictive, that 90% of your process of making art, or building a new product, can be entirely stifled by the way the laws are setup. Copyright laws suck the fucking joy out of everything. Put up a doodle of a Star Wars character and sell $1 stickers of it and see what happens when Disney & LucasArts jump down your throat with lawyers. To them, they’re Santa, and your life is their chimney. To everybody else, it’s a fucking burglar in the night.
Copyright laws are so insanely complex that the public generally cannot afford to understand them, and protect themselves - these laws are intentionally prohibitive for public comprehension, and easily abused by those that practice this corner of law. Now think about how new AI is, and how absolutely unprepared we are as a society to handle implications of an AI scraping media, and having some powerful parties determine how that should be regulated. Who is making these decisions? Are they qualified? Does Gary Gensler know the difference between a blockchain transaction and how an email is transmitted? The answer is no. We’re fucked!
Most creatives and agencies are still working with Creative Commons licenses to use work under oddly specific guidelines. What do you think will happen if you ask an AI to generate a new image from a CC image you bought, and then cycle that image to create a new image, and sell it with no license, effectively placing it under general copyright law? Let that loop through your brain for a minute. Take your time.
Here’s one more to think about: people (like me) do a bunch of research and write articles based on information they learned from other articles. We take care not to plagiarize, but we are essentially paraphrasing a lot of what we read. How is that any different than what a text AI does when it generates answers to queries? Well, at least I have some style to my writing, even if it means I cuss a lot, while the AI is just annoyingly polite (read: sterile).
Obsolescence
The question is always “is this good or bad?” The black and whiteness of that question is fucking obsolete when we’re talking about bleeding edge technologies. If you think that AI should be stopped because it’s going to kill [insert occupation]s, you can check out this awesome community that is probably perfect for you!
While I’m not saying certain jobs will not become defunct, I do believe that many jobs will need to adopt to this unstoppable tidal wave that is approaching at a terrifyingly rapid rate. I’m going to end this article with a scenario related to education, because I personally believe that a lack of education is one of the cornerstones of societal strife, and misunderstanding. The following example may or may not be based in real life events.
If You Can’t Beat Them…
Carry McHopium gets accepted to a prestigious 4 year digital design program. To cover the cost of her bachelor’s degree she goes into debt. The DingleFuck University markets itself through various private companies that getting a degree here results in getting a great job! Look at all the happy people in the ads, wearing GAP and H&M classics, wow!
Carry graduates DFU with a 3.9 GPA. She lost the missing 0.1 to perfection, to having to spend time working her way through college, so she can afford the books and software necessary to complete her degree. This is in addition to the $60K debt she has incurred. She’s stoked, she’s got job interviews lined up and can’t wait to show the world what she’s learned.
She lands a job that starts at a $20K salary. On day one she realizes that all the software she learned to use, is now 3, 4, and even 5 versions ahead of what she had learned. There are even tools in Photoshop she didn’t even know existed. She spends the next 3 months learning all the new industry standard shit while working 60 hour weeks, and posts happy IG posts saying “learning never ends!”
Others much younger than her get ahead very rapidly in the company, without college degrees, because they spent 6 months out of high school binging youtube tutorials, and freelancing. They complete multiple freelance gigs simultaneously with great speed by hiring folks on fiverr to do their work for them.
Carry says “fuck this,” and goes to find a better job. She has a degree and some cool stuff on her resume now. She lands a $30K job. The company emails her regularly at 6PM on Fridays to get that graphic revised for the 50th time in time for the influencer party that’s gonna be bussin, no cap, over the weekend. She gets zero credit for the work, and her bosses do lines of whatever off their clients’ partners’ asses on Saturday nights. Carry spends her Saturday nights starting at job listings with a bottle of $5 Merlot.
One day, Carry wakes up and has a nervous breakdown. She goes doom-scrolling online in her depression, on the unpaid sick days she took off from work, and finds chatGPT. Through this simple AF chatbox, she learns more about the marketing industry in 2 hours than she did in 4 years at DingleFuck U. She asks at one point, “what is the best way to get ahead quickly in the graphic design industry?”
The next day, she quits her job, goes to one of her now-ex-company’s parties, parties really hard with one of their clients, and shows them everything she’s done for the past couple years. She gets a $90K salary job with them using cutting edge AI tools for predictive marketing. She pays her college debt off in under 4 years. She changes her last name to McCopium. DingleFuck U goes out of business a few years later.
Her college teachers didn’t know better, but should be thinking about what they’re going to be doing next. Industries like this have had to get their shit together for years. Begging for help from government programs, taking money from jobless students, and getting in the sheets with corporations to wipe out liberal arts to replace them with engineering and sportsball is not the fucking way. If anything, AI has the potential to be the wake-up call for institutions like these, so that they can stop grifting the masses just to keep their employees paid.
Does AI scare you? I think it should. Should this paralyze you? Fuck no. Lean into your fear, and go learn how to use these tools; they are inevitable. Next week, I will publish the third, and final article in this series, and it will be focused entirely on what this all means for us as artists. I might even cuss a little less (probably not).
Alrighty, I’m done yelling into the void. Let me leave you with a question:
Can an AI teach you the techniques behind painting, creating 3D graphics, or even motion design? If yes, how is this different from watching YT tutorials?
That was a long one. If you’re still here, damn! I’m impressed. I really hope you’re all planning to touch some grass, and not let AI and the web consume your life this weekend.
I’ll be dropping my next Transmissions from the Void episode with Boxhead this weekend! Be sure to check it out, because it’s absolutely not about AI, and mostly about art and life! Go subscribe if you haven’t already, there are so many good interviews coming up.