Here's a weird one: I CAUGHT AI BEING PRO-FREEDOM!
I've mentioned before (#876782) that I'm currently translating a 1922 book by a Swedish economist. Not a luddite, I'm happy to use the tech tools available at my disposal—in this case DeepL, the self-described "most accurate translator" out there.
Anyway, today the author discussed the history of guilds and their influence on economic outcomes.
Guilds, and their modern equivalents labor unions, have an interesting status in the ~econ literature; they are clearly monopolistic ventures in that they, per FEE.org, "manipulate the market to extract an advantage for those involved is certainly a monopoly-type goal." 1
There's a big empirical fight in economic history whether guilds historially were
- on net harmful (because they were anti-competitive measures that reduced labor supply and increased prices/wages); or
- on net beneficial (because they also operated as a sort of apprentice system in societies with very little formal or informal schooling).
Result undetermined.
But from a strictly economistic point of view, a guild—or indeed, a modern-day labor union—insofar as they are effective, bid up wages above their equilibrium level, thus generating surplus income to their workers while reducing the total market (and thus contemning outsiders to unemployment).
It is not uncommon for free marketers to hold unions in very low regard; they're the flipside of a monopoly, extracting unfair value from an idealized competitive system.
Now, what's all of this got to do with AI and translation?
Ah, so I asked DeepL for this sentence (in Swedish):
"Skråväsendets historia erbjuder obegränsad lärdom på denna punkt."
To my absolute amusement, DeepL interprets guilds as crooks. The word ("skrå") and "väsende" (system) turns into:
"The history of the crooked system offers unlimited lessons on this point."
It's absolutely wonderful, because from my/free marketeers' point of view, that is precisely what guilt systems/labor unions are! Crooks.
So DeepL, an AI-type neural net, captures the factor-analysis type core of how many many humans up and down the centuries have discussed guilds in these two languagesand—pretty much correctly in my mind— judges that it is a CROOKED system. Ha-ha, confirmation bias is a powerful drug.
Sometimes AI models hallucinate; sometimes the ideological biases of their programmers/trainers shine through; ...and sometimes the core, big picture meaning of a word—rather than its literal meaning— gets captured and included in a translation.
Fucking lovely.
Proof I ain't fooling you:
Or, maybe this was all just a lucky coincidence, or perhaps I'm reading too much into this wonderful turn of events. Probably, that's the case.
I asked ChatGPT to double-check the work, and unfortunately this AI properly restored the translation order:
"The history of the guild system offers unlimited lessons on this point."
Oh well, it was very fun for a while.
Footnotes
-
Benjamin Rogge, "The Labor Monopoly" FEE, Dec 1, 1964: https://fee.org/articles/the-labor-monopoly/#:~:text=It%20may%20be%20true%20that,certainly%20a%20monopoly%2Dtype%20goal. ↩