I think, like the two interpreters were saying at the end, that the real boon is going to come for small businesses and social situations where you wouldn't otherwise have an interpreter at all. (Or perhaps just "Sally the bookkeeper who took a few years of high school French, but is the only one in the office who knows any French at all.") For high-level diplomatic stuff or high-stakes interactions, then I think professional human interpreters are still the way to go. But if you're on vacation and want to figure out how to get back to your hotel, this might be a good substitute.
@@andres3665 I am going to give you an example (And I am a professional interpreter), For years, airplanes have been able to take off, fly and land by themselves; also, air crashes today are mostly, (almost 100%) human error. However, if you are sitting on a plane and the door is open and the cabin personnel say the following "Ladies and gentlemen, this flight is totally automated and anyone who is not confortable with a flight without pilots can leave now", most of us will get off the plane. That AI did a beeter job than many of my colleagues, but the acceptance of the technology will take at least a decade. BTW, another example are today's subway systems, which are automated; ther is no engineer or conductor operating the trains and nobody seems to care. It's not the tech, it may be ready in 1, 2 or 5 years, but users will not trust it for a while. Also, some companies decided to spearhead this and tried when the tech was not really ready and it was a costly lesson (in the tune of millions of dollars due to errors). It will take some time for them to try again.
@@ivojara You are proving the point. If you go into the direction of translation/interpreting, newsflash, you're wasting your time because you'll be replaced either by the time you're done learning or a few years after. If you learn a language as your hobby or you want to move to another country, sure, go ahead.
@@andres3665 That is not the case, some of my customer have gone that route and came back, the problem is not the performance of the machines, the problem is the performance of the speakers, they are horrible aat speaking and the accents and poor grammar are very challenging for the machines.
The issue with AI translating (not interpreting) well-written speeches is that it's just translating a perfect text that is recorded clearly and reading it out. As a conference interpreter, I encountered people who read their speeches out and the majority don't even listen because it's robotic and fast and too dense for the average listener to take in. It's not verbal communication. It's something for the press to analyze later usually. However, there are a lot of cases where people just speak normally and communicate their thoughts. Sentences break or things get repeated, there are idioms, jokes, moments where they can't be heard clearly because they don't speak to the microphone. This is where AI cannot function properly (at the moment), because a human is needed to understand what a human wants to convey. A human's life experience is needed to figure out what the problems or mistakes or jokes are. And interpret not just what is being said, but also how. Is the speaker cheerful? Where is the emphasis? Are they using silences in a certain way? What is their body language/facial expression saying? What is the context, what is happening in the world at that moment? What is the gravity of the situation? So AI may be able to _translate_ well, but interpreting is much more than just translating words.
I would love to see this with Japanese. Subjects get left out all the time, euphemism is everywhere, and social context is king. How would a computer do with that?
@@rasen7721 Most LLM AIs are trained on everyday conversations so I think it would do better than you might think. Its trained on exactly those euphemisms and casual speech. It will probably struggle more than with other languages but English also uses a lot of those things that make translation hard so I dont know if it would be that different.
like your brain does by algortihim emulating your brain functions 1000 of years of practice in mere weeks will leave us in dust i think current Ai age is like 3 year old but it will learn more and more
I'm an American living in China. AI-powered translation apps have made my life significantly easier. That being said, there is a reason why they did this with English/Spanish. When you start doing harder or more significantly different languages like Chinese to English or lesser known languages like Estonian to Navajo, the quality of the translation takes a nosedive. It will continue to get better and better very quickly but it isn't quite there yet, especially when you get into metaphorical language or idioms.
Yeah enes is arguably the best pair when it comes to machine translation, probably because the US where the tech is mostly made is basically one of the biggest Spanish speaking countries in the world.
The US and ISAF forces would have killed for this tech in Afghanistan and Iraq. So many situations where this could have been useful. Hopefully this becomes common tech for everyone soon.
I‘m a pediatric neurosurgeon in Austria where we treat a lot of Ukrainian refugees for a year now and google translate completely changed the interaction. If somebody brings a medical report I can just take a picture and instantly read the complete page, I don’t need emotions or the correct word choosing for that and it is way faster than interpreters. On the other hand, when talking with patients about surgery and prognosis I really like interpreters since patients tend to ask more questions than with a program. Tablet video interpreters are a blessing where before we had to make an appointment. But to be honest most interpreters are not as world class as these two.
Being a pro medical interpreter+translator I've had a few instances of veeery serious Google translation mistakes. I do hope you double check at least the more important bits affecting the medical decisions to be made...
I have seen veeery serious Google Translate mistakes in different types of translations, medical among them, some not really hard, so I don't trust it and I don't recommend it.
Google Translate may be great to give anyone an idea of the subject in a document, not necessarily the details. And, definitely, interpreters add more to the mix.
@@GioLester in my experience it is the opposite. I have a rough outline of the story by basic communication but sometimes I need specific implant or other surgical details which can by especially hard to trace when letters are written in cyrillic. I had interpreters asking to end the conversation because their medical English wasn’t profound enough - one mother once asked about a specific type of glue in an arterial embolization process. I understand that interpreters are a great and viable option on paper but google translate proved to be a reliable option in my everyday practice.
This was just Spanish to English, which given Kudo's location (NYC), makes sense. I'd be curious how it would do with Nepali to French, or really any two languages that are not super common to the country where Kudo was created.
They are all horrible at translating russian. We as professionals translators have practice and we laugh at the inability to translate basic sentences. The best one we have seen is yandex auto translator.
This was an amazing video! I hope you show the strengths/weaknesses with translating lesser-spoken or grammatically complex languages, or even explore whether AI is to the point where signed languages can have the same level of interpretation (or, if not, why the heck not?) As a linguist, I worry about tech bros and popsci blogs running with this notion that language is a static and fully observable thing that you can pop out of one mouth, run through an algorithm, then pop it whole-cloth into someone else’s eyes or ears. So much of communication is metalinguistic, and we barely understand the mechanisms ourselves. It’s a black-box programming a black-box.
I’m English and speak a little French. I find the ability to speak and think in multiple languages fluently amazing. At university in the nineties I had a Portuguese friend called Cristina who obviously spoke Portuguese, English, Spanish , French, German and Italian fluently. She also decided to learn Croatian for fun in her spare time as well as gaining a first in Politics. Some people are so naturally gifted I envy them.
As a speaker of Korean, I can tell you that AI and online translators will never be as good as people for translating Korean into English perfectly. As a teacher here, I can tell when a student has used Papago (the most popular translation app/website). It has terrible trouble with subject pronouns because they are not always written in Korean so it guesses what the subject may be and mixes them up in sentences. Also given that, like other languages, there may not be a direct translation between phrases and words, meaning it can make a mistake with them too. At the moment, it's about 70% effective, but that 30% means that it's very obvious when someone has used a translator.
That’s true but I think the latest technology (which I’m not sure that either Papago _or_ Kudo employs) handles the dropped pronouns _better_ (but not perfectly) because it can account for the _context_ better (as people do). (Of course, if there is _no_ context, it might make a guess, which could easily be wrong-but people would do that, too.) And, for those instances where there might not be a direct translation, these large language models, with billions of text examples, including correspondences between Korean and English, can _probably_ come up with something that some actual human interpreter or translator has employed before. That doesn’t mean that these LLMs _will_ be as good as people at translating Korean into English but probably the 30% gap will get much narrower.
I translated your comment into Korean using GPT-4, how did it do? 한국어를 구사하는 사람으로서 AI와 온라인 번역기가 한국어를 영어로 완벽하게 번역하는 데에 사람만큼 좋을 수 없다는 것을 말해 드릴 수 있습니다. 여기서 교사로서, 학생이 파파고(가장 인기 있는 번역 앱/웹사이트)를 사용했을 때 알 수 있습니다. 그것은 주어 대명사에 대해 큰 문제를 가지고 있는데, 이는 그것들이 항상 한국어로 작성되지 않기 때문에 그것이 주제가 될 수 있는 것을 추측하고 문장에서 섞어버립니다. 또한 다른 언어처럼, 문구와 단어 사이에 직접적인 번역이 없을 수 있어, 그것은 그것들에 대한 실수를 범할 수도 있음을 의미합니다. 현재로서는 대략 70%의 효율성을 가지고 있지만, 그 30%는 누군가가 번역기를 사용했을 때 매우 눈에 띄게 만듭니다.
@@tyronew5464 I translated it back using GPT 4 lol As a Korean speaker, I can tell you that AI and online translators cannot translate Korean into English as well as a human can. Here, as a teacher, you can tell when a student has used Papago (the most popular translation app/website). It has a big problem with subject pronouns, as they are not always written in Korean, so it guesses what can be the subject and mixes it up in the sentence. Also, like other languages, there may be no direct translation between phrases and words, which means it can make mistakes with them. Currently, it has about 70% efficiency, but that 30% makes it very noticeable when someone has used a translator. at the very least, its good at doing GPT 4 korean to english lol
I believe the true potential will be realized when we, as interpreters, can leverage AI to enhance our interpretation. If the AI can precisely identify all complex terminology for me, it would free up cognitive resources, allowing me to make on-the-spot decisions to either trust the AI or opt for a more suitable phrasing during human interpretation.
What would it need you for? You could train it on situations the same way you train a human. And once trained it will make far less mistakes than a human.
A great and well-informed video (although professional conference interpreting started in 1919, not Nuremberg). The final analysis is spot on. AI interpreting is going to widen access to interpreting and might help professionals but it is not even nearly a replacement and, based on current algorithms, cannot be one.
Interpreters will survive since someone will have to bear the responsibility when something goes wrong, and that's also probably the promoting mechanism that lies in the interprting circle.
Great overview! Would've loved to hear the full records of original speech, interpreters' and AI's interpretation to get a better idea though. Do you plan to publish it at some point?
Well, AI certainly helped summarize this video faster than watching it: "AI speech translation can be useful in informal situations with low consequences for mistakes. However, for critical and high-stakes scenarios like courtrooms, medical interactions, and diplomatic negotiations, human interpreters remain essential due to their ability to process complex nuances, metaphorical language, and ensure accurate communication."
I'd love to see how Barry would do on the speed test speech if he had Kudo's running translation in front of him *while* he did his translation. I'd be curious to see if it would help, or simply be a distraction.
I am an interpreter, for me, even my colleague wrting down somewthing on a paper pad beside me is a distraction, I can cope, however when the speed is ove 150 words per minute, I just need to "zone out" and match the speaker mentally, remove all the repetitions and then follow the speaker. It is a lot easier form spanish to english because of the amount of words BTW. 😉
Excellent segment I worked as translator of 19c documents in the past and I wonder what would ai do. Google translate was and isn't accurate enough in professional circles or environments. I'd love a segment focusing on translation of medieval, early modern, and 19c, pleasssssssse.
Yes, I'm aware of that. Like that's obvious and widely known I'd say. Google Translate is not that good with complex language or historical documents So I wonder how AI translates in comparison. That's the meaning of my question :) @@shakur07
@@sock1Remember you can watch videos at x1.5 and x2.0 speed. There are some extensions/add-ons which let you go x3 and beyond. He could've simply skipped or watched at x2.
@@lugh.i i know lol but it doesnt make sense that they skipped more than half of the video or put it on 2x, either way they wouldnt have understood anything. and they didnt do neither of those things anyway lol they used ai to give the general idea or smth
AI still routinely struggles with and comes up with nonsense sequences for more synthetic languages or languages with a more complex sentence structure. It would be nice to finally see at least one study that is not done on English or Spanish.
10 years ago, AI translation was trash. Now, it is good for a series of uses. Let's see what happens in the next 5 years. Oh, and ChatGPT is great for translating texts at light speed. When it comes to texts, the best translators will become revisors. The others will have to look for another job.
A.I is going to end up heavily influenced by opinion not fact. Give it time you will see. The whole world is running on opinions and no longer facts. It will follow the pattern/notion that the most noise and more popular thinking, will dominate the algorithms. Over time it will become more inaccurate, more diluted and require more maintenance through editing and refining by humans to counter balance this trend.
I don't get why it always has to be a VS thing? Why isn't anyone thinking about how we can make things more accurate and efficient by combining the two positively and efficiently instead of pitting against each other?
As a graduate student majoring conference interpreting, I would say AI will replace most average interpreters because machines can do a much better job of understanding foreign languages, terminologies, accents, and so on. Imagine what will happen in the future since AI tools nowadays are not fully trained in all languages and specifically adapted for translation.
This is misleading, no one saying AI driving will replace F1 racing drivers, job replacement always starts at lower level jobs. Of course AI will not replace interpreters for now, but lower level streaming video translators will be replaced sooner than expected.
I propose a TURING challenge. If for a pair of language most humans cannot tell the difference between AI and human translation..whether it is speech to speech or document..then AI has won. AI should be used in MOST situations for translation.
In other words, AI translators are good for people traveling on vacation asking shopkeepers how much a product costs. But, AI translators are not yet close to good enough for high-level use like in corporate negotiations or diplomatic councils.
It'd be interesting to see how A.I. handles speakers of non standard dialects, accents, and people who have disturbances in their language production ex. TBI, a stroke, or Language Deprivation Syndrome.
The bigger concern is knowing whether it's AI or human, not the translation per se. If the intended listeners do not know, that could be an issue. We are used to being deceived by public speakers already, but this makes it doubly bad, whether it's a speech, news article, photograph or video. AI is indeed a tool, and any tool will always be used by those in power to gain more control. This has always been the case.
Give it 5 years, these minor details and nuances will be smoothed over. I mean lets be honest, with the translation tools already in place pre-a.i-- it will only get better
Yeah there isn't really a threat to translators and interpreters because they lack the human element, they translate everything when you don't need every part of the sentence. Your whole job as an interpreter is to carry the essence of a sentence/message and make it understandable to the listener, cutting out unnecessary words but still retaining the core message/idea. And obviously the tone and emotion is difficult for an A.I to emulate accurately 100% of the time since it can't understand emotions like a human. Half the meaning of what you say is based solely on tone and facial expressions, that is how you know if someone is asking you or telling you to do something. It is a very helpful tool and can aid translators/interpreters but in most situations, it is not a complete replacement.
Kudo uses software that translates in real time and is backed up my human translators. AI has not tackled real time translation, however, and if you wanted to reassure people that "AI isn't coming for your jobs", this would be an ideal test. Translation tasks for documents and text are already giving way to AI Large Language Models, like ChatGPT. They are very accurate for this task and extremely fast. It will not be a long time before LLMs can handle the speed requirements for real time translation, but it's dishonest to use this as the benchmark. It's an extremely small segment of the translation business market. Wired should have done better.
Then, take a look at the FMRIs of interpreters while they are working and compare them with FMRIs of normal people speaking. And yes, I said "normal people", the brains of interpreters are being studied in several universities because they light up as christmas trees when working. I took part in one of those studies and the data will be very surprising for you.
I think it is just a matter of time. Soon AI will take over the interpreter's job. Believe me! It might take a longer time for some serious conversations to be trusted to AI's interpretation, as no one wants to lose any serious deal, but once AI proves it can do this job, (and it will be able by the time, by building more data) people will trust it with every interpretation, whether it is political, legal etc.
It is strange to me that Barry considers AI good enough for asylum cases that are also very important in terms of the life of the asylee hanging in the balance where a misunderstanding can get their case rejected.
Hmmmm. AI's are basically bureaucrats -- unemotional, literal, robotic and intrinsically sociopathic. I hate videos that are AI narrated. Some of it is just awful.
Sorry...not buying this, even 6 mos ago - this is very old tech being demonstrated. The latest publicly available LLM technology will easily do better than humans. And certainly AI wins in non-realtime interpretations, simply due to the amount of material it can interpret, in addition to the quality. Not only can current tech perform the translations, but there are apps that can actually seamlessly replace the foreign speakers mouth movements, and body language, so that the speaker appears and sounds like they are personally speaking in the target language. Sadly, it's only wishful thinking that humans will continue to be needed for interpretation once people get comfortable with the new tech.
They don't get it, and they got the least advanced AI to perform this test. I think that a lot of people are so afraid of this themed that they don't want to see the potential that AI has, they think that it will remain the same when they point out a simple mistake, AI always evolves.
Wired is getting it all wrong! It’s not about AI vs humans. It’s about humans with AI vs humans. Like one of them said, AI got all the content covered in which it was not humanly possible. If that person could supplement his own translation with the AIs version, he would probably have a better overall translation with AI than his own efforts alone.
Remember 2017? "Ai will never be able to translate language to an useful level" - almost every translator. 2023 - "it will never be able to translate the sentiment". 2025 - AI can translate at least as good as the best of us
Exactly. For higher standards and higher important cases(e.g, diplomat, business), should never use AI . Because AI will never have human sense about complicated thinking
I have grown up translating English into Gujarati or my parents. And have also tried AI it’s not as good and lacks the cultural etiquettes requirements. ⚡️🤴🏽🕉🔺➕
So far I rather the real life translator than AI🤷🏽♂️ I can't connect with AI emotionally. And the speech of the president of El Salvador was pre-recorded and made faster during the editing process. So not fair to the real life translator.
This is nothing short of extraordinary. I recently stumbled upon similar material, and it was breathtaking. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
In the human's closing statements, that's what a boomer would say. For balance and fairness, you should also have had the AI explain their thoughts as well.
Where the AI seems to suffer the most is prosody. The AI interpretation of King Felipe’s speech does not sound compassionate or reassuring; it’s actually a little _unnerving_ to listen to.
Which, ironically, makes it more accurate since Felipe's compassion and reassurance is a put-on as befits a parasitic unelected aristocrat (son of an elephant hunter appointed by Franco who revived the monarchy) steeped in privilege for whom "his" people's suffering is an abstraction that he'll never experience.
I am so happy that most people are excepting blindly how “wonderfully” AI can do the jobs of human life. I applaud you sheeple for your “open mindedness”. That is all.
there would never be a human transater 2.0 but there would hundreds of versions of machine translation in the future (which would inevitably beat the humans)
Some languages are just spoken like that. It's up to the speaker to consider the context and the role of interpreters to slow down to make their job easy.
Now is the time to *hope that religion is right* I do not really follow any religion, but if it really is the case that humans have souls and god created them and only god can create them, then only will A.I. not replace us. Otherwise, it is very likely that A.I. surpasses us in all realms once we push to develop its sentience and consciousness (and we'll break the philosophical barrier that governs it aswell, no problem, because we're very persevering creatures, us humans).
though i love learning foreign languages, unfortunately i dont think there is a future for the professions of translation and interpretation. thats not to say demand will be zero or they they will never be needed, I just think AI being 80-90% as good as a human is enough to decrease the demand for these professions massively. and lets also not pretend like human interpreters are flawless either. if AI makes mistakes, it would be foolish to think humans don't.
I think, like the two interpreters were saying at the end, that the real boon is going to come for small businesses and social situations where you wouldn't otherwise have an interpreter at all. (Or perhaps just "Sally the bookkeeper who took a few years of high school French, but is the only one in the office who knows any French at all.") For high-level diplomatic stuff or high-stakes interactions, then I think professional human interpreters are still the way to go. But if you're on vacation and want to figure out how to get back to your hotel, this might be a good substitute.
Yeah, human interpreters will still have their jobs...for like a year.
@@andres3665 I am going to give you an example (And I am a professional interpreter), For years, airplanes have been able to take off, fly and land by themselves; also, air crashes today are mostly, (almost 100%) human error. However, if you are sitting on a plane and the door is open and the cabin personnel say the following "Ladies and gentlemen, this flight is totally automated and anyone who is not confortable with a flight without pilots can leave now", most of us will get off the plane. That AI did a beeter job than many of my colleagues, but the acceptance of the technology will take at least a decade. BTW, another example are today's subway systems, which are automated; ther is no engineer or conductor operating the trains and nobody seems to care. It's not the tech, it may be ready in 1, 2 or 5 years, but users will not trust it for a while. Also, some companies decided to spearhead this and tried when the tech was not really ready and it was a costly lesson (in the tune of millions of dollars due to errors). It will take some time for them to try again.
@@ivojara You are proving the point. If you go into the direction of translation/interpreting, newsflash, you're wasting your time because you'll be replaced either by the time you're done learning or a few years after. If you learn a language as your hobby or you want to move to another country, sure, go ahead.
@@2manygamestoplay I am 57 years old, and an interpreter, I will retire before being replaced.
@@andres3665 That is not the case, some of my customer have gone that route and came back, the problem is not the performance of the machines, the problem is the performance of the speakers, they are horrible aat speaking and the accents and poor grammar are very challenging for the machines.
The issue with AI translating (not interpreting) well-written speeches is that it's just translating a perfect text that is recorded clearly and reading it out. As a conference interpreter, I encountered people who read their speeches out and the majority don't even listen because it's robotic and fast and too dense for the average listener to take in. It's not verbal communication. It's something for the press to analyze later usually. However, there are a lot of cases where people just speak normally and communicate their thoughts. Sentences break or things get repeated, there are idioms, jokes, moments where they can't be heard clearly because they don't speak to the microphone. This is where AI cannot function properly (at the moment), because a human is needed to understand what a human wants to convey. A human's life experience is needed to figure out what the problems or mistakes or jokes are. And interpret not just what is being said, but also how. Is the speaker cheerful? Where is the emphasis? Are they using silences in a certain way? What is their body language/facial expression saying? What is the context, what is happening in the world at that moment? What is the gravity of the situation? So AI may be able to _translate_ well, but interpreting is much more than just translating words.
yeah you right
Yeah language interpretation, specially realtime interpretation, is way too unpredictable for ai to handle as of now.
I would love to see this with Japanese. Subjects get left out all the time, euphemism is everywhere, and social context is king. How would a computer do with that?
it would simply fail miserably and either be nonsensical or very confidently wrong (actual j->e translator speaking)
@@rasen7721 Most LLM AIs are trained on everyday conversations so I think it would do better than you might think. Its trained on exactly those euphemisms and casual speech. It will probably struggle more than with other languages but English also uses a lot of those things that make translation hard so I dont know if it would be that different.
like your brain does
by algortihim emulating your brain functions
1000 of years of practice in mere weeks will leave us in dust
i think current Ai age is like 3 year old but it will learn more and more
Not
@@Ghost-pb4ts turns out all your friends are smarter than you and you just have a fixation with AI lol
I will always agree with the sentiment that AI is a tool and not a replacement.
for now...
@@nickamodio721 Exactly, this types of tecnology are improving, and very fast. Humans barely improve once they reach a certain point.
Lets make it short. He didnt interpret the video.
😂
how can you say “always” when you have no idea what the future will be like.. now everything else you say is discredited
I'm an American living in China. AI-powered translation apps have made my life significantly easier. That being said, there is a reason why they did this with English/Spanish. When you start doing harder or more significantly different languages like Chinese to English or lesser known languages like Estonian to Navajo, the quality of the translation takes a nosedive. It will continue to get better and better very quickly but it isn't quite there yet, especially when you get into metaphorical language or idioms.
Yeah enes is arguably the best pair when it comes to machine translation, probably because the US where the tech is mostly made is basically one of the biggest Spanish speaking countries in the world.
Americans will accept any trash translation if that means they don't have to pay a human for their service.
The US and ISAF forces would have killed for this tech in Afghanistan and Iraq. So many situations where this could have been useful. Hopefully this becomes common tech for everyone soon.
i wonder how many people there are in the world who are able to interpret Estonian to Navajo
Same goes for humans. Not many can speak Estonian to Navajo.
The risk is too high in diplomatic negotiations, and a human needs to make sure it is correct.
Also A.I. can be manipulated in some ways I think. So a human correction will always be needed.
@Humans can be manipulated, too ...
@@susanne5803 You are right but what I meant by that was changing the A.I. in some way that can change the meaning a bit and it can cause problems.
@@susanne5803can be" nah uh they're manipulated all the time.
@@susanne5803 human mistake
I‘m a pediatric neurosurgeon in Austria where we treat a lot of Ukrainian refugees for a year now and google translate completely changed the interaction. If somebody brings a medical report I can just take a picture and instantly read the complete page, I don’t need emotions or the correct word choosing for that and it is way faster than interpreters. On the other hand, when talking with patients about surgery and prognosis I really like interpreters since patients tend to ask more questions than with a program. Tablet video interpreters are a blessing where before we had to make an appointment. But to be honest most interpreters are not as world class as these two.
Being a pro medical interpreter+translator I've had a few instances of veeery serious Google translation mistakes. I do hope you double check at least the more important bits affecting the medical decisions to be made...
I have seen veeery serious Google Translate mistakes in different types of translations, medical among them, some not really hard, so I don't trust it and I don't recommend it.
There is St Jude Global interpreting charity that translates Ukrainian refugees medical papers free of charge
Google Translate may be great to give anyone an idea of the subject in a document, not necessarily the details. And, definitely, interpreters add more to the mix.
@@GioLester in my experience it is the opposite. I have a rough outline of the story by basic communication but sometimes I need specific implant or other surgical details which can by especially hard to trace when letters are written in cyrillic. I had interpreters asking to end the conversation because their medical English wasn’t profound enough - one mother once asked about a specific type of glue in an arterial embolization process. I understand that interpreters are a great and viable option on paper but google translate proved to be a reliable option in my everyday practice.
This was just Spanish to English, which given Kudo's location (NYC), makes sense. I'd be curious how it would do with Nepali to French, or really any two languages that are not super common to the country where Kudo was created.
or to Finnish
African varieties of international languages too: French, English, Portuguese...
They are all horrible at translating russian. We as professionals translators have practice and we laugh at the inability to translate basic sentences. The best one we have seen is yandex auto translator.
The AI would do worse...much worse.
This was an amazing video! I hope you show the strengths/weaknesses with translating lesser-spoken or grammatically complex languages, or even explore whether AI is to the point where signed languages can have the same level of interpretation (or, if not, why the heck not?)
As a linguist, I worry about tech bros and popsci blogs running with this notion that language is a static and fully observable thing that you can pop out of one mouth, run through an algorithm, then pop it whole-cloth into someone else’s eyes or ears. So much of communication is metalinguistic, and we barely understand the mechanisms ourselves. It’s a black-box programming a black-box.
I’m English and speak a little French.
I find the ability to speak and think in multiple languages fluently amazing.
At university in the nineties I had a Portuguese friend called Cristina who obviously spoke Portuguese, English, Spanish , French, German and Italian fluently. She also decided to learn Croatian for fun in her spare time as well as gaining a first in Politics.
Some people are so naturally gifted I envy them.
There are not gifted people, only people with interests and dedication
What's up with so many comments being jolly about the idea of interpreters being replaced by computers?
1:40 he seemed to gloss over the fact that the screen shows that 1020 as the year it picked out of the speech.
As a speaker of Korean, I can tell you that AI and online translators will never be as good as people for translating Korean into English perfectly. As a teacher here, I can tell when a student has used Papago (the most popular translation app/website). It has terrible trouble with subject pronouns because they are not always written in Korean so it guesses what the subject may be and mixes them up in sentences. Also given that, like other languages, there may not be a direct translation between phrases and words, meaning it can make a mistake with them too. At the moment, it's about 70% effective, but that 30% means that it's very obvious when someone has used a translator.
That’s true but I think the latest technology (which I’m not sure that either Papago _or_ Kudo employs) handles the dropped pronouns _better_ (but not perfectly) because it can account for the _context_ better (as people do). (Of course, if there is _no_ context, it might make a guess, which could easily be wrong-but people would do that, too.) And, for those instances where there might not be a direct translation, these large language models, with billions of text examples, including correspondences between Korean and English, can _probably_ come up with something that some actual human interpreter or translator has employed before. That doesn’t mean that these LLMs _will_ be as good as people at translating Korean into English but probably the 30% gap will get much narrower.
I translated your comment into Korean using GPT-4, how did it do?
한국어를 구사하는 사람으로서 AI와 온라인 번역기가 한국어를 영어로 완벽하게 번역하는 데에 사람만큼 좋을 수 없다는 것을 말해 드릴 수 있습니다. 여기서 교사로서, 학생이 파파고(가장 인기 있는 번역 앱/웹사이트)를 사용했을 때 알 수 있습니다. 그것은 주어 대명사에 대해 큰 문제를 가지고 있는데, 이는 그것들이 항상 한국어로 작성되지 않기 때문에 그것이 주제가 될 수 있는 것을 추측하고 문장에서 섞어버립니다. 또한 다른 언어처럼, 문구와 단어 사이에 직접적인 번역이 없을 수 있어, 그것은 그것들에 대한 실수를 범할 수도 있음을 의미합니다. 현재로서는 대략 70%의 효율성을 가지고 있지만, 그 30%는 누군가가 번역기를 사용했을 때 매우 눈에 띄게 만듭니다.
but like in what time frame
5 year 10 year or 50 year
i think in a decade you would be fooled by whos human or Ai
so you're dead wrong
@@tyronew5464 I translated it back using GPT 4 lol
As a Korean speaker, I can tell you that AI and online translators cannot translate Korean into English as well as a human can. Here, as a teacher, you can tell when a student has used Papago (the most popular translation app/website). It has a big problem with subject pronouns, as they are not always written in Korean, so it guesses what can be the subject and mixes it up in the sentence. Also, like other languages, there may be no direct translation between phrases and words, which means it can make mistakes with them. Currently, it has about 70% efficiency, but that 30% makes it very noticeable when someone has used a translator.
at the very least, its good at doing GPT 4 korean to english lol
Never say never
I believe the true potential will be realized when we, as interpreters, can leverage AI to enhance our interpretation. If the AI can precisely identify all complex terminology for me, it would free up cognitive resources, allowing me to make on-the-spot decisions to either trust the AI or opt for a more suitable phrasing during human interpretation.
What would it need you for? You could train it on situations the same way you train a human. And once trained it will make far less mistakes than a human.
@@dumbahhpersonAnd it will replace him by the way
The pore you use it , the more you train it. ;)
Interpretation is translation of emotions rather than words
Yes!!
A great and well-informed video (although professional conference interpreting started in 1919, not Nuremberg). The final analysis is spot on. AI interpreting is going to widen access to interpreting and might help professionals but it is not even nearly a replacement and, based on current algorithms, cannot be one.
Interpreters will survive since someone will have to bear the responsibility when something goes wrong, and that's also probably the promoting mechanism that lies in the interprting circle.
And interpreters will be less needed
Human tone and Psycho would never be replaced by IA.
Great overview! Would've loved to hear the full records of original speech, interpreters' and AI's interpretation to get a better idea though. Do you plan to publish it at some point?
Well, AI certainly helped summarize this video faster than watching it: "AI speech translation can be useful in informal situations with low consequences for mistakes. However, for critical and high-stakes scenarios like courtrooms, medical interactions, and diplomatic negotiations, human interpreters remain essential due to their ability to process complex nuances, metaphorical language, and ensure accurate communication."
AI's always lack the most important thing about us. Humanity
BS
you know even today many humans prefer Aivtubers instead of humans
Naive
@@Ghost-pb4tsespecially when employees engage in strike action calling for more rights as human being
It's only a matter of time until the software catches up
I'd love to see how Barry would do on the speed test speech if he had Kudo's running translation in front of him *while* he did his translation. I'd be curious to see if it would help, or simply be a distraction.
I am an interpreter, for me, even my colleague wrting down somewthing on a paper pad beside me is a distraction, I can cope, however when the speed is ove 150 words per minute, I just need to "zone out" and match the speaker mentally, remove all the repetitions and then follow the speaker. It is a lot easier form spanish to english because of the amount of words BTW. 😉
It's not yet there, but its a matter of time.
as a professional interpreter, I actually look forward on how AI can help me do better at my job
And replace you by the way
@@shakur07 HAHAHAHAHHA nice one.
Excellent segment I worked as translator of 19c documents in the past and I wonder what would ai do. Google translate was and isn't accurate enough in professional circles or environments. I'd love a segment focusing on translation of medieval, early modern, and 19c, pleasssssssse.
Google translate is not AI, chatgpt is
Yes, I'm aware of that. Like that's obvious and widely known I'd say. Google Translate is not that good with complex language or historical documents So I wonder how AI translates in comparison. That's the meaning of my question :) @@shakur07
7:20 Perfectly explained
this video was posted like 3 minutes ago lol
@@sock1 Used GPT to give the gist with timestamps skipping out all the redundant parts of the video.
@@ankitb3954how do you timestamp a video with GPT? 😮
@@sock1Remember you can watch videos at x1.5 and x2.0 speed. There are some extensions/add-ons which let you go x3 and beyond. He could've simply skipped or watched at x2.
@@lugh.i i know lol but it doesnt make sense that they skipped more than half of the video or put it on 2x, either way they wouldnt have understood anything. and they didnt do neither of those things anyway lol they used ai to give the general idea or smth
This was the first time I’ve seen professionals actually criticize AI instead of steadfastly rejecting the idea AI could replace them in any capacity.
AI still routinely struggles with and comes up with nonsense sequences for more synthetic languages or languages with a more complex sentence structure. It would be nice to finally see at least one study that is not done on English or Spanish.
AI won't be able to INTERPRET jokes, figures of speech, jargon, some technical language/terms, and emotions
Yet
Just like me fr
Nah, chatgpt is already able to explain jokes. Some tests where made where it was even able to explain memes and what the joke is
Some can definitely catch subtleness. But regardless, it will 100%, soon.
This is only time for this…it learned the hardest before
Why didn’t they give it something with terrible grammar or a heavy accent or proper nouns or toponyms?
10 years ago, AI translation was trash. Now, it is good for a series of uses. Let's see what happens in the next 5 years.
Oh, and ChatGPT is great for translating texts at light speed. When it comes to texts, the best translators will become revisors. The others will have to look for another job.
A.I is going to end up heavily influenced by opinion not fact. Give it time you will see. The whole world is running on opinions and no longer facts. It will follow the pattern/notion that the most noise and more popular thinking, will dominate the algorithms. Over time it will become more inaccurate, more diluted and require more maintenance through editing and refining by humans to counter balance this trend.
I don't get why it always has to be a VS thing? Why isn't anyone thinking about how we can make things more accurate and efficient by combining the two positively and efficiently instead of pitting against each other?
Barry, this is brilliant, am going to share! Bravo!!
is the version of Kudo they're using when the video was made actually using AI, or is it just doing machine translation?
As a graduate student majoring conference interpreting, I would say AI will replace most average interpreters because machines can do a much better job of understanding foreign languages, terminologies, accents, and so on. Imagine what will happen in the future since AI tools nowadays are not fully trained in all languages and specifically adapted for translation.
It would be cool to just have an AI tool that provides real-time accurate subtitles of what of what is being said! No more need to take notes!
as an interpreter myself, i agree with the profesionals on this interview
I would really have wanted to see the human translator performing the translation of the last (fast) speech. We were getting like 3 seconds. Too bad!
so far theres no AI that can cover all languages well, so translator and intepreter still stay
Great job guy's..
This is misleading, no one saying AI driving will replace F1 racing drivers, job replacement always starts at lower level jobs. Of course AI will not replace interpreters for now, but lower level streaming video translators will be replaced sooner than expected.
People will always prefer a human's mistake over a computer's.
wait 5 to 10 years
you wont even know whos human
@@Ghost-pb4tsI give it 2 years
Maybe not in 10 days or 10 months, but AI will gradually take some functions of our jobs some day.
I propose a TURING challenge. If for a pair of language most humans cannot tell the difference between AI and human translation..whether it is speech to speech or document..then AI has won. AI should be used in MOST situations for translation.
In other words, AI translators are good for people traveling on vacation asking shopkeepers how much a product costs. But, AI translators are not yet close to good enough for high-level use like in corporate negotiations or diplomatic councils.
They are not very accurate and become less accurate with regional accents and dialects
It'd be interesting to see how A.I. handles speakers of non standard dialects, accents, and people who have disturbances in their language production ex. TBI, a stroke, or Language Deprivation Syndrome.
Give it a few years, and it will be better than a human.
This was 5 months ago. AI has boomed since then. We get AI phone calls now.
The bigger concern is knowing whether it's AI or human, not the translation per se. If the intended listeners do not know, that could be an issue. We are used to being deceived by public speakers already, but this makes it doubly bad, whether it's a speech, news article, photograph or video. AI is indeed a tool, and any tool will always be used by those in power to gain more control. This has always been the case.
What headset is Barry, the 1st interpreter, wearing?
2:04 The way he pronounces the "h" reminds me of Stewie
I think it has already replaced many below mediocre interpreters.
Give it 5 years, these minor details and nuances will be smoothed over. I mean lets be honest, with the translation tools already in place pre-a.i-- it will only get better
Yeah there isn't really a threat to translators and interpreters because they lack the human element, they translate everything when you don't need every part of the sentence. Your whole job as an interpreter is to carry the essence of a sentence/message and make it understandable to the listener, cutting out unnecessary words but still retaining the core message/idea. And obviously the tone and emotion is difficult for an A.I to emulate accurately 100% of the time since it can't understand emotions like a human. Half the meaning of what you say is based solely on tone and facial expressions, that is how you know if someone is asking you or telling you to do something. It is a very helpful tool and can aid translators/interpreters but in most situations, it is not a complete replacement.
Aha, and you think that AI will stay the same forever? What will happen when the AI be able to carry those emotions?
Kudo uses software that translates in real time and is backed up my human translators. AI has not tackled real time translation, however, and if you wanted to reassure people that "AI isn't coming for your jobs", this would be an ideal test. Translation tasks for documents and text are already giving way to AI Large Language Models, like ChatGPT. They are very accurate for this task and extremely fast. It will not be a long time before LLMs can handle the speed requirements for real time translation, but it's dishonest to use this as the benchmark. It's an extremely small segment of the translation business market. Wired should have done better.
listening to this as a data scientist student ✍🏽✍🏽
I highly recommend the book Interpreters vs Machines, (Routledge, 2019).
Then, take a look at the FMRIs of interpreters while they are working and compare them with FMRIs of normal people speaking. And yes, I said "normal people", the brains of interpreters are being studied in several universities because they light up as christmas trees when working. I took part in one of those studies and the data will be very surprising for you.
Interesting. Can you share the study ?
I think it is just a matter of time. Soon AI will take over the interpreter's job. Believe me! It might take a longer time for some serious conversations to be trusted to AI's interpretation, as no one wants to lose any serious deal, but once AI proves it can do this job, (and it will be able by the time, by building more data) people will trust it with every interpretation, whether it is political, legal etc.
Thank you for this video!
Companies have got stuff cookin in their kitchens that are gonna improve on this significantly in a year or two
Hi, commenting a year later. Nope. Lol
There are NATURAL features to language that AI can't ever really replicate.
It is strange to me that Barry considers AI good enough for asylum cases that are also very important in terms of the life of the asylee hanging in the balance where a misunderstanding can get their case rejected.
underrated...
Hmmmm. AI's are basically bureaucrats -- unemotional, literal, robotic and intrinsically sociopathic. I hate videos that are AI narrated. Some of it is just awful.
Sorry...not buying this, even 6 mos ago - this is very old tech being demonstrated. The latest publicly available LLM technology will easily do better than humans. And certainly AI wins in non-realtime interpretations, simply due to the amount of material it can interpret, in addition to the quality. Not only can current tech perform the translations, but there are apps that can actually seamlessly replace the foreign speakers mouth movements, and body language, so that the speaker appears and sounds like they are personally speaking in the target language. Sadly, it's only wishful thinking that humans will continue to be needed for interpretation once people get comfortable with the new tech.
They don't get it, and they got the least advanced AI to perform this test. I think that a lot of people are so afraid of this themed that they don't want to see the potential that AI has, they think that it will remain the same when they point out a simple mistake, AI always evolves.
Wired is getting it all wrong! It’s not about AI vs humans. It’s about humans with AI vs humans. Like one of them said, AI got all the content covered in which it was not humanly possible. If that person could supplement his own translation with the AIs version, he would probably have a better overall translation with AI than his own efforts alone.
Remember 2017? "Ai will never be able to translate language to an useful level" - almost every translator. 2023 - "it will never be able to translate the sentiment". 2025 - AI can translate at least as good as the best of us
They just talk about the now and not the evolution, they point out mistake and they think that will remain the same forever
Exactly. For higher standards and higher important cases(e.g, diplomat, business), should never use AI . Because AI will never have human sense about complicated thinking
you need to try this again with new chat gpt 4o
chatgpt 4o is wining 😅
I have grown up translating English into Gujarati or my parents. And have also tried AI it’s not as good and lacks the cultural etiquettes requirements. ⚡️🤴🏽🕉🔺➕
It shpuld be called Artificial Stupidity.
just remeber that tecnology is advancing ...
This channel will always be the best. We will support them no matter what.
Is that nana on your pfp
Why do you sind like trolls, and who is we?
Great news. 💥Ai replace translater in anime industry
So far I rather the real life translator than AI🤷🏽♂️
I can't connect with AI emotionally.
And the speech of the president of El Salvador was pre-recorded and made faster during the editing process. So not fair to the real life translator.
A.I can sure do the job but you want to hear emotion and feeling, something A.I can't match at all.
For the moment
@@shakur07 shouldn't we do something about this as a human being for our future
Poor Olsen, Barry slaughtered them.
This is nothing short of extraordinary. I recently stumbled upon similar material, and it was breathtaking. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
In the human's closing statements, that's what a boomer would say. For balance and fairness, you should also have had the AI explain their thoughts as well.
AI has no thoughts.
AI are NOT real human beings
Dude name is Slaughter
AI will never be able to be smarter than the user providing the input into it
that's the thing. it will be faster, more productive, but never more complex in thought.
user providing the input into= internet
Illuminating
this colombian president is fun
Dwight v Dunder Mifflin Infinity
Where the AI seems to suffer the most is prosody. The AI interpretation of King Felipe’s speech does not sound compassionate or reassuring; it’s actually a little _unnerving_ to listen to.
Which, ironically, makes it more accurate since Felipe's compassion and reassurance is a put-on as befits a parasitic unelected aristocrat (son of an elephant hunter appointed by Franco who revived the monarchy) steeped in privilege for whom "his" people's suffering is an abstraction that he'll never experience.
I am so happy that most people are excepting blindly how “wonderfully” AI can do the jobs of human life. I applaud you sheeple for your “open mindedness”. That is all.
there would never be a human transater 2.0
but there would hundreds of versions of machine translation in the future (which would inevitably beat the humans)
The real question is why is that guy speaking so fast in the first place
Some languages are just spoken like that. It's up to the speaker to consider the context and the role of interpreters to slow down to make their job easy.
If interpreters would only translate words, most meetings wouldn't come to expected conclusions.
Interpretation goes beyond words
Now is the time to *hope that religion is right*
I do not really follow any religion, but if it really is the case that humans have souls and god created them and only god can create them, then only will A.I. not replace us.
Otherwise, it is very likely that A.I. surpasses us in all realms once we push to develop its sentience and consciousness (and we'll break the philosophical barrier that governs it aswell, no problem, because we're very persevering creatures, us humans).
its off-putting to hear the AI take short breaths while talking.
interesting
Hopefully, he can interpret the phrase 'Your services are no longer required' in multiple languages.
Not perfect,but are we?
though i love learning foreign languages, unfortunately i dont think there is a future for the professions of translation and interpretation. thats not to say demand will be zero or they they will never be needed, I just think AI being 80-90% as good as a human is enough to decrease the demand for these professions massively. and lets also not pretend like human interpreters are flawless either. if AI makes mistakes, it would be foolish to think humans don't.
The mic on this video is so gross. It's like asmr
Let AI interpret Trump
AI is trash.
"AI has been threatening everyone's jobs"
Literally no it hasn't what kind of weird Fox News headline is this.
Timestamp?
it hasn't, but it will.
Literally yes.
I know 🤦♀