I think the most crucial thing not noted here is that there are still many non-AI translation jobs available simply for the reason that not all companies want their confidential and proprietory data to become some AI/machine translation company's property and grist for their algorithms via being fed to a MT application, such as by some unaware and careless translator.
Firstly, with the advent of generative AI like ChatGPT, we are likely to see more and more non-native speakers confidently producing their own texts in foreign languages or post-editing MT output themselves. The demand for translations may well decrease considerably – especially for translations into and out of English.
Secondly, the efficiency of MT is a myth.
While some translators find that editing MT output increases their productivity, many experienced translators find that post-editing takes them more time than a normal translation (see e.g. Kränzler, Artificial Intelligence in Technical Translation, table 2 on page 5, 201217_sof8_ai-in-technical_translation.pdf). Post-editing MT output requires comparing two existing texts, assessing the quality of the target text and, if found to be necessary, editing it. This activity is more complex than grasping the meaning of just one text and rendering it in the target language in the grammatically and stylistically most appropriate way. Ironically, as MT technology gets better, it produces more plausible output and spotting mistakes in the MT output requires more concentration rather than less.
Any claims about the efficiency of MT in comparison to professional human translation appear even more doubtful if you consider the enormous amount of resources that go into building and operating MT systems – not just in terms of computing power (hardware and electricity) and data, but also in terms of brain power for programming and training.
So far, it is not clear how to make money with artificial intelligence based on Large Language Models. Still, investors betting on what they hope will prove to be a revolutionary technology are pouring huge amounts of money into this technology – disrupting and distorting the translation market.
MT systems are expensive. Training and fine-tuning them is costly. Language Services Providers (LSPs) feel the need to invest in MT to stay competitive, but struggle to recoup their costs – last, not least because their customers have come to expect paying 40% less for translations. The only cost factor left for making savings appears to be translators’ pay. The beauty of being hit with a 30-40% rate reduction is that 10% inflation barely register.
By introducing new benchmarks (“adequate” or “fit for purpose” translations) LSPs’ (ab)use of MT has driven down translation quality and prices as well as translators’ rates and job satisfaction. Translators are considering career changes and students’ interest in translation courses is in decline.
This development could perhaps be halted if translators’ rates were not determined depending on how the target text is produced, but depending on the expected quality of the translation. According to the relevant ISO standard 18587:2017 “full post-editing” is the “process to obtain a product comparable to a product obtained by human translation“.
On principle, a product of the quality of a human translation should always be remunerated like a human translation. Whether translating from scratch or doing full post-editing, professional translators deliver what MT will never be able to do. Only humans are able to grasp and render the full meaning of a text, to explain their translation decisions and to accept liability for their mistakes. It should go without saying that their work needs to be remunerated accordingly.
Few people would expect their hairdresser to charge them less if they supplied him with a pair of scissors – or presented him with the results of a robot’s first attempt at cutting their hair. If translation is to remain a sustainable profession, translators need to take a page from the hairdressers’ book.
Hi Angela! I appreciate you sharing your thoughts. They are similar to those of translators I spoke to for my piece: "Many of the translators I spoke to remain skeptical of this approach. They told me that machine translations are often so bad that it takes more work to fix them than it would have taken to produce a translation from scratch. Some found the experience so frustrating that they’ve stopped accepting this kind of work."
Actually, the point I was making is that it is the very process of post-editing that is more cumbersome than translating "from scratch" - not the dubious quality of the MT output. And the more plausible the MT output appears the more mentally exhausting the post-editing process becomes.
Professional translators enjoy the creative aspects of coming up with good translation solutions and producing texts they can be proud of. It is precisely the most enjoyable part of our work, however, that is being replaced by MT.
It is often said that AI will do the boring bits of work and set humanity free to focus on the interesting stuff. Well, for most experienced translators translating is the interesting part, while checking and editing translations is boring. Adding insult to injury, we are asked to do our work in a more tedious and time-consuming fashion and at a lower price.
MT is a fantastic tool for anybody unable to translate because it opens up the chance of getting - almost instantly- a fair idea or at least a glimpse of the meaning of texts that would otherwise be incomprehensible. Used in that way, e.g. in order to decide whether to order a professional translation of a foreign language text, MT is a useful tool. It is not particularly useful for professional translators.
If you ever bought cheap Chinese consumer electronics from the likes of Ali-Express, you have probably read some really awful and incomprehensible user guides. That's because the small Chinese manufacturers can't afford to hire professional translators, and they just do what they can with either non-professional office workers or maybe some really dated translation software. ChatGPT or other LLM based translation software would be a marked improvement.
If you want to use a metaphor, think of it as McDonalds food versus gourmand food prepared by the best professional chefs. The latter is indubitably better, but the former is what most of the world can afford.
What % of other people in the AI space are as optimistic as you? I fully admit that I am a raging AI pessimist due to being extremely anxiety-prone and intolerant of uncertainty, coupled with a complete lack of computer/software engineering knowledge. So I just go off of what I’m reading on the internet and it is TRULY all over the place, haha!
It probably depends on which parts of the "AI space" you inhabit. My own perspective is heavily influenced by the views of economists, who I think may have the best perspective on the economic impact of new technology. I'd guess a majority of economists would find a view like mine to be plausible.
I don't have as good a sense for people closer to AI like engineers, scientists, or company executives, but I wouldn't be surprised if they are more pessimistic, on average, than the economists.
Consider programming. AI definitely helps programmers, but is it really any different than access to Stack Overflow, optimizing compilers, common libraries and APIs, or, going farther back, to structured programming and compilers instead of assembly, to IDEs and debuggers instead of punch cards.
All those things have meant that people who don't really understand computational complexity and Big O notation, or pointer arithmetic, or other topics, are perfectly able to program to solve other tasks. The resultant code may be less perfectly optimized and efficient than the most talented coder could do, but I don't think it's resulted in a lack of jobs for top coders, rather in an explosion in the number of places where software is used.
People have been lamenting the death of the Real Programmer for as long as their have been computers.
Reminds me of the legal field. I think many, many fewer lawyers will be needed over time, but I don’t see how drafting everything using a large language model is going to save significantly more time than using previous filings to draft new briefs....”Chat GPT can write contracts” yeah and every law firm already has all the contract templates they need lol. As a paralegal, I’ve tried drafting some letters using Chat GPT (changing client names/and tweaking identifying facts) and of course I knew the output would need editing, but Chat GPT’s work product was way more fucking awful than I ever could have imagined. Did not save me any time at all. Maybe it’ll be a lot better in a few months or years though...
Do you not worry about giving workers false confidence here? AI performance is accelerating more than linearly, no? More generally, what evidence are the assertions in your final paragraph based on?
One specific note: the finding that median translator pay fell by 8% from 2010 to 2021 is already adjusting for inflation. That is, it is already netting out the consumer benefits of automation for those workers, the fact that "automation in other industries lowers the cost of the goods and services we buy."
edit: My own conclusion paragraph would have reflected on your subtitle: "Productivity is up and real wages are down." Econ 101 teaches the intuition that productivity and wages are proportional, but that's not true when the technology in question is more *substituting* for human labor than *complementing* it. Historically, technological progress has eventually led to the marginal human having a higher marginal product of labor (albeit after a painful adjustment period, perhaps), but there is nothing in economic theory that guarantees that will continue to be the case. Instead, the signs are pointing in the opposite direction, and it is imperative to recognize those signs and plan ahead, as a society.
The evidence for the last paragraph is 200 years of economic history. This translators story is one example, but there are lots of others. You can look at how long it took for companies to reorganize to take advantage of the PC and the Internet between 1980 and 2020. You can look at how long it took factories to adapt to the potential of electrification in the early 1900s. Real-world industries and markets are complicated, and it takes time for industries to reorganize.
Can I prove that AI won't be different? Obviously not. Nobody knows for sure. But people have been saying "this time is different" for many decades and so far it never has been.
And sorry to be contrary, but inflation adjustment does not "net out the consumer benefits of automation for those workers." Quite the contrary, if goods and services get cheaper over time, that will be measured as deflation, which will increase inflation-adjusted incomes.
On the inflation, though, I think you have it mixed up. A "real wage" is meant to be a measure of the real goods the nominal wage rate can procure. So an 8% decrease in real wages means the median translator can afford 8% less in goods and services in 2021 than they could have in 2010. (One can question whether a given inflation adjustment is accurately capturing differences in quality, composition shifts, etc., but that's a separate issue, a measurement issue as opposed to a conceptual issue.) In the usual notation, PQ = WL, so that the real wage is W/P = Q/L. Indeed, goods and services getting cheaper over time means higher real wages, other things equal. And the fact that we don't see real wages rising for translators suggests that the benefits of automation have not outweighed the costs, for them.
Ah I see what you mean now about inflation. Sorry for not getting it before.
I think your point is basically right, but it's a little hard to know what the counterfactual is. For example, maybe increasing scarcity of resources (like oil for example) means that in the absence of technological progress everyone's real income would be declining. On the other hand, maybe there are other factors (such as globalization and the Internet) that would be pushing down translator wages even in the absence of globalization.
So we don't know how real translator wages, or the wages of other occupations, would be changing in a counterfactual "no automation" world. Without that baseline, we can't say whether automation has been a net benefit or harm to translators over the last decade.
Yes, 100%. I thought about adding an "other things equal" caveat but felt my comment was already getting too long. Looking back, though, I should have. Given these other factors (as well as the substantial measurement questions around inflation), a decline in measured real wages is not sufficient to determine the net effect of economy-wide automation on a set of workers. For that, something more like this is needed: https://onlinelibrary.wiley.com/doi/full/10.3982/ECTA19815
AI performance might be accelerating more than linearly but implementation isn’t going to grow at the same exponential rate, is what I think the argument is here. His final paragraph probably leans on the same amount of evidence most AI reporting leans on - not much. I don’t mean that in any disparaging way at all, we just simply don’t know what’s going to happen so he’s guessing just as much as anyone else putting out educated guesses.
If so, I would ask that he phrase those educated guesses more modestly. Adding the word "likely" is not a sufficient hedge if there's no evidence for the claim that something is more likely than not. More adequate would be to say "I suspect" or "if this analogy holds true."
I understand the argument that implementation lags the basic tech, but does that lag really transform exponential growth into something linear (or quadratic)? I suspect not.
I have written about this myself from the perspective of a human translator: https://alakasa.substack.com/p/how-will-gpt-4-affect-white-collar
I think the most crucial thing not noted here is that there are still many non-AI translation jobs available simply for the reason that not all companies want their confidential and proprietory data to become some AI/machine translation company's property and grist for their algorithms via being fed to a MT application, such as by some unaware and careless translator.
That's a good point thanks!
This article misses a few crucial points.
Firstly, with the advent of generative AI like ChatGPT, we are likely to see more and more non-native speakers confidently producing their own texts in foreign languages or post-editing MT output themselves. The demand for translations may well decrease considerably – especially for translations into and out of English.
Secondly, the efficiency of MT is a myth.
While some translators find that editing MT output increases their productivity, many experienced translators find that post-editing takes them more time than a normal translation (see e.g. Kränzler, Artificial Intelligence in Technical Translation, table 2 on page 5, 201217_sof8_ai-in-technical_translation.pdf). Post-editing MT output requires comparing two existing texts, assessing the quality of the target text and, if found to be necessary, editing it. This activity is more complex than grasping the meaning of just one text and rendering it in the target language in the grammatically and stylistically most appropriate way. Ironically, as MT technology gets better, it produces more plausible output and spotting mistakes in the MT output requires more concentration rather than less.
Any claims about the efficiency of MT in comparison to professional human translation appear even more doubtful if you consider the enormous amount of resources that go into building and operating MT systems – not just in terms of computing power (hardware and electricity) and data, but also in terms of brain power for programming and training.
So far, it is not clear how to make money with artificial intelligence based on Large Language Models. Still, investors betting on what they hope will prove to be a revolutionary technology are pouring huge amounts of money into this technology – disrupting and distorting the translation market.
MT systems are expensive. Training and fine-tuning them is costly. Language Services Providers (LSPs) feel the need to invest in MT to stay competitive, but struggle to recoup their costs – last, not least because their customers have come to expect paying 40% less for translations. The only cost factor left for making savings appears to be translators’ pay. The beauty of being hit with a 30-40% rate reduction is that 10% inflation barely register.
By introducing new benchmarks (“adequate” or “fit for purpose” translations) LSPs’ (ab)use of MT has driven down translation quality and prices as well as translators’ rates and job satisfaction. Translators are considering career changes and students’ interest in translation courses is in decline.
This development could perhaps be halted if translators’ rates were not determined depending on how the target text is produced, but depending on the expected quality of the translation. According to the relevant ISO standard 18587:2017 “full post-editing” is the “process to obtain a product comparable to a product obtained by human translation“.
On principle, a product of the quality of a human translation should always be remunerated like a human translation. Whether translating from scratch or doing full post-editing, professional translators deliver what MT will never be able to do. Only humans are able to grasp and render the full meaning of a text, to explain their translation decisions and to accept liability for their mistakes. It should go without saying that their work needs to be remunerated accordingly.
Few people would expect their hairdresser to charge them less if they supplied him with a pair of scissors – or presented him with the results of a robot’s first attempt at cutting their hair. If translation is to remain a sustainable profession, translators need to take a page from the hairdressers’ book.
Hi Angela! I appreciate you sharing your thoughts. They are similar to those of translators I spoke to for my piece: "Many of the translators I spoke to remain skeptical of this approach. They told me that machine translations are often so bad that it takes more work to fix them than it would have taken to produce a translation from scratch. Some found the experience so frustrating that they’ve stopped accepting this kind of work."
Actually, the point I was making is that it is the very process of post-editing that is more cumbersome than translating "from scratch" - not the dubious quality of the MT output. And the more plausible the MT output appears the more mentally exhausting the post-editing process becomes.
Professional translators enjoy the creative aspects of coming up with good translation solutions and producing texts they can be proud of. It is precisely the most enjoyable part of our work, however, that is being replaced by MT.
It is often said that AI will do the boring bits of work and set humanity free to focus on the interesting stuff. Well, for most experienced translators translating is the interesting part, while checking and editing translations is boring. Adding insult to injury, we are asked to do our work in a more tedious and time-consuming fashion and at a lower price.
MT is a fantastic tool for anybody unable to translate because it opens up the chance of getting - almost instantly- a fair idea or at least a glimpse of the meaning of texts that would otherwise be incomprehensible. Used in that way, e.g. in order to decide whether to order a professional translation of a foreign language text, MT is a useful tool. It is not particularly useful for professional translators.
If you ever bought cheap Chinese consumer electronics from the likes of Ali-Express, you have probably read some really awful and incomprehensible user guides. That's because the small Chinese manufacturers can't afford to hire professional translators, and they just do what they can with either non-professional office workers or maybe some really dated translation software. ChatGPT or other LLM based translation software would be a marked improvement.
If you want to use a metaphor, think of it as McDonalds food versus gourmand food prepared by the best professional chefs. The latter is indubitably better, but the former is what most of the world can afford.
What % of other people in the AI space are as optimistic as you? I fully admit that I am a raging AI pessimist due to being extremely anxiety-prone and intolerant of uncertainty, coupled with a complete lack of computer/software engineering knowledge. So I just go off of what I’m reading on the internet and it is TRULY all over the place, haha!
It probably depends on which parts of the "AI space" you inhabit. My own perspective is heavily influenced by the views of economists, who I think may have the best perspective on the economic impact of new technology. I'd guess a majority of economists would find a view like mine to be plausible.
I don't have as good a sense for people closer to AI like engineers, scientists, or company executives, but I wouldn't be surprised if they are more pessimistic, on average, than the economists.
Consider programming. AI definitely helps programmers, but is it really any different than access to Stack Overflow, optimizing compilers, common libraries and APIs, or, going farther back, to structured programming and compilers instead of assembly, to IDEs and debuggers instead of punch cards.
All those things have meant that people who don't really understand computational complexity and Big O notation, or pointer arithmetic, or other topics, are perfectly able to program to solve other tasks. The resultant code may be less perfectly optimized and efficient than the most talented coder could do, but I don't think it's resulted in a lack of jobs for top coders, rather in an explosion in the number of places where software is used.
People have been lamenting the death of the Real Programmer for as long as their have been computers.
Reminds me of the legal field. I think many, many fewer lawyers will be needed over time, but I don’t see how drafting everything using a large language model is going to save significantly more time than using previous filings to draft new briefs....”Chat GPT can write contracts” yeah and every law firm already has all the contract templates they need lol. As a paralegal, I’ve tried drafting some letters using Chat GPT (changing client names/and tweaking identifying facts) and of course I knew the output would need editing, but Chat GPT’s work product was way more fucking awful than I ever could have imagined. Did not save me any time at all. Maybe it’ll be a lot better in a few months or years though...
Do you not worry about giving workers false confidence here? AI performance is accelerating more than linearly, no? More generally, what evidence are the assertions in your final paragraph based on?
One specific note: the finding that median translator pay fell by 8% from 2010 to 2021 is already adjusting for inflation. That is, it is already netting out the consumer benefits of automation for those workers, the fact that "automation in other industries lowers the cost of the goods and services we buy."
edit: My own conclusion paragraph would have reflected on your subtitle: "Productivity is up and real wages are down." Econ 101 teaches the intuition that productivity and wages are proportional, but that's not true when the technology in question is more *substituting* for human labor than *complementing* it. Historically, technological progress has eventually led to the marginal human having a higher marginal product of labor (albeit after a painful adjustment period, perhaps), but there is nothing in economic theory that guarantees that will continue to be the case. Instead, the signs are pointing in the opposite direction, and it is imperative to recognize those signs and plan ahead, as a society.
The evidence for the last paragraph is 200 years of economic history. This translators story is one example, but there are lots of others. You can look at how long it took for companies to reorganize to take advantage of the PC and the Internet between 1980 and 2020. You can look at how long it took factories to adapt to the potential of electrification in the early 1900s. Real-world industries and markets are complicated, and it takes time for industries to reorganize.
Can I prove that AI won't be different? Obviously not. Nobody knows for sure. But people have been saying "this time is different" for many decades and so far it never has been.
And sorry to be contrary, but inflation adjustment does not "net out the consumer benefits of automation for those workers." Quite the contrary, if goods and services get cheaper over time, that will be measured as deflation, which will increase inflation-adjusted incomes.
Fair enough on the economic history.
On the inflation, though, I think you have it mixed up. A "real wage" is meant to be a measure of the real goods the nominal wage rate can procure. So an 8% decrease in real wages means the median translator can afford 8% less in goods and services in 2021 than they could have in 2010. (One can question whether a given inflation adjustment is accurately capturing differences in quality, composition shifts, etc., but that's a separate issue, a measurement issue as opposed to a conceptual issue.) In the usual notation, PQ = WL, so that the real wage is W/P = Q/L. Indeed, goods and services getting cheaper over time means higher real wages, other things equal. And the fact that we don't see real wages rising for translators suggests that the benefits of automation have not outweighed the costs, for them.
Ah I see what you mean now about inflation. Sorry for not getting it before.
I think your point is basically right, but it's a little hard to know what the counterfactual is. For example, maybe increasing scarcity of resources (like oil for example) means that in the absence of technological progress everyone's real income would be declining. On the other hand, maybe there are other factors (such as globalization and the Internet) that would be pushing down translator wages even in the absence of globalization.
So we don't know how real translator wages, or the wages of other occupations, would be changing in a counterfactual "no automation" world. Without that baseline, we can't say whether automation has been a net benefit or harm to translators over the last decade.
Yes, 100%. I thought about adding an "other things equal" caveat but felt my comment was already getting too long. Looking back, though, I should have. Given these other factors (as well as the substantial measurement questions around inflation), a decline in measured real wages is not sufficient to determine the net effect of economy-wide automation on a set of workers. For that, something more like this is needed: https://onlinelibrary.wiley.com/doi/full/10.3982/ECTA19815
AI performance might be accelerating more than linearly but implementation isn’t going to grow at the same exponential rate, is what I think the argument is here. His final paragraph probably leans on the same amount of evidence most AI reporting leans on - not much. I don’t mean that in any disparaging way at all, we just simply don’t know what’s going to happen so he’s guessing just as much as anyone else putting out educated guesses.
If so, I would ask that he phrase those educated guesses more modestly. Adding the word "likely" is not a sufficient hedge if there's no evidence for the claim that something is more likely than not. More adequate would be to say "I suspect" or "if this analogy holds true."
I understand the argument that implementation lags the basic tech, but does that lag really transform exponential growth into something linear (or quadratic)? I suspect not.