Episode Transcript
[00:00:07] Speaker A: Welcome to the Week in review, your source for the latest news in the language industry. My name is Eddie Carrieta. I will be your host today. We have an amazing guest with us today, Shashi Shikar Vempati. I will introduce you to him shortly. But let's begin with the news that we have for this week. Let's get started with how the IBB translates important eclipse safety information in eleven languages. The IBB is the International Institute of Buffalo and has translated important safety information into eleven of the top different languages spoken in the region. The idea is that people have enough information to safely enjoy a once in a lifetime event. And I don't know about you, but I've seen a couple of solar eclipses and they are spectacular. So if you are a speaker of Arabic, Bengali, Burmese, Dari, French, Pashto, Somali, Swahili and Ukrainian, you will be able to get safety information to be prepared for this amazing opportunity in the United States of America. And more news. Well, you can see more of this information in Spectrum localnews.com, but gadgets 360, it's showing us information on inline text translation so Google Circle to search picture. It's updated so smartphones that already support Google Circle a little search bar at the bottom, which is a circle is going to be able to translate certain information. And as you can see on this article in gadgets 360, once you have the information, and Mila, maybe you can just click play on that video. You will be able to see that independent on the content that you have. You will be able to click on it and they will be automatically translated. And like I said, you can read more about this gadgets three all more news for us. The NIMSi 100 2024 was released last week. 100 plus companies done by five plus researchers. You can download the report for free and you should be able to see where we are as an industry and what is expected from the NIEMC team. Just as a matter of fact, you can see a graph in this link.
Nimsi.com Nimsi 100 LSP and in 2023, the industry did $67.9 billion in revenue. And of course, with the whole conversation about artificial intelligence and large language models, we should be able to see how the industry is going to evolve in the upcoming years. NIMC 100 is going to give you that information. So please go ahead and find it. Now let me introduce you an amazing guest that we have today, Shashi Shakar Venpati. He's the co founder of Deep Tech Bharat foundation and the CEO, former CEO of Prasar Bharati, of course, he has had many illustrious positions around the world. And he was a chairperson of external experts group in the University Grants Commission of India. He was a board member of the AJK Mass Communications Research center. He is also right now, as I mentioned, the co founder of Deep Tech for Bharat foundation. And I had seen a previous one which is vice president and member of the board for the Indian Broadcasting foundation and the vice president of the Asia Pacific Broadcasting Union. Without any further ado, let me introduce you to Shashi Shakar Benpati. Shashi, thank you so much for joining us today.
[00:03:59] Speaker B: Thank you. Very happy to be here. Very excited to meet all of you.
[00:04:03] Speaker A: Fantastic. And you consider yourself, and of course I would love to hear more about how you've got into your current position and situation. You consider yourself a technocrat, a columnist, a policy thinker and an author. Tell us about that. How did you get to become all of this? Where are you from and how did you develop this professional career?
[00:04:28] Speaker B: Yeah, thank you. I think. So if I trace my career path back, I started off in the it industry in India. I was with Infosys, which is a well known indian information technology company. And for several years I worked at the United States for Infosys.
Around 2012 2013, I returned to India. And that was around the time Prime Minister Modi, who was then chief minister Modi, was preparing for his campaign for the national elections. And I was part of his team which for the first time used digital technologies to engage the citizens in the campaign for the 2014 elections in India. And so that was the first transition from technology to public affairs at the intersection of technology and public affairs. And so I was running this digital media platform called Neeti Digital, which managed a part of Mister Modi's campaign at that time. And subsequently I was asked to serve on the board of India's public broadcaster, Prasad Bharati. And then I in 2017, took over as the CEO of Prasar Bharati. And that's how I made the transition from technology all the way into public affairs and public policy. And one of the flagship content programs on the public broadcaster is the monthly radio broadcast of the prime minister called Monkey Bath. And so that was the subject of my book, collective spirit, concrete action, which basically that's how I ended up being an author, talking about this book and.
[00:06:11] Speaker C: How it took, you know, radio broadcasting across India and various aspects. Perhaps we'll talk about some of that.
[00:06:19] Speaker A: Fantastic. And thank you so much because that was part of what we were talking just a few minutes ago. Backstage, and we were discussing specifically what inspired us having this conversation. So a few weeks ago, I read an article that you can find on the Internet, of course, and it says 2024 polls stand out for the use of artificial intelligence to bring language barriers. So in this article, you start digging a little bit deeper into how technology has evolved in the political landscape and the role that large language models, and more than large language models, artificial intelligence technologies are having in political campaigns.
[00:07:06] Speaker C: So can you tell us a little.
[00:07:07] Speaker A: Bit more about the evolution that you have seen in political campaigns around the world with the use of technology and how you land into the conversation about the use of artificial intelligence to bridge language barriers around the world, and specifically in India? Of course.
[00:07:25] Speaker B: Sure. So if you look at across the world, especially the western democracies, the use of technology has been very, very advanced over the decades.
I recall extensive use of, you know, social media, YouTube and Facebook over the years. Right now in India, the path has been kind of.
In India, the important thing to recognize is it's a billion people democracy. You're talking of nearly 900 million plus voters in India.
[00:07:58] Speaker C: You know how complex it is to communicate in India, given the language diversity, you have 23 plus official languages.
You have hundreds of dialects, which are spoken across the country. And so any kind of public communication is not challenging. And if you look at the history of elections in India with 900 million plus voters, the use of technology has gradually increased over the past two decades. In 2009, for example, when India went to elections, we saw for the first time blogs being used to communicate. So bloggers kind of played a role in the campaign. 2014, I would think, was the first election where technology was used in a very big way with the advent of YouTube for live streaming, the use of Twitter and Facebook for engagement. And we also had a digital platform for volunteers, students, professionals who otherwise would not associate with a political campaign to participate. So you saw a whole lot of use of technology in 2014 and 2019 kind of took it to the personal level with WhatsApp playing a significant role in a person to person messaging, where a lot of the campaign messages were disseminated using WhatsApp.
What is interesting with 2024 is artificial intelligence, as you rightly pointed out, with large language models, generative AI, all the buzz around them. So one of the use cases that I wrote about is how Prime Minister Modi is using AI to have his speeches delivered, which he delivers in Hindi, translated into multiple indian languages.
And that has been very important, because what typically happens is his speeches would get translated with a significant delay where manually someone would have to, you know, translate, transcribe the speech, translate it, and then dub with, you know, the local language and then have it, you know, broadcast in that particular region. So that would introduce significant delay, more than, you know, 24 to 48 hours to get his message out. Now, with AI, he was able to turn around very, very quickly. And so by the evening, when he delivers a message, the AI has generated his speech in five, six, seven languages. And those versions are out on social media almost immediately. So that's a huge shift. And I think that is where I see AI playing a significant role for the first time in the general elections.
[00:10:38] Speaker A: And let me understand a little bit more about how the process was done before. So before the speeches were recorded and then they were sent to interpreters, and were those speeches dubbed into a different language or just the transcripts were released? How did that work before and how is it working now?
[00:10:58] Speaker C: So let me give you the example of the monthly radio program, because that is where it all started. Prime Minister's Modi's monthly radio program, which is a radio broadcast which lasts for 30 minutes, would typically air on the last Sunday of every month.
And this 30 minutes broadcast would be at 11:00 a.m.
Very fixed time slot every last Sunday. And he would do that in Hindi. And what would happen is that particular 30 minutes monologue or dialogue in Hindi would get transcribed and translated into the 20 different languages, and the interpreters would then have a presenter readout that transcript in those languages, end up on the original video, and then the language version is produced. And then this would be then broadcast in different radio stations and television stations across India with that specific language audience in mind. So that is the process, how the radio broadcast started. Now, when it comes to political speeches, because they're not scripted, they're largely impromptu. The challenge is even more unlike the radio broadcast, which, where there's a, you know, prepared script that he more or less takes to with some improvisation. But the political speech is. It's rhetoric. You know, it's part of the movement stuff, right? So, so the challenge is that you've got to listen to all of it, transcribe it, and then translate. So that introduces significant delays. And so the process would take, you know, more than a day or day and a half. So now that whole thing has been crunched with AI, and I think that is a huge, huge benefit in terms of being able to get your message across without barriers.
[00:12:43] Speaker A: And let's talk a little bit about the relevance of the content rather, rather than the relevance, the omeletic part of it, which is how people are receiving it.
Now that you're using artificial intelligence, are you using tools that also emulate a bit of the tone, the original tone of the speaker and the emphasis that the speaker does? How is that working so far?
[00:13:13] Speaker C: Yeah, I don't see that happening yet, though. I know I've seen tools which can do it, and I've seen tools that do it fairly well. But given the sensitivity around campaign messaging, they're not yet experimenting with those features yet. But I wouldn't be surprised in the times to come that they start using that. So, for example, today it is just a voiceover. While you have AI tools like OpenStream and others who can actually do the lip sync and make it look like you're talking in different languages.
So they're not doing that, they're just doing a voiceover.
Because there's also this other concern. PM Modi has spoken about this multiple locations, including most recently at an AI summit which was held in India about deepfakes, how he himself was a victim of deepfake video. So it becomes very important that you retain the original video and the original soundtrack in the background while you do a voiceover with AI in the language that you want the video to be presented.
[00:14:15] Speaker A: And we'll dig deeper into the risks because I think that's going to be very important, as well as the benefits that we receive from this. But right before we do that, I want to ask you a little bit about what you think and what you've seen. The role of interpreters. It's becoming, because a lot of the fears that we see in the language industry is that then all of a sudden, interpreters now don't have a job. It's been argued that, in fact, now they do have a very important job, which is to guarantee that the interpretations are done correctly, that the messages and the meaning is being transferred. But from your experience of what you've seen so far in politics, how is this evolving?
[00:14:54] Speaker C: Yeah, I think that's a great question.
I think you brought out a very important point. Does this make people redundant or does it give people more capacity, more ability to do their job? There was a recent interaction between Prime Minister Modi and Bill Gates where he brought this aspect out that AI, it is not about AI versus humans, but it is about humans using AI to expand their ability, be more creative, more, you know, more efficient and more productive. So to my sense, the interpreters job is now far more important, as you rightly pointed out, because now they have to make sure that the algorithms are doing it correctly, doing it effectively, especially in context, where if you do a literal translation, you lose the context, you lose that, the intent behind the messaging, and it's very important to capture that. So it's not always a literal translation. You will have to use a metaphor. You'll have to use something which is very contextually appropriate and relevant. And that is where I think the creativity of the interpreter becomes very important. Making sure the translation is right as well as the context is not missed out.
[00:16:03] Speaker A: And if you can tell us a little bit more about the relationship that you've seen between translators, interpreters, and politicians and broadcasters and communication agencies in India, what has been the relationship like? If you can attest to something like that, sure.
[00:16:21] Speaker C: It's a very historic legacy.
And that legacy goes back to all India Radio, the radio broadcaster, public broadcaster of India, which, incidentally, is several decades older than modern India itself, because it was radio broadcasting started well before India became independent.
And radio historically in India has spoken in multiple languages. All India radio broadcasts in more than 100 languages and dialects across India.
So they had this long tradition of people who are what are called readers and translators. So they would specialize in reading out text in different languages and being able to interpret and translate very quickly into the specific language of their expertise.
And this has been a very institutionalized mechanism within all India Radio for several decades now.
And that is the mechanism, the infrastructure that was utilized for the monthly radio broadcasts of this prime minister and even in the past, several other prime ministers, when they do their broadcast to the nation, this is the mechanism that would be used to translate their speeches across the country.
So it's a historic relationship in that sense, and it's also a skill that is, over the years, with the advent of technology, it has kind of diminished. But now languages are again being rediscovered with artificial intelligence. So it's a very interesting moment in time where, especially with a lot of sensitivity on preserving culture, preserving languages, I think technology is coming to the aid of societies to preserve these languages. And so some of these jobs, which had become less fashionable as being a radio translator, is now going to be all the more interesting and important.
[00:18:27] Speaker A: That's very, very exciting, especially for the community of interpreters and translators that had thought that artificial intelligence or large language models were all of a sudden going to eliminate their jobs. I think I agree with you. Their roles have become much more prevalent. And this talks a little bit about the risks and benefits, and we've talked a little bit about the benefits. If you want to dig deeper we can do so as well. But those are very obvious from your perspective. And, of course, you have a technical background, a communications background, and a high level view of things background as well. What do you think are some of the things that we could do? Well, first of all, what do you think are the risks of this? And you've spoken a little bit about deepfakes, and, of course, that the biggest risk, which is putting information out there, that's not fair. That's not really what people are saying. So what are the risks that you see? And perhaps there are some that we have not touched upon today. And what are the technical things that we could do to avoid those risks? And what are the human things that we could do to also work on those risks as well or mitigate those risks?
[00:19:34] Speaker B: Sure. Sure.
[00:19:35] Speaker C: So one of the significant risks that I see is, of course, apart from deepfakes and malicious content, it is the inherent biases within these models. The last language models, if you look at the models that are dominating the discourse today, they've all been conceived, designed, trained and developed in the west, and they don't necessarily capture these cultural sensitivities. The local context, which can be very unique and very different. Like, for example, in India, you know, every 200 cuisine changes, the dialect changes, and these nuances are not necessarily captured in these last language models, which have been trained on data which is in the public domain on the Internet. And so the sheer fact that a lot of local content is not necessarily digitized and it's not necessarily out there on the public Internet means that several of these models have missed out on being trained on those contexts. And I think that is a big risk and a challenge. Now, there are several efforts in India to address that. The government has created a platform called Bhashini, which I mentioned in my article. There are several leading institutes in India which are developing these language models. But there is a challenge. If you look at the kind of infrastructure, compute infrastructure with the GPU clusters that are needed to do these kinds of large models, they're not necessarily available everywhere in the world. India is only now building up that capacity. So there is that divide that is already creeping in. And in the past, we would talk of digital divide, and we would talk about Internet divides and so on, where access is not available. But with AI, I think it's a much more significant gap because the technology is feeding on itself. It's exponentially improving. Right. So how do other societies play catch up? And if a large democracy like India itself is seeing this gap, what about smaller nations? What about Pacific island and all of the language diversity in all those communities?
So that is, I think, a significant risk that I see from a generative AI standpoint and the, that we've taken.
The other risk would be more in terms of skills and skill enablement.
And I think, again, that's where it's very important that a lot of content gets technical, content gets created in different languages, and we are able to train and enable people. One of the institutes that I was involved with, which you mentioned in my introduction, is the University Grants Commission, and they're looking at using AI to create massive online courses which can be delivered in different languages for different personalized learning journey.
And that can be a big help for the skill enablement that we're talking about. So use AI to train people on AI.
So we'll have to get creative on how we mitigate these risks, handle these challenges.
[00:22:49] Speaker A: Fantastic. And what do you envision or want? And I know we're getting here in the realm of speculation and opinions, but what would you want technology to go in terms of, of course, enablement of information to be transferred in many different languages. What would you want to see or what do you envision things are going to go to in the next few years?
[00:23:12] Speaker C: Well, the pace of evolution of technology is very rapid, and we are most of the time playing catch up to keep pace with where technology is.
So one of the things that I would really like to see is greater global coordinated efforts, because just like there's all this effort to coordinate actions around climate, I think AI also merits that kind of coordinated local action. Otherwise you will have societies that will be left out, communities that we left up. So to me, I think that would be the most important thing from a technology evolution standpoint, that there is greater global coordination. And if you look at the current approach of the large model, that is massive data centers, massive computing, lot of power being consumed.
So this is not a sustainable model. So ultimately, I think AI will have to come down to the edge, to the devices, to your mobile phone, your tablet, your laptop or your wearable device. So we'll have to look at smaller targeted models that are optimized to run on your personal devices. Because once you come to the rest of the world, which doesn't have great Internet access, which doesn't have massive computing power, to be able to empower them with the benefits of AI, we'll have to bring the technology down to their personal.
[00:24:40] Speaker A: Shashi, thank you so much. This has been great.
If you have any questions, anyone looking at us on LinkedIn YouTube. If you have any questions for Shashi, this is the moment to share those questions. I'm going to switch gears a little bit here. You are also an author and we are going to be sharing your book, collective spirit, concrete action. Can you tell us a little bit about how the book came to be and what were your intentions behind writing this book?
[00:25:07] Speaker B: Sure.
[00:25:08] Speaker C: So the radio broadcast by the prime minister, the monthly radio broadcast, it started in 2014. And last April, April of 2023, when this book was launched, we had hundred episodes of this monthly radio broadcast.
So this book was intended to take a look at those hundred radio broadcasts and put them into perspective in terms of India's developmental journey. Because what is unique about this program was that there is no politics in it. He has kept the focus of this radio broadcast completely on development and on societal change in India and how he did that with communicating through local metaphors, cultural metaphors, talking about different languages, talking about poets, authors, iconic figures of different communities, and using those examples to motivate change, be it on environment, be it on, you know, saying no to plastics, being on water conservation, forest conservation. It's a whole bunch of issues. And I think that has been the focus of this book. What is interesting is because this program is translated and broadcast in multiple languages. It was a very clean and high integrity data set to train models. So you had researchers at IIT Kanpur, which is one of the premier institutes in India II Hyderabad. They collaborated using the monkey Bath corpus of audio and text in multiple languages to build AI models. And that is the basis of what we're seeing today with these real time translation of speeches. So one of the chapters in the book talks about that as well.
[00:26:58] Speaker A: Shashi, thank you so much for your time today. Sashi is the co founder of Deep Tech at the Bharat foundation, author, as we just saw, of an amazing book, and also the author of the article 2024, Paul's stand out for use of AI to Bridge language barriers. You can go ahead and read it from his LinkedIn profile or ask us for the link and we'll find it there. And definitely we're gonna send Shashi an invite for writing a multilingual magazine. Tracy, thank you so much for joining us today. It's been a pleasure.
[00:27:28] Speaker B: Sure.
[00:27:29] Speaker C: Thank you.
[00:27:30] Speaker A: All right, everyone, thank you so much for joining us here in the week interview. Remember to leave your comments and questions below and share this content with your contacts. You can also join multilingual.com to receive our weekly newsletter, the weekly Digest, and keep yourselves safe. The weekend reading multilingual magazine, the April issue, the digital version coming out today. My name is Eddie Avieta. Thank you so much, Mila Savogal, for the production of this show. See you all next week. Bye.
