Episode Transcript
[00:00:04] Speaker A: And we are live. Welcome to the week in review, your source for the latest news in the language industry. I'm your host, Eddie Arrieta. Let's jump right in. And we will begin with highlighting our weekly digest, which will work out today. We have three specific news we're going to be diving in. Mila, if you can augment the site so that everyone can read the three news. We have one regarding interpreters and a lawsuit around that. We will also be talking about climate, climate survey. Mila, can you put it back in? Yeah. Thank you so much. So alleges court interpreters, identity theft. We will also everything you need to know about subtitling and climate survey. We invite everyone to join the survey and we will go from there. You can also check our news outside. News Motor Language Day was celebrated this past week, which is a celebration promoted by UNESCO since the year 2000. It was an initiative of Bangladesh, celebrated all around the world. You can read more about this in languagemagazine.com. Also, Kansas City language access.
It's improving.
One out of every 20 Kansas City residents have limited proficiency in English, and the city council is looking for ways to improve those that have limited proficiency in English, which is around 20,000.
And you can read more about this in kcur.org, which is the official Kansas City NPR member station. Also, the Pentagon exploring military uses of large language models. So Washington's top military AI officials are gathering with industry executives to learn what are some of the things that they can do with this. Researchers are saying that some potential uses of the military are training unsophisticated wargaming and also helping with real time decision making. But in general, it's going to be used for summarizing conversations, especially in the battleground. We know how difficult this could get. So those of you, if you have any comments, please make sure that you leave them. We are live right now on LinkedIn, YouTube and Instagram. Mila, who is helping us in the back end. Mila, if you see any comments, please put them in the front so that we can see them. You can also check out press releases and hello, Jennifer. Which you can also check out press releases on the news from this week and multilingual.com press releases. And you will see their news from blend and their acquisition of manpower language solutions. You can see e arabization opening offices in Riad, and you can see also safe AI. We're going to learn more about save safe AI today. So without any further ado, we want to welcome everyone in the audience, and we also want to welcome our guests who are going to be joining us today. So we have Bill Rivers from the safe AI, let's, let's call it task.
[00:03:35] Speaker B: Force safe AI task.
[00:03:41] Speaker A: Both part of this amazing task force welcome today in this space. And thank you so much for joining us. How's the day going?
[00:03:52] Speaker C: Great.
[00:03:54] Speaker B: Yeah, very good.
[00:03:55] Speaker A: Excellent.
[00:03:56] Speaker C: Thanks for having us.
[00:03:57] Speaker A: Absolutely, absolutely. And I want to get started. We can do this with Mila first, and then Bill, can you tell us a little bit more about your background, your professional background, and how did you landed at the safe AI task force so that the audience can start learning a little bit more about that?
[00:04:13] Speaker C: Well, I started working in the industry.
Actually, my first interpreting assignment was in the late eighties, believe it or not. And then so I started as a professional interpreter and a translator, and in 1993, I founded Masterword Services, which is a language services provider. And so since the beginning of my career, I've been very involved with different professional organizations in the industry. And the way I got involved with safe AI was a group of industry professionals. Veterans, I would say, were on a call when the whole hype about AI was happening in the spring, and we said, you know, how do we respond as an industry to this?
And then now let Bill tell the next steps of the story, how we responded.
[00:05:11] Speaker B: Yeah, like Mila, I've been in the industry in one or another way for a very long time. My first interpreting assignment was also in the tail end of the Cold War, 1988, Russian, English, English, russian conference interpreting.
So I've been a lobbyist for the last twelve years. I've done extensive work and standards. I was a CTO of the company. So I've done a lot of different things, sort of have kind of CEO add, as I think a lot of us do.
But yeah, we have this call and there's all this hype around chat GPT in March of last year, February, March, and a lot of folks coming back from the gala meeting saying, okay, we're going to be displaced by AI. That's what all the technologists are saying. And, you know, my personal response was, well, let's kind of think that through, right, translators have not been replaced by technology. We're not replaced in the last 20 years, but the technology changed how that work is done, changed how localization is done. Can we think through what the possible impact of AI might be and where it might make sense, what pieces of the workflow, what situations, and can we pull enough a diverse body of stakeholders together so this isn't just, say, interpreters and language companies looking to defend their turf. Not that that would be what we're doing, but that would kind of be the perception. But can we bring in, you know, can we bring in the customer end, you know, the medical providers, the healthcare systems? Can we bring in regulators? Can we bring in academic experts? Can we bring in some of the technology vendors themselves? And in varying degrees, we're able to bring an enormous group of stakeholders together. We put out a survey for volunteers, had more than 600 people volunteer, created from that, a stakeholder assembly to kind of serve as the oversight of the effort. And there's more than 50 people on that. There's an administration subset that does meets weekly to make sure everything's on the rails and developed a roadmap of what we were going to do with the end goal of, you know, we know that this is going to be regulated. We know that the Biden administration has put out an executive order. We know there's a law in the U on AI. We know that the Department of Health and Human Services, for example, is looking how, looking at how AI is going to be used in healthcare.
Well, let's make sure that they hear from us, they hear from the language experts about what needs to be done around the use of AI in our space, whether that's in language access, healthcare, education, legal system, whether it's conference interpreting, et cetera, and really looking at interpreting because at least in my view, that the impact on the translation and localization world kind of, you know, swallows the technology whole and integrates it very quickly into the workflow. I mean, I'm oversimplifying, but that's when you step back and look over the last 20 plus years that it's a sort of a different space for different reasons. And, you know, translation versus interpreting, even though I've done both and even though in some languages, like Russian, it's the same word, Pirivood, you know, you can say, you know, it's just one word for translator and interpreter that, you know, how would this play out in interpreting? Right.
[00:08:34] Speaker A: Thank you so much for sharing that insight.
And you mentioned 600 people, 50 people. Could you speak a bit about the texture of the comments and the reception of the initiative? What are those that are within saying, and I know, Mila, we've changed a few messages, and I see the passion and I see the speed at which we are communicating. Maybe you can speak a little bit about what are the interpreters saying, those that are joining and having these discussions, what are the words that they are using to.
[00:09:05] Speaker C: So first of all, we at the very beginning we wanted to involve, and we identified eleven stakeholder groups. So interpreters are just one of those eleven. We also wanted to go to the end users and consumers. And the whole idea behind the task force is not going language professionals against the technology, but how do we embrace the technology responsibly and with accountability? And bill mentioned the translation industry kind of taking technology very, very quickly, and it's true, but translation industry is also a lot more mature in the way, has a lot more years behind. It's felt, in a way, in adopting machine translation, newer machine translation and text. And translation can be reviewed after, in interpreting, text cannot be reviewed after it's live. And if a particular error occurs, depending on the setting and the risk, it can have very dramatic outcome. Let's say if medical interpreting is handled incorrectly, a patient can die. So the overall perception, I believe, is that technology will change the way we work.
Technology is welcome because it's actually going to expand language access in places where it has never been provided before.
And my personal view, not the survey view, I think is going to exponentially expand language access.
And then we're looking at ways of what is the responsible way. How do you have accountability for those high risk scenarios? I don't think anybody doubts in the fact that in low risk encounters, AI and machine interpreting is welcome. And then there's a second point that was not covered in the survey, but I'm going to bring it up because many people hear me talk about it. And that's marginalized languages, the other half of the world where their language is not even digitized in many of cases. So then what is the digital disparity we are creating for the other half of the world if they don't have access to it? And I'm going to. I think, Bill, it would be great if you can answer the same question because you have.
[00:11:33] Speaker B: Yeah, it ties directly back into mother language day, which was yesterday. For seven years, I was the president of the board of directors of 7000 languages, which uses a licensed technology that was very generously given to 7000 by transparent languages, a company, New Hampshire, that builds online language learning. And we've done 50 different courses, 50 different languages, roughly, of endangered languages, First nations, native american and other endangered languages around the world.
But the question of data and training for large language models is actually a big deal. And there are languages where there's very little data available, very little, whether it's online or not. There are certain languages of the austronesian languages of the Philippines where the only extant data are translations of the gospels. Well, that's if you're going to train a language model you're dealing with now. It's something, a rather limited vocabulary and a limited register, a limited set of subjects that it doesn't get you the kind of richness of data. Now, I know Google claims it's got an approach.
You have to see how well that works. Mila mentioned a survey. And so what we did, we got together, we did our first survey to get volunteers in an incredible response, global response, primarily from the US, but still, we had people from all over the world involved. And then the safe AI commissioned a survey, a perception survey, that was also then accompanied by a qualitative research effort by the Deaf advisory group to safe AI. So the deaf, we invited the registry for interpreters of the Deaf in the National association for the Deaf, the Council of Interpreter Training Programs, which focuses on ASL interpreter training. And we have a number of prominent deaf american scholars and computational linguists and ethicists and philosophers involved. And there's now a deaf advisory group, because the issues are, there are some analogous issues, but there's some very different issues in terms of language access for the deaf and hard of hearing. They did a qualitative study of some more than 100 hours of webinars and interactive forums on AI in their world. In the deaf language access world, common sense advisory did a survey with 2700 plus respondents on perceptions of AI in interpreting among those eleven stakeholder groups, including translation into the top ten most common spoken languages in the US, outside of English. And you see, you know, we were looking at how is it trusted and in what cases and what is the, are people using it? So, you know, it's early days, the adoption levels are pretty low.
One of the drivers for adoption, when you look at the survey data, is clearly cost, when you know, the customers have that front of mind, because language access in particular is an unfunded mandate in the United States. And so it has to be done by law, under title vi of the Civil Rights act and under section 1557 of Obamacare. But it's not paid for, so that comes out of overhead. Okay. But then when you look at what's trusted and what's trusted most in terms of AI technologies are the sort of real time automated in language captioning. Like if you're on a Zoom type product and you're getting the transcript done by Otterpilot or one of the other apps, everybody's speaking English, and the transcript in English comes out, you know, pretty good, and then it goes down from there all the way to. I think there were four, four categories of technology that we looked at to real time speech, to speech, which is not trusted very much and where the perceptions of quality aren't very high. And then you get into, okay, well, what do we mean by quality? That's a big question. What do we mean by accuracy?
Something could be 98% accurate, but if that 2% is. Oh, oops, we removed the wrong kidney. The tumor was actually on the other one. Okay, that's a problem.
Or an example that one of our colleagues, Gabrielle Lemoyne from Span, a language advisory, posted an article yesterday looking at the argentinian president Millay's speech in Davos, where the AI solution translated everything pretty well, but had missed a very important and subtle nuance when he was talking about collective economic action that might lead you to believe he supports it, where in fact. Whereas, in fact, when you listen to the Spanish or you listen to the human, now the human, you know, the english interpreter, now, it's a little bit far fetched because any, anybody that's making decisions based on that, say the US government wants to think about its policy towards Argentina. They're going to have humans involved, at least now.
But if you look at it from purely point of view of accuracy, it could be seen as very accurate but still misleading. So the survey tries to get at all of this, and it's an enormous amount of data. There will be two webinars. Right. Neil, I think you want to talk about those next week where these results from both the quantitative survey and the qualitative research by the deaf advisory group are going to be presented to the public.
[00:16:47] Speaker A: Yeah, and I love to hear more about the two events. Mila, I want to take a second to lead everyone to also know that we have an article. We are sharing those in the comments of our social media so that you can read the information on the two events that we're just going to talk about. And Mila, after giving us this recap of the events that we're going to have, it'd be amazing if you can also tell us a little bit more about the objective and also the reason to be the raison desert of safe AI. Is it about the safe use and embracement of the technology? Is it also looking to protect the interpreters? Is going to be my question. But let's dive into the events because those are very exciting and I'm probably going to join as well.
[00:17:35] Speaker C: So the events are going to be on February Central. There'll be an event and it's posted in the information releases and press releases that you guys, in the article that you're publishing. It's also posted on safeaitfourtaskforce.org. So safeaitf.org dot so the first event is going to have mainstage presentation is going to be Helene P. Almir with CSA Research. It's about a 350 page report with a lot of data points that she's going to talk about.
And then the second event is March 1, and the center stage will be the Deaf advisory group who have conducted an incredible study, qualitative study, with a lot of information, a lot of new information. Actually, I've been in the industry for a long time. Some information is very new and different than expected. So expect surprises.
And that will be on March 1, the following day.
The purpose of the task force is not to promote one group over another. Not to say technology is better or humans are better or this or the other. The purpose is to get a perception and have a way or guidelines or recommendations so that we have a guidelines committee that is working hard now that the results came out to ensure responsible adoption of new technology. There's no question that this technology is being adopted, but how do we adopt it? With responsibility and accountability that we can measure. And how do we have this enter into that new age where it's a combination of a human and a machine? So we'll be having a lot more press releases, a lot more events, because right now it's the results of the survey and the study, and the next one will be results from the guidelines committee that's coming out, and there'll probably be a lot more next steps.
So please stay tuned as an entire industry of the spoken world is changing in a way.
[00:20:00] Speaker A: Thank you so much. And I do have a follow up question regarding the purpose of safe AI, and I hope you can both bear with me on this one. And I'm thinking, what is a safe use of AI considering interpreters? Are we thinking of all the elements about the safe use? Are we thinking about the technical elements of the use of the technologies and embracement of the technologies? Are we looking about the ethical components of it as well? And we were talking about it right before the show with Mila, who is in the backstage. Her name is also Mila, but from Camila, not from Ludmila, but Camila. And the discussion we were having was all the interpretations that have been recorded around the planet can be used to train artificial intelligence models. Are those interpreters being recognized for their effort? So I posed a question to you that have already, of course, looked into this more deeply.
[00:20:59] Speaker B: So that's actually a huge question for large language models in general. And there are several lawsuits that have been brought against some of the developers of large language models because they are using other people's data, presumably with, you know, these are for profit companies. So they're going to use these data for a reason, without attribution, without payment for the data, and that's going to play out over the next several years.
I think Google now gives attribution of where things come from, which is actually very helpful when you're asking.
I think it's called Gemini now, to write a one page around language access. And the first time I asked chat GPT to do that was just to see how it worked. It ended up quoting something to me verbatim that I had written years ago in an academic publication and not attributing it to me.
We know that that's an issue and that's a broader issue, not just for large language models, but the idea of data being this important driver for the Internet economy. Well, who owns the data? And that comes back to the interpreters, right?
And just as much as it does for translators. When you're building a translation memory for a client, you know who's owning that? If you're using, you know, we know that machine translation solutions often end up having to be put into walled gardens because the client does not want their data being used by the bigger machine translation engine to train. Right. For patients. This becomes a real issue because there's federal laws in the United States, there's HIPAA, there's GDPR in Europe. There may be patient privacy laws in other european countries. There's a very strong patient privacy law in the United States that the AI solutions are going to have to respect, because that's the law.
All of these questions are really complex and all tangled up together. But, you know, we do want to focus on the safe and ethical use. And you ask, what are the safe uses? Part of that is we need to put into the center of our attention, we need to focus on the actual people for whom the interpreting is being done right. Not necessarily the interpreter, not necessarily the company, not necessarily the healthcare agency or the court system, but the individuals whose languages, whose speech is being interpreted.
What is the best thing for them? And I think that's still in some ways an open question. We don't have good answers for you. We're starting down this road with the perception survey, with the qualitative work that the deaf advisory group has done. We don't yet have, because it's such a new technology. In some ways it's a new technology, some ways it's not. You know, we've been talking about statistical machine translation, neural machine translation, large language models, and evolution of this technology over the last 30 years. But it's so new as an application and so new in and of itself that we don't yet have good answers. But, you know, if I put it in the context of language access in the United States, something that I do a lot of work on as both a consultant and an advocate and lobbyist, you have to put patient safety and privacy first. And in my view, AI has to prove that it is appropriate for that, that it's, you know, is there going to be a comparison with humans? Now? I get pushback. Sometimes humans make mistakes. Well, yes, we do. I had a couple times where I've stumbled on a word or misspoken. I catch it immediately. I backtrack, even in this conversation.
But we are capable of recognizing mistakes. Don't always ask. Ask my family. Capable of correcting them. Again, don't always do that. Whereas the AI is an algorithm doesn't know it's made a mistake unless you tell it. And that means that immediately, logically entails having a human involved, human in the loop, human in control, something that people like Yoseci have been talking about for 25 years with Mt.
So I don't want to equivocate, I don't want to sound like I'm talking around it, but it actually is kind of a big question, what are the safe uses? And you can sort of see emerging from the perception survey, places where it would be trusted and wouldn't be trusted. That's kind of the first place to start.
[00:25:16] Speaker A: Thank you.
[00:25:18] Speaker C: I can jump on, summarize it and say the name. It's stakeholders advocating for fair and ethical AI in interpreting. So a brief answer to your question is all of the above, whatever, you know, all of the components of that. That's what we're looking at.
[00:25:38] Speaker A: Fantastic. We're going to wrap up in just a couple of minutes, but I want to take a second to thank everyone who is in the audience in our different social media. If you have any questions for Mila and Bill, this is the time to do them. We're going to be wrapping up soon.
Where do we want this to go?
What are the ultimate expectations of the task force, if I may say? And I know this is probably an ongoing conversation, and it will be forever happening, and we will want policy and we will want different things. But what are some of those expectations that you have?
[00:26:14] Speaker C: So I think one of the expectations is because this is a technology that's evolving very, very fast. The important thing is that as an industry, we are on the same page and we are reacting off the same page together. And so safe AI gives that forum and the foundation for everyone to come together and be kind of figuring out how to safely drive a car with at least a foundation behind it. Unified foundation.
[00:26:46] Speaker B: And, Bill, what's, I take that even a step further. We want to be in a place where we're providing thought leadership, where this is not driven by the technology, where we are not in a position of reaction.
Again, I look at things through the legal and regulatory framework of language access, and I can see immediate applications to federal regulations around language access, for example. But that gets kind of both down into the weeds and there may be intermediate objectives along the way.
But trying to shape this conversation in such a way that it's not driven solely by the technology vendors, nor is it driven solely by people reacting against them. I'm oversimplifying tremendously, but that's sort of where I would start in terms of, you know, what our purpose and goal is, where we are.
[00:27:40] Speaker A: And that is, that is really great to hear. Can you remind us just to end the conversation, where can we read more information, the website and also the dates and times for the upcoming two events.
[00:27:54] Speaker C: So it's www. Safe a I T, as in task force, you know, tf.org dot safeaitf.org. And the events are on February Central and March 1, 10:00 a.m. Central will be, two reports will be presented.
[00:28:21] Speaker A: All right. Thank you so much. Any final words, Bill and Mila, you want to give to the audience?
[00:28:26] Speaker C: Stay tuned.
[00:28:28] Speaker B: Exactly. Stay tuned. If you want to get involved, you can go to the webpage and send the, there's an email link. If you want to get involved. You want to learn more, you want to serve on one of the committees. Always happy to have people. It's all volunteer. We've been very lucky to have a number of sponsors, large and small, that have supported the survey, but none of us are getting paid to do this. This is all a volunteer effort trying to make our industry a better place.
[00:28:55] Speaker A: All right. Thank you so much. Bill Rivers from WP, Rivers and associates, principal there, and Ludmila Golovine, president and chief executive officer at Master Word Services, both part of the Safe AI Task Fork initiative. Thank you for being with, here to be here with us today and to wrap up. We want to thank our guests, of course, and our audience for joining us today. Catch up on more news press releases from companies like eravisation blend and Memoq by visiting multilingual.com. This has been the week in review. Remember to leave your comments and questions below, and subscribe to our channels for more language industry updates. Until next time, Goodbye.
