I didn’t want to be talking about this again but a couple of recent events have really driven home to me some truths about the state of things in magic.
I know roughly when AI was first used to promote a magic trick, essentially as soon as the available models were good enough to get away with it.
I know how deeply ingrained it has become, now on every magic shop storefront, in countless products, and in magicians promotional materials, often to an insane degree.
What I didn’t know was when this would start to be a problem.
Obviously it has always been a problem – human creativity, energy and water usage, brain rot, duplicity, etc. What I mean is when would it bite the people who don’t care about any of that shit?
There have been two recent cases of AI use which I think are just baffling but before I get into that I want to talk about my holiday in Portugal a few months ago. We stayed in a lovely coastal town called Alvor and I ate a lot of seafood. It’s no secret that I love seafood and in Portugal I ate shrimp, squid, muscles, scallops and my personal favorite, crab¹.
The hotel we stayed in had a schedule of interesting things that made every day different. Even in what was technically the off season, one night in the dining hall they had a person freshly carving a parma ham, and one night you could line up and get some freshly cooked pasta tossed in a big wheel of cheese. One night there was a doomed chocolate fountain that I wanted no part of. Many people had no idea of food hygiene around that thing, it was fucking rank. Im focusing on the food because the other thing they did differently were the activities and I had zero interest in any of their live DJs, karaoke, bingo, or club singers, and quiz night was oversubscribed by the time we arrived.
All these events had posters however, and now we get to the topic of todays post.
See any tell tale signs of todays topic here?
How about this one? Do you like to ᏤIN PRIZES?
Shoutout to ჵ℥Ձ, one of the numbers of all time.
What always baffles me about these things is: why does no one notice during the production? I see this stuff in the wild more and more, how can the people paid to produce it not see it’s wrong? I used to assume they just didn’t care. Small cog, big machine, no accountability, alienated workforce, the collapse of late stage capitalism.
But in magic, the assumption is that we’re all artists, we care about aesthetics and we connect our personal brand to things. So why then would the magic cirlcle put out something with the following symbol on it:
What’s your favourite zodiac sign? I bet it’s 8. Mine is probably either ⚝ or ⚚.
Full disclosure here, I don’t know where this was used because I became aware of it from a thread on the Magic Circle members’ Facebook page.
People have actually complained about the club’s use of AI and all the committee members who were given Disney style stable diffusion caricatures to use on social media will hopefully now go to some photoshoot they’re doing to get proper pictures again. I’m not sure if they can force people to use the new photos if they think the AI one looks better, though. After all the magic circle is almost entirely run by volunteers. You can’t make them do anything.
What about on the business side of things? Hey did you hear John Carey has a new book out? Let’s see what people have to say about it…
“The way Carey thinks about magic is exactly how we all should-clear, efficient, and deeply magical.”
– Ben Earl
Wow. High praise indeed. Except he never said that. This was a fabricated² quote in an AI generated product page from the distributor, Murphy’s Magic, which was then copied onto the store page of every magic shop in the world. Props to Ben Earl for steppng up and actually telling people that’s not a real quote though, because there’s always the danger that in offering a correction you may offend the person in question.
One time I was away during a magic contest and when I got back I assumed a recent recruit who was a young dynamic sleight of hand expert would absolutely mop the floor with the competition. But when I expressed surprise that another member, who I’d never seen do anything but some fairly regular card tricks, had actually won, he assumed that I hated his guts. You often can’t even tell someone you don’t think they are the absolute best without them thinking you’re the worst.
So for someone to mistakenly think you’ve said their work is clear, efficient and deeply magical, and you have to announce that you don’t think that at all must be utterly mortifing. So much so that this may not even be the first fake quote to sneak it’s way into magic ad copy. It just happens to be the first one that the quoted individual was willing to stand up and say so.
Many commentators have expressed concern at this quote somehow making it through every single person whose job it would have been to spot things like this. I, on the other hand, wasn’t even surpised.
Because I know the 4 step AI adoption curve:
- Person writes, person checks.
- Person writes, machine checks.
- Machine writes, person checks.
- Machine writes, no one checks.
It may be the same person, it may be multiple passes or it might just be someone meticulously planning each sentence. Either way everyone involved is deeply aware of the content.
This probably started with spell checkers, and then tools like grammarly. When LLMs came along some people would feed everything they wrote into the machine to get a fancier more grammatically correct, more engaging version.
As soon as AI was able to generate a text from a short summary or series of bullet points, people wanted to use it to shortcut the writing process. But people didn’t quite trust it yet, so they would read it back to make sure it actually said what they wanted it to.
After a certain number of successful uses of an LLM, they checking gets less thorough and eventually stops. This is the dream of corporate AI productivity. If no one even has to check what the computer writes, you can essentially get the manager who would have told the staff to write things tell the AI to do it, and you can have the manager be the last human in the chain so the machine goes straight from specification to production.
The ultimate dream of course is that the AI can also be the manager that tells the AI what to write, and also the next manager up who tells that AI what copy is needed.
The perfect company from a shareholder’s point of view is one where every employee from the CEO down to the call centre operators are a box of wires that cost less than a minimum wage hire.
This dream doesn’t quite mesh with the reality of current LLMs costing far more to operate than they can charge³, but that’s not the only reason it will fail.
We are currently seeing AIs replacing the cheapest layer of labour. Not the creative visionaries, not the meticulous engineers, not the complex coordinators. The people who are currently being let go are people told to write 500 words of gushing praise for a magic book they haven’t read. 500 words which would not typically be scrutinized too deeply.
Those people are being replaced and already there are issues cropping up, outright scandals in the small ponds they occupy.
Imagine if someone who made important decisions or worked to tighter specifications had their work instead performed by an AI system? The danger, I have heard, is not that an AI can do your job. It’s that your boss will believe some AI spokesperson who claims that it can.
More worryingly, when you get away from the impact on org charts, if you take humans out of the loop, nothing is anybody’s fault anymore. No one is to blame when some part of your life blows up because of a glitch in a computer. Companies survive by throwing employees under the bus, sacrificial lambs to the slaughter to detoxify their image in the face of scandal.
But if Murphy’s Magic or the Magic Circle get caught using an AI system producing shoddy work, who gets blamed?
Many members shook their heads at the Magic Circle but Murphy’s caused a lot of grief with their error.
But thats what happens when a big org tries to thin the workforce with AI, with the goal of eventually being one person atop a throne of non deterministic software. What if someone goes the other way; starting as a solo creator and attempting to expand into a big employee free business using AI⁴?
Well let’s shift focus for a moment, and talk about scope creep.
Scope Creep
As a person descends through the AI adoption curve, another thing happens. Just as a machine used for grammar punch ups can be used to write from whole cloth, it can also make images, it can also be a friend, it can also be a confidant, it can also be a therapist, and I cannot stress enough that it is bad at all these things.
There was a tragic story in the news recently of a teenager who killed himself at the behest of an AI. Many people say the AI drove him to it, but that gives an LLM credit for far more agency than it is capable of weilding. Rather the AI fed a troubled teens angst bat at him, amplifying it through agreement and parroting the kinds of conversations around suicide which were present in its data set. It agreed with him about the world being unfair, life being pointless and such, even helped him research methods and write a note. It correctly informed him that if he actually wanted to do it, he had to keep it a secret, as leaving any kind of clues around would be seen as a cry for help and ultimately someone may attempt to stop him. I don’t know any more details than I glaned from the opening minutes of this youtube video because I was not going to open myself up to knowing any more about it.
The sad part of this is that he originally started using the AI to help with his homework. But he found it was ‘helpful’ in other ways, leading to everything I just described.
Now… I try my best to stave off what I have identified as a Cassandra complex in myself. I cannot afford such high brow things as a psychotherapist, and I am not about to entrust such a role to some kind of LLM, but I do my best to identify the difference between the usual kind of predictive inferences that people make⁵ and the sort of multi stage predictions that would make a person watching the world trade centre fall think to themselves “This is going to make a great movie. And that movie will be called Fifty Shades of Grey.”⁶
But sometimes I feel like Eric Finch from the Wachowski Sisters adaptation of V for Vendetta, in the scene where he says he can almost see the entirety of V’s plot from beginning to end.
Craig Petty started using AI to make thumbnails for his videos, then box art for his products. Recently people have questioned if he’s using it to write some scripts for his topical youtube videos because his voice has changed. Not his timbre or tone but the pacing and choice of his words.
John Carey appears to have used AI cover art for his most recent works. It sort of raises the question of whether it was actually Murphy’s MAgic that used GPT for the promo copy or if it went to them like that. It’s possible that there are so few people in the loop that no one can actually say either way. But has he used any AI to write the books? Who could say.
Alakazam is using one of the image generating AIs to create graphics on their website and product boxes, but I know that at least one newsletter writer uses GPT to write monthly articles and also to generate text for several books he has released. There is almost certainly AI generated art on some of their card effects.
The tireless volunteers who work behind the scenes at the Magic Circle were all gifted AI generated profile pics a little while ago, and we’ve now seen some AI generated imagery end up in some magic circle publicity. Two AI epxerts recently unpicked an election rigging scandal using AI to check writing style⁷. They put out a monthly magazine, how many of the writers or editors of their monthly magazine or using it for a little productivity boost?
The magic circle, now that I mention them, has a mental health advocacy group. Did you know that? It’s not really spoken about outside of the club but every recent copy of the magic circle has had an advert for a mental health app and call centre. This is because a few years ago someone realised that while Magicians don’t suffer the same kind of mental health issues as artists, they do suffer a lot of mental health issues that drive them to become magicians in the first place.
But I’m sure that an ever growing number of magicians with possible mental health problems making increasing use of AI with an ever increasing scope of purpose will have no ramafications whatsoever. Right?
My only hope is that the OpenAI driven stock market collapse happens before one of these guys goes too deep down the rabbit hole and briefly becomes the most famous magician on the internet for very bad reasons.
¹ The first seafood place we went was called Ribeira, which is Portuguese for River, and I had the crab. Listed on themenu as “per kg” When I asked how much for a small crab they went to a big tank, pulled out a live one, weighed it in at 750g and asked if I wanted it. Mortified at the notion of giving thumbs down on this aquatic colosseum I sheepishly nodded and then sat there as they took it away to cook. I’m glad they did it that way though because if I had to point to one in the tank and literally choose who would die for my lunch I don’t think I could have done it. I’m not one of these meat is murder people, but ultimately meat is natural, and nature is brutal.
² I say fabricated rather than the more popular term “hallucinated” because machines cannot hallucinate. A hallucination is an experience and the machines lack the divine spark of sentience required to experience… anything. That said, machines have been used for fabrication since the days of the Jacquard loom.
³ It has been said that so much money has been invested into AI development that the construction of mass data centres is both a corporate money go round and simultaneously the only thing keeping the US economy afloat. Once all those data centres are built the services they offer will have to produce eight hundred billion dollars of annual revenue. If they don’t, all the companies running them will go bust and probably tank the financial sector faster and harder than the dot com bubble or the sub prime mortgage crisis, taking all our savings and salaries with it.
⁴ You may recall in a previous post I encountered someone who was doing this exact thing.
⁵ Such as the downside to staying up all night watching Gravitation before a day trip to the London Anime Club on public transport, a mistake which I will not make a second time.
⁶ Okay so the story goes that it was witnessing the attack on the World Trade Centre on 11/9/2001 inspired Gerard Way to form the band My Chemical Romance. This band’s music went on to inspire Stephanie Meyers to write the Twilight books. E.L. James then wrote some erotic AU fanfiction of Twilight called Master of the Universe in which the power balance of an immortal being dating a school girl is replaced with that of a billionaire CEO deciding to invite his new intern into a seedy bondage relationship. Later as the work grew large and horny enough to be worth publishing the names of all the characters were changed creating the Fifty Shades of Grey franchise. Boosted by Kindle sales (a format in which you could read a book in public without the cover on display) this became enough of a household name to get adapted into a movie. I lied about it being great though. I haven’t watched it but everyone seems to agree that it’s terrible, both as an erotic film, a story, and a basis for a BDSM relationship.
⁷ Okay so they actually did this, and while I personally agree with the committee’s findings, no court in the land would accept an AI analysis of the writing style used to match an email allegedly sent by a third party pretending to be a council member. There was enough other evidence without having to resort to digital tea leaves.