Idiomdrottning’s homepage

Denying the ML antecedent

Normally going online and into the li’l echo chamber created by my steady stream of blocks is cozy & comfy because everyone is sane about everything (climate, bodily autonomy, criticizing capitalism, eating plants, etc) and sometimes I even learn new things (like how bad cryptocurrencies are, and yes, they’re super bad).

One recent hot topic where I started out in full agreement because my initial take was also super negative is ML & generated art. But I’m a li’l scared that the steady stream of super bad arguments and outright misinfo will gradually push me into a more pro-ML-stance.

Sorta like how a really bad vegan argument (“we need to go into the jungle and stop all the spiders in there from eating flies! And the flies need voting rights!”) can make some people really crave a burger. Even though there are plenty of actual good arguments for sticking to plants on the plate.

That is such a fallacy. A formal propositional fallacy called denying the antecedent.

It goes like this:

If all those bad anti-ML arguments were legit, ML would would be bad.
Those arguments don’t seem very legit.
It’s fallacious to then conclude that that makes ML good & OK, because there might be other arguments against it that are better.

For me the three main remaining problems with ML are:

  1. Resource usage. Electricity and e-waste has unaccounted for externality costs. “ML wouldn’t be used if it weren’t the best tool for the job, that’s how market works” isn’t true because of these externalities. But this isn’t universally the case. There are cases where using a NN is way more efficient than any other algo, even when factoring in the training, but until energy is rationed it’s easy to accidentally get super wasteful.

  2. Increasingly concentrated ownership of very powerful means of production. Spinning Jenny squared. This warps political and economic power away from the people and into the hands of the few. Now, models that are open source or runs or can be created or trained locally are counter-arguments to this but there still are some petabyte models out there that’s way beyond what we have in our garages. This is a disaster. We need to fix our economic system. We need to do that with or without ML.

  3. Yes, it does feel a li’l bad that stuff that’s fun to do by hand now can be done by a machine. But painting was (and is) fun yet there still are practical uses for cameras also. Handwriting feels good yet sometimes we use typewriters or even digital text.

For bad arguments, I have an older post about that. The one bad argument I see all the time these days, “the machines can’t be original”, I didn’t originally have in there but I’ve added it now.

Taking some of those other bad arguments to their extreme, we would also have to oppose automatic typesetting like any TeX or word processor does. Even fonts that use automatic interpolation of edges using beziers (i.e. all vector fonts) would be not OK. Only handwritten documents from now on. And of course using a camera is right out.

Maybe that would be a fun world to live in, actually. I’m not making a very good case here. 😅