A.I. Ruins Everything Part #473: Online Recipes

Remember the science fiction of the 1950s, when people imagined the role robots and artificial intelligence could play, dramatically improving our lives by taking on menial tasks and unpleasant jobs so that we would have more spare time, not need to work every hour, and increase our personal luxury? Weird how, in the moment of this fiction being realized, we’re instead using it to force people into poverty and destroy all art. But also, it turns out, online recipes.

Online recipes have long been weird. Anyone who’s looked up the best way to make fluffier pancakes or slow roast a brisket will know that instead of just being told how, you’re treated to the author’s entire life story for page after page of utterly pointless text, before eventually reaching an ingredients list just before you’ve scrolled to the core of the Earth. Now imagine that, but the recipe at the end is entirely fictional, the images of the food are imagined into existence by a plagiarism machine, and the method is entirely untested and deeply unlikely to work. Welcome to now!

I always feel it’s important to distinguish what sort of AI we’re talking about when it comes to these rants because there’s the very real likelihood that “AI” could be bringing us massive breakthroughs in medicine, given its ability to manage vast amounts of information in ways humans, and even regular computation, cannot, and thus save millions of lives. This, however, is not the sort of “AI” techbros are currently losing their collective minds and wallets over. That’d be Large Language Models, which are fed all of human creativity and then spit out weird, gloopy slop. Whether it be ChatGPT and its incessant outpouring of drivel (if you know anyone who trusts it, get them to ask it to create an anagram), or image creating tools that can’t count to five, or the deeply unsettling sloppy videos in which no one’s face remains the same for more than a few seconds, LLMs are art-in-garbage-out machines, and they’re wrecking everything

Including, as the wonderful YouTuber Ann Reardon points out in her latest video, recipes.

Reardon and her channel How To Cook That is best known for her debunking videos, in which she is alerted to dubious recipes posted to the internet and then attempts to recreate them in her kitchen. It’s always a joy, as she not only demonstrates how content farms and wannabe TikTok influencers alike just make up any old crap they feel like for clicks but also then shows you how to successfully create the same food! But wow is her job about to get harder with the current flooding of socials with AI-created recipes, where every element, from ingredients to method to images, is the delusion of an LLM.

As Reardon points out, alongside trying to make a couple of the recipes, this is not only a problem social media companies aren’t trying to address, but rather one being actively encouraged! Facebook is working on something called “inverse cooking,” where its AI will “create” a recipe based on a photograph, driving such utter rubbish on its own site.

It’s worth noting that while there will almost certainly be examples of AI-generated recipes that would create lethal meals, the vast majority of them will be almost right. That’s the magic of plagiarism, as the LLMs have been fed every single legitimate recipe on the internet, and will approximate an average of them all. Ask it how to make pancakes, and via taking everyone else’s efforts without paying for it, it is capable of blindly reproducing an approximated median method for making pancakes. But, given it’s not “intelligence” in any meaning of the word, it’s also not capable of checking for accuracy. Hence the “not quite right” results in Reardon’s video above.

The issues arise because we now have no means of knowing what’s a legitimate, tested recipe written by a human and what’s an android dream with faked, photorealistic images before pages and pages of machine-generated life story atop a predictive-text recipe. It’ll only be in the act of trying to follow it will we realize we’ve been duped.

And thus, online recipes—unless posted by a writer you’ve followed for a long time and can trust—are ruined. Googling for the best way to make a butter chicken curry could be gambling with food poisoning, and that might just be the Gemini result it insists on forcing on you before you can click on any of the links. Thank god for actual, physical recipe books, but what a blow for a more recent massive convenience, even if you did wear through the skin on your finger scrolling to find the actual recipe.