AI without human oversight
« previous post | next post »
Despite the panic over AI we're seeing in many sectors of society, including academia, the juggernaut rolls on, seeming set to crush everything in its way:
"EU gives more power to AI translation machines"
The European Commission has launched a pilot project to translate some press releases without any human oversight.
POLITICO (6/15/23)
BRUSSELS — The EU’s translation unit is doubling down on artificial intelligence.
For the first time, high-tech machines will translate press releases without any human oversight, in order to cut waiting time for journalists and expand the number of languages available to the public.
This marks a new frontier in the Commission’s drive to automate its translation department, one of the largest and oldest among the multilingual Brussels institutions.
My take for the moment: AI can do the grunt work, but higher level, creative, esthetically satisfying thought still rests with humans.
Selected readings
- "The AI threat: keep calm and carry on" (6/29/23)
- "ChatGPT has a sense of humor (sort of)" (6/10/23)
- "InternLM (6/10/23)" — with a long bibliography
[Thanks to Kent McKeever]
Ferdinand Cesarano said,
July 5, 2023 @ 7:11 pm
The potential for error and for mistranslation is obvious.
However, my guess is that the use of AI translations for press releases is predicted on the public's intuitive understanding that press releases typically contain verbiage that is equal parts anodyne and self-serving.
Therefore, when the inevitable mistranslation occurs, the vast majority of the public will no doubt assume that the minister or the corporate figure in question did not actually say "For this reason, I'd very much like to massage your auntie."
AntC said,
July 5, 2023 @ 7:34 pm
AI can do the grunt work
Hmm. MarkL in particular seems to be gung-ho about the broad adequacy of machine translation.
My experience travelling in Taiwan was epic fail in something over 30% of cases for entirely everyday shop/restaurant signs. 'Epic fail' as in blank incomprehension from my interlocutors — or from me of their translation. (And since back-translating often enough gave nonsense in the source language, I had no way to know whether the translation made sense.)
Of course, the danger is not of producing ungrammatical nonsense, but of producing something that appears to make sense but is just plain wrong: Use a metal blade to open this electrical equipment. Switch the power on first.
I guess EU bureaucratese is so opaque anyway, noone could tell whether this was the original or an adequate translation or out-and-out word-salad.
Chester Draws said,
July 5, 2023 @ 8:24 pm
Another example, say Poland wants to expand its military, and puts out a release saying it is going to expand by 10 artillery divizions (sic). That will be inevitably translated as "divisions". So out by a factor of up to 10 in terms of the number of artillery pieces.
No-one will spot that mistake. Why should they, because they expect artillery to come in divisions.
My specialty is false friends in military terminology, and so I check machine-translations very carefully for them, but I would bet that there are plenty in economics too.
Typing Test said,
July 6, 2023 @ 6:06 pm
AI's progress in translating without human oversight is impressive, but human creativity is irreplaceable. While AI handles repetitive tasks, higher-level thinking and aesthetics are uniquely human. Balancing AI's efficiency with human intellect is crucial for transformative outcomes. Embracing AI as a tool, not a replacement, fosters a future where human ingenuity thrives alongside technology.