Cat blindfolds, salted caramel, and algorithmic kitsch

This story is fascinating. So there’s this e-commerce company that basically asks second- or third-tier Chinese manufacturers to do their worst, and advertises it to you. If you’re willing to wait 14 days for delivery, it’s dead cheap. This suggests they may even be manufacturing on demand and air freighting, eliminating inventory as far as they can. But then Facebook introduced a new advertising product. Rather than pick products to push, and buy Facebook ads for them, you give Facebook the whole catalogue and it runs A/B tests, trains a model, and pushes the stuff that breaks the Internet.

And so they ended up paying for a huge ad campaign for cat blindfolds.

The variety and weirdness of the product line has to be seen to be believed. Some products are obvious sex toys – the strap-on knickers with attachment points for no fewer than three dildos – but plenty aren’t. Except…they are. Everyone I showed them to thought this. It’s recognisably fetish-y, but not any fetish you’d recognise, as if the products are artefacts of perversions that haven’t been invented yet. And thinking about it, this does make a kind of sense. They are combinations of unrelated tropes that get your attention, that maximise some sort of response. Fetishes don’t have to make sense. The interaction between Wish, Facebook, and us is essentially scanning through a large collection of attributes looking for combinations that push our buttons.

Isn’t that like…everything? This strikes me as being a major cultural force. Why not put some crispy bacon on that lobster ice cream? Some salted caramel on that juicy burger? Cooks may have been in the lead here. Having identified a core set of basic tastes, bitter, salty, sour, astringent, sweet, pungent, and umami, the temptation to load up a heaping horse dose of all of them and yell “Service!” was hard to resist.

It’s much wider than that. You want to dump characters you like from movie X into comic Y? Right. Art has long understood this point. A working definition of kitsch is the idea that you get something good if you take all the individual tropes people like and dump them in at ten times the usual dosage. Wish ads are a special case of a wider culture of algorithmic kitsch. People have internalised a shockingly impoverished notion of creativity in which you optimise for response by adding more tropes.

It would be easy to blame this on a culture of targets and metrics, but it has run well ahead of their deployment. Fan cultures tend to be really heavily like this and they aren’t chasing any numbers. Rather than all critics now, perhaps we are all cynical executive producers now, mentally distilling everything we see into tropes and thinking how we could squeeze in a bit more cowbell.

Speaking of which, I don’t mean to idealise original authorship versus remix culture. A good remix brings out more than was in the original, very often by honing away the unnecessary. That’s the opposite of algo kitsch. A bad one usually chucks in more stuff. That’s algo kitsch. Pick the most opposed pairings, and maximise them! Then iterate. Going back to the kitchen, the problem is that what you get is good the first time out in a sickly way and then dull.

Neurons that fire together wire together, and then tire together. The future according to those Wish ads may be incredibly boring.

6 Comments on "Cat blindfolds, salted caramel, and algorithmic kitsch"

  1. “It’s recognisably fetish-y, but not any fetish you’d recognise, as if the products are artefacts of perversions that haven’t been invented yet.”

    This reminds me of the TV adverts for food products obviously made on production lines which show chefs in kitchens making stuff and have the small caption ‘product development kitchen’, you can imagine the advert for these in a well equipped dungeon with people being flogged in the background with these items being made by hand with the subtitle ‘fetish development dungeon’


  2. Mandatory XKCD reference
    Algorithmic design like this seemed infeasible to me due to the huge amount of iteration and human feedback required, but if you can exploit facebook clicks then that’s trillions of hours to leverage.


  3. WRT feedback; Feedback here is relatively low effort relating only to views rather than consumption. Where consumption itself is relatively cheap then the effectiveness of such feedback is subject to a multiplicative effect:


    and as stross points out, that’s just the lowest of low hanging fruit.


Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.