Poll: What is the most effective way to reduce bias in AI language models?
Thread poster: ProZ.com Staff
ProZ.com Staff
ProZ.com Staff
SITE STAFF
Sep 17

This forum topic is for the discussion of the poll question "What is the most effective way to reduce bias in AI language models?".

View the poll results »



 
Maria Teresa Borges de Almeida
Maria Teresa Borges de Almeida  Identity Verified
Portugal
Local time: 10:12
Member (2007)
English to Portuguese
+ ...
N/A Sep 17

Asking this question to us translators is useless, whatever our opinion. Shouldn’t it be addressed to computer engineers? A brief Internet research led me to this MIT paper:

https://news.mit.edu/2024/researchers-reduce-bias-ai-models-while-preserving-improving-accuracy-1211


Jennifer Levey
expressisverbis
Claudio Machado Junior
Sara Daitch
Luis M. Sosa
Maria Laura Curzi
Zea_Mays
 
Alex Lichanow
Alex Lichanow
Germany
Local time: 11:12
English to German
+ ...
The most effective way... Sep 17

...is to toss "AI" language models out of the proverbial window, forget about them and let human professionals do their job.

Claudio Machado Junior
Daryo
Maria Laura Curzi
Sean O'Byrne
Maria Teresa Borges de Almeida
Géraldine Fourrier
expressisverbis
 
Claudio Machado Junior
Claudio Machado Junior  Identity Verified
Brazil
Local time: 06:12
Member (2014)
English to Portuguese
+ ...
I really don't care Sep 17

Really.

Maria Laura Curzi
Philip Lees
 
Sean O'Byrne
Sean O'Byrne
Germany
Local time: 11:12
German to English
Just don't use them Sep 17

Just don't use them

Maria Teresa Borges de Almeida
Géraldine Fourrier
Angus Stewart
expressisverbis
 
expressisverbis
expressisverbis
Portugal
Local time: 10:12
English to Portuguese
+ ...
My view Sep 17

I'm not anti-AI. I use AI myself sometimes, and I can see its value as a tool and an assistant only.
What I am against is treating AI as an opportunity to reduce the amount of work available to translators and, at the same time, push their rates down.
A translator is not simply a human quality-control layer for AI.
We are qualified professionals who can analyse, research, understand and translate complex material themselves.
AI can assist the translator, just as CAT tool
... See more
I'm not anti-AI. I use AI myself sometimes, and I can see its value as a tool and an assistant only.
What I am against is treating AI as an opportunity to reduce the amount of work available to translators and, at the same time, push their rates down.
A translator is not simply a human quality-control layer for AI.
We are qualified professionals who can analyse, research, understand and translate complex material themselves.
AI can assist the translator, just as CAT tools, terminology databases and other technologies do. But it shouldn't automatically mean replacing part of the human work, reducing the volume of work available, or lowering the value placed on that work.
Perhaps reducing bias in AI translation also means treating translators as human professionals with the knowledge, skills and capacity to do their job properly, and recognising their right to be treated with the same dignity and professional respect as any other skilled professional.
Collapse


Philip Lees
Dan Lucas
Maria Teresa Borges de Almeida
 
Philip Lees
Philip Lees  Identity Verified
Greece
Local time: 12:12
Greek to English
Wipe 05:13

Wipe their memory clean.

Seriously, this is a daft question. Once an AI already has bias - which they all inevitably do - it's not possible to "reduce" it. All you can do is add more bias towards an alternative direction, or alter the angle at which the bias tilts by playing around with its data sources.

That's because "bias" is an entirely subjective concept, dependent on the weighting attached to different factors by different points of view. So while your view of som
... See more
Wipe their memory clean.

Seriously, this is a daft question. Once an AI already has bias - which they all inevitably do - it's not possible to "reduce" it. All you can do is add more bias towards an alternative direction, or alter the angle at which the bias tilts by playing around with its data sources.

That's because "bias" is an entirely subjective concept, dependent on the weighting attached to different factors by different points of view. So while your view of something might seem to you to be completely balanced and objective, to me it might appear biased - and vice versa.

The best we can hope for, as with humans, is that an individual, whether person or AI, is able to recognise their own biases and compensate for them when appropriate.

I have little optimism that such a capability is among the top priorities of the tech companies developing the latest LLMs.
Collapse


expressisverbis
Maria Teresa Borges de Almeida
 
Thayenga
Thayenga  Identity Verified
Germany
Local time: 11:12
Member (2009)
English to German
+ ...
Avoid them... 06:31

...like the plague whenever/wherever that is possible.

Maria Teresa Borges de Almeida
 


To report site rules violations or get help, contact a site moderator:

Moderator(s) of this forum
Jared Tabor[Call to this topic]

You can also contact site staff by submitting a support request »

Poll: What is the most effective way to reduce bias in AI language models?






Pastey
Your smart companion app

Pastey is an innovative desktop application that bridges the gap between human expertise and artificial intelligence. With intuitive keyboard shortcuts, Pastey transforms your source text into AI-powered draft translations.

Find out more »
PerfectIt
Check translations faster for consistency, quality and style guide compliance

PerfectIt helps professional translators deliver error-free documents in Microsoft Word and PowerPoint. It helps improve consistency, ensure quality, and enforce style guides. ProZ members receive a 30% discount off individual purchases of PerfectIt.

More info »