2951: RWS Group: How AI and Human Synergy is Enhancing Creativity and Efficiency

[00:00:00] Have you ever wondered how AI is reshaping the landscape of language services and content creation? It's something we're hearing more and more about in our news feeds. Well, in today's episode I'm joined by Maria Schnell, the Chief Language Officer at RWS Group. And

[00:00:20] I want to take everyone on a journey into the fascinating world of AI and its integration into language services and uncover how this technology is not a replacement for human talent, but rather a powerful tool that can enhance our capabilities. Maria is going to

[00:00:40] share her insights on RWS's innovative machine-first approach, the critical role of human oversight in ensuring quality and ethics, and also the importance of embracing curiosity, experimentation to truly leverage AI's potential. But before I get today's guest on, I want to give a big

[00:01:01] shout out to the sponsors of The Tech Talks Daily Podcast because they help me pay the bills that come with a daily tech podcast and hosting fees of 365 interviews a year. And also shine a light on how legacy managed file transfer tools are dated and lack the security

[00:01:19] that today's remote workforce demands. And companies that continue relying on that outdated technology put their sensitive data at risk. But the good news is you can ensure top-tier security with Kiteworks MFT Suite. Their software security hardening includes an ongoing bounty program,

[00:01:36] regular penetration testing to minimize vulnerabilities, the convenience of a one-click appliance update, and avoid the hassle of managing multiple unhardened servers. And as an ex-IT guy, I know the benefits of that. So why not choose Kiteworks for the highest level of security in

[00:01:55] managed file transfer? And remember, Kiteworks is also FedRAMP Moderate Authorized. So step into the future of secure managed file transfer with Kiteworks by visiting kiteworks.com to get started. That's kiteworks.com to get started today. But enough rambling from me. Let's get today's guest on.

[00:02:14] So buckle up and hold on tight as I beam your ears all the way to Germany, where we're going to explore the symbiotic relationship between human and artificial intelligence and how it's paving the way for a new era in content creation.

[00:02:33] So a massive warm welcome to the show, Maria. Can you tell everyone listening a little about who you are and what you do? Sure I can do. First of all, thank you for having me.

[00:02:43] My name is Maria Schnell. I'm the chief language officer of the RWS group. What that essentially means is I'm in charge of everyone producing language. And what I mean by that is we have a

[00:02:55] large team of in-house linguists. We work with a very large network of external translators that support us, but it really also covers all of the other things that have to happen for content to be localized. So things like software organization, testing engineers, people who do voiceovers,

[00:03:13] subtitles, and essentially all of the people that help us lay out as well. So anything that helps us to truly transform content in all of its dimensions for another language or another culture is part of my team. A team I'm very proud of, very united nations.

[00:03:32] I'm glad that you've sat down with me today and agreed to join me on the podcast because every single day I always explore a different area of how technology is impacting a business. RWS has got this huge reputation for being the world's leading provider of technology-enabled

[00:03:49] language content management and intellectual property services. But I've got to ask, how has the perception of AI evolved over the past year for you at RWS? Especially in relation to fears of it replacing human workers. It's a narrative we see a lot on our news feeds,

[00:04:07] but from the conversation I'm having, quite the opposite seems to be happening. But I'm wondering what you're observing here. To be perfectly honest, it feels like the rest of the world has caught up a lot more than anything else. So particularly in the language

[00:04:21] experience delivery team, which is my team, we have always essentially had a machine-first approach. Well, not always. Since 2017, we have a machine-first approach to localization. So we've been interacting with particularly neural machine translation engines for quite a while,

[00:04:38] which essentially means part of the change curve that AI brings is behind us. A big part of the change curve is behind us. So a lot of the fear we have very well survived also because we've

[00:04:52] learned in the process that any AI application, frankly, it's just another tool in our tool set. It takes a human to make sure that the tool set is appropriately deployed. So what you learn when you essentially are in a machine-first process, part of the machine-first

[00:05:08] process is you learn very quickly that the machine has its limitations, if you want to put it that way. It is still a very almost statistical process as in the quality of the majority of

[00:05:19] the data will influence the output. And it ultimately falls to the human to optimize how that tool is deployed and when it is deployed in the best possible way. So I think what we've learned is actually the machine will never replace us, but it does

[00:05:33] actually change what we do. And it changes it quite substantially. It doesn't mean that the world around us changing us didn't affect us. We suddenly had clients calling our linguists, many of our client reviewers, particularly our indirect contacts with our in-house linguists,

[00:05:48] we literally had clients calling our linguists asking them, so what are you planning to do with the rest of your life given you will be fired soon? Which of course is a fantastic moment.

[00:05:57] And so of course it did affect the teams and we did what we always do. We essentially started playing with that new thing that had come along with large language models to try and figure out

[00:06:07] how does it work? How could we use that new tool? And what does it change in the overall way of working? And we've come back pretty much with a very consistent perception at this point in time,

[00:06:21] based on using the combination of neural machine transition and large language models every day. It is essentially the next evolution of the tool that we already know. And we now have to think

[00:06:31] about how we best deploy it and we found ways of best deploying it, but it's also increasingly obvious that again, the human will not be replaced. The way the human interacts with the machine changes. And it's a very mutually beneficial relationship, if I'm perfectly

[00:06:47] honest, as in the human quite literally brings the human touch, understands context, makes content engaging, makes it relevant, makes it valid and accurate while the machine just predominantly helps accelerate. And thanks to that large language model,

[00:07:04] even helps accelerate very, very complex and convoluted processes that we couldn't really touch before. So it's more for us, it's more season two of a series that we already knew. It really does. And I recently wrote an article about collective intelligence and how

[00:07:22] the magic really happens when humans and machines collaborate, combining their unique strengths to achieve that neither could do completely on their own. And one of the reasons I invite you

[00:07:33] on the podcast today is I know a subject very close to your heart is that AI content just is great, but it's just not the same without that additional human input. So can you expand on that

[00:07:43] and also how human creativity and the oversight actually enhance AI generated output? Because you need both, don't you? Absolutely. I mean, we call that concept genuine intelligence actually, because it's a lot more about understanding the binary aspects of language and adding to

[00:08:03] that understanding the binary aspects of language, all of the more complicated context-driven, etc., human, very, very human aspects of language and culture. So combining the two is almost intelligence on steroids. So we call it genuine intelligence, but we're thinking exactly the same

[00:08:19] way. So in a nutshell, what we've learned over the last couple of years, and particularly in the last year, is even with large language models, we still have the same fundamental problem. And what

[00:08:30] I mean by that is it still is to some degree a fundamentally statistical process. So the amount and quality of data that you input into a machine learning process will very much drive the quality

[00:08:43] of the output of machine learning processes. And it is actually up to the human to make sure that their output is not just driven by the majority, which is an even bigger problem for large language

[00:08:56] models, specifically because large language models absolutely are being driven by so much larger data sets. So it's essentially in very, very simplistic terms they've learned from the internet and the internet is so full of bias. The internet is frankly dominated by,

[00:09:12] I would say, eight to 10 languages max. Most of our customers buy 80 plus target languages, so eight zero plus target languages now, which essentially means that automatically gives you massive weaknesses in some of the outputs that you see. If a market is comparatively small,

[00:09:30] the amount of hallucinations that you'll have to deal with as you talk about a market or an area subject meta expertise really denies the efficiencies that you can get if you're smart

[00:09:41] about it from AI and large language models. But my favorite example of that one is we have one of our customers generates target language content across a multitude of languages for travel destinations. And I loved what the engines had created about Slovenia. Slovenia is, as we all

[00:10:01] know, a comparatively small market. The level of improvisation essentially was mind-boggling. It invented parks in Ljubljana. It's essentially claimed that an existing Slovenian punk band had won the Eurovision Song Contest a year ago. They haven't won in the first place,

[00:10:20] aka Slovenia as a nation hasn't won, and that punk band definitely hasn't participated. Now, you need to know so much more about what is really going on in the ground on specific languages or in specific markets than to even notice that this is wrong because the text

[00:10:40] read beautifully, really, really eye-wateringly beautifully. So this is what the human brings, understanding the limitations of the tool and working with it. Many of the AI capabilities they really can't do humor because humor is so context and timing dependent. And

[00:10:57] AI, again, is always a little bit behind times because you need to reach a supercritical mass of data to be able to represent a specific concept adequately. So humor is so timing dependent. That's what particularly if you think about creative text localization or even copywriting,

[00:11:17] that's what a human will always have to contribute, humor, creativity, because creativity lives in the spur of the moment and lives in the context of the content consumer. So you need to understand, for example, in what specific environment, through what channel, etc.,

[00:11:37] a particular piece of content will be read. And that's when it comes to creative content. But it's also, for example, if you think about content for medical devices or pharmaceutical content, you cannot assume that patients have access to the same means and information channels

[00:11:54] wherever they digest the information that they have to digest. A patient in a hospital in Afghanistan probably will not use an iPad to, for example, input information around a clinical trial. So knowing things like that is ultimately what a human can always contribute,

[00:12:11] depth in cultural expertise, expertise about the local context in which content is consumed, and also just, again, creativity and a sense of humor. 100% with you on that. And on this Daily Tech Podcast, I also always invite my guests to maybe

[00:12:27] bust a few myths that they see out there. Maybe they frustrate you. So I've got to ask you, what are some of the common misconceptions about AI's capabilities, particularly in fields like journalism, copywriting, and digital art? Because we're seeing this on our news feeds a lot at the

[00:12:42] moment. But any myths out there that surround this that you see a lot? I think the whole AI can replace a copywriter, can replace creativity topic is that I've mentioned

[00:12:53] before is a big one. The next one is that AI can essentially give you the same depth of subject matter expertise that, for example, journalists can give you. Oh, I can tell you from the

[00:13:04] large language model generated content that my team reviews, the amount of research that you need to do to even validate some of the stuff that gets flushed up is impressive and is now a new

[00:13:17] value added service. So don't underestimate the amount of work that needs to go into making content bias-free, generating ethical content and generating actually valid content, which is really critical for most of our customers use cases. The other one that eternally amuses me

[00:13:35] is that AI comes for free. It doesn't. It's a technology that costs money to host, to maintain, to keep alive, to keep relevant, et cetera, et cetera. I wish that the rest of humanity got there.

[00:13:49] It does not come for free. And that is me only speaking of the direct cost of AI. There's an indirect cost of it as well. Going back to the, you need to keep bias in check. You need to

[00:13:59] validate the validity of content, et cetera, to make sure that it does not essentially take you to a place that you don't want to be in as an organization. Yeah, that we could almost dedicate an entire episode that we've got the energy that AI consumes

[00:14:16] as well, especially in an era where we're talking about sustainability, et cetera. In your own role, how do you address concerns that AI might replace jobs by illustrating its need for human input and deliver that quality of work? Because I would imagine there are a few

[00:14:33] worried faces around. And how do you address those concerns? I think the biggest, to me, the easiest way to deal with fear, fear of the future is to just play with the future, figure it out for myself, and then ultimately build a path moving forward,

[00:14:49] which is what we did. We just had a comparatively large team of linguists work together with our AI development teams and just figure out how it works, how it's best deployed, what brings the best value for money for our customers. Value truly is also about making sure

[00:15:05] that we have an ethical approach to AI in the first place. We have a huge responsibility as an organization of our size towards our internal linguist teams, as well as our external supply chain. We need to be there to help guide them through this situation and make

[00:15:22] them understand that this is not an opportunity, not a threat. So that's what we did. We essentially played with it and positively identified the potential, built processes and guidelines on how to go about it constructively, and also have been quite vocal about the potential,

[00:15:38] the amount of acceleration, even acceleration of comparatively complex processes that we see through the use of AI and through an AI-first process is really a major benefit given that we see an explosion of source content, so English content. We also see an explosion of needs to

[00:16:00] cover more than five, six, seven, eight, nine, ten languages. I said before, 80 plus languages. The trend now goes towards more than 100 languages for most of our customers. You cannot cover that without the use of AI, and it's particularly true if you think about, for example, African languages,

[00:16:16] Southeast Asian languages. Not a lot of content is out there to help you generate AI for those languages, but at the same time, you need it to be able to cover all content that our customers

[00:16:27] expect us to localize for them. So it is something that actually facilitates bringing content to consumers closer to whatever they're trying to consume. So the whole unlocking global understanding thing that we keep going on about is something that we really, really care about, and that is

[00:16:49] realistically only possible if we embrace AI and the capabilities to accelerate and to make processes more robust if adequately combined with humans that have deep, deep, deep language and expertise as well as cultural expertise. And you mentioned a few moments ago the problem with

[00:17:08] AI hallucinations, especially with the example of the punk band in Slovenia, but are there any other closer examples to Homer about hallucinations and bias, how they can, the impact that AI generated content can have, and also how human intervention ultimately mitigates these issues?

[00:17:27] My favorite example that winds me up no end, I tell you, is AI is the German term, the English term nurse in most languages actually needs you to be specific once you translate it,

[00:17:39] and whether it's a male or a female nurse. Across the globe, AI will assume that a nurse is always female and that a doctor is always male. As I said, that winds me up no end. However, that bias goes

[00:17:53] so much deeper than that. One of the observations that we've made, for example, is the person that gives advice is always male. The person that receives advice and the person that is grateful is always female. Male people apparently are always ungrateful. So think of that.

[00:18:11] So it is things like that that are, again, as a woman in business, for me personally, terribly infuriating because it just perpetuates stereotypes that we should have come over given the last 100 years. A lot of those biases just go so much deeper than that.

[00:18:30] Do you really always want to perpetuate notions that may go completely against what your brand stands for? So that's what you actually need humans for. You need to make a human understand

[00:18:42] what you as a brand of your organization want to stand for if you want to stand for inclusivity. These are the kinds of things that you actually need to think about as you devise how you want to build your localization process.

[00:18:55] And you are someone that is a tech optimist and there is so much doom and gloom around the future of work and the impacts of technology. So I'd love to try and restore the balance in the universe by

[00:19:06] sharing a more positive future. So in what ways do you think AI and humans together can collaborate to create more of a symbiotic relationship that benefits both parties? And what do you think that

[00:19:19] means for the future of work? I think what it means for the future of work is that you actually start almost increasing the value that a human brings to the equation because a lot of the

[00:19:33] localization processes require a lot of thinking and structuring. And sometimes not all of that structuring can happen and can happen fast enough. Given, as I said, how much more content there is

[00:19:46] out there for localization, I have been that the direction of travel is up when it comes to word volume, I can tell you. Given how much more volume is out there to localize and given how little

[00:19:58] time we increasingly have, like when I started in localization as a project manager in 2006, the average turnaround time project for me was over 30 working days. It's now 24 hours and it's increasingly going to six to 12. You cannot do that without the use of technology. So technology

[00:20:17] helps us accelerate even very complicated processes now. And all of the tedious stuff that linguists often used to have to do, for example, copy and paste content from one platform to the next, sealing bizarre amounts of Excel spreadsheets to deal with terminology and whatnot.

[00:20:36] AI can do that for you now. Even researching now becomes materially easier because you can use AI. So what the human now focuses on is where they add most value, be truly creative, research, learn, validate. Like for me, as you may have noticed, fundamentally curious

[00:20:56] a human being that is incredibly appealing from my perspective for the future of work. It also means that we need to be very responsible of how we deploy it. And then just because it

[00:21:08] accelerates you doesn't mean that there is the sky is the limit in how fast you can get. So be clear about what are the areas that you should and could responsibly accelerate and what are

[00:21:20] the areas where you better make time because humans are not machines. Humans need time to be creative. Humans need time for proper research and validation. So that is where I see the symbiosis.

[00:21:33] You essentially use the machine for all of this stuff that are a waste of what the human has to offer and help them to accelerate and streamline. Use that to accelerate and streamline and use

[00:21:44] the humans to spend their time where they add most value, which is intelligence, creativity, research, etc. I completely agree with you. And as you said, humans are not machines. But ironically, many people have been almost treated like machines stuck in repetitive and mundane roles where

[00:22:03] they're performing robotic tasks every single day, whether it be in a spreadsheet or on a factory floor. And I think in some ways we've got to get back to being human, being curious,

[00:22:13] being creative. And that's the skills that we need. So we'll have business leaders listening all around the world. How do you think these businesses can better prepare for their or prepare their future

[00:22:24] workforce to leverage AI? And also, what kind of skills do you see are essential for workers to effectively integrate AI into their roles? My gupteal is embracing AI. A lot of people are currently very risk-averse when it comes to AI, which I totally get. Like you shouldn't embrace

[00:22:42] something just because it's new and shiny. But you should do your research. The minimum you need to do is do your research, and do your research with the right mindset, which is, could I make

[00:22:51] this work for me? And not, is this scary? That's a wrong default setting. So have a positive, embracing curiosity as you research and be open to learning new things and be open to use

[00:23:05] technology for your own benefits. Give your teams the freedom to experiment and come up with new ideas and come up with ways of funding that willingness to play in an experiment. What I

[00:23:18] mean by that is, look at what they're doing and start thinking about what exactly brings business value to you, your customers, and then try to hone all of the playing and experimentation down to

[00:23:31] almost like two, three, four, five core topics that you want to research and then just go after it. For me, this is a lot more mindset than anything else. And that's what I'm currently

[00:23:43] discussing with my teams as well. Play with it, play with it, come up with reasons why this makes sense for us and be open to use your brain again in the way it was originally intended.

[00:23:56] So focus on time for creativity, focus on time for research, etc. That is a fundamental mindset shift and it's a change journey that based on my experience takes a couple of years, but it's totally worth sparking that initial curiosity now and taking it to the right place.

[00:24:17] Just to further ease fears, if we were to look in the context of translation at RWS, what would you say are the uniquely human elements that make the role more than just a find and replace task? And how AI can actually enhance elements without actually replacing them?

[00:24:35] I think it'd be great to bring to life the positive impact it's having at RWS. I think the fundamental benefit that we see again is acceleration and making processes that are innately incredibly complicated look very simple. That's the fundamental benefit that

[00:24:54] AI as a tool brings to us because essentially the complexity moves into the background. The process complexity moves into a background and gets simplified by essentially robust AI-based processes. What the human contributes to this is the deep, deep language and cultural expertise,

[00:25:14] the awareness of how content gets consumed in the target cultures. Not sure you're aware, but most of our linguists actually sit in the country of their native language. Being aware how the content will be consumed, in which channel, by what kind of demographics,

[00:25:32] in what cultural context brings an enormous benefit to most of our customers. To give you an example of what I mean by that, we often tend to have retail customers, for example, that try to position a specific product in exactly the same way in every market. That is

[00:25:55] not a great idea. It's an example that I use a lot and that I've used a couple of years ago as well. In the context of the pandemic, we had a retailer notice that a lot of people are buying sweatpants

[00:26:09] now. They assumed that the use cases for buying sweatpants, and that's how they built their campaigns, was for example, things like home improvement projects, doing yoga, playing with your family, sitting and doing Netflix and chill. The feedback that many of

[00:26:29] our linguists, for example, in Germany and Japan gave was, no, we don't do home improvement projects in sweatpants. In Germany, because Germans are very health and safety oriented, they would essentially use any power tool wearing sweatpants, first reaction.

[00:26:47] The Japanese said, we don't have the space to do home improvement projects in most instances. Totally irrelevant use case for us. That is the kind of value that linguists living in the country actually bring to you. It's also deep subject matter expertise. We have linguists in our

[00:27:06] teams that have such deep subject matter expertise when it comes to pharmaceutical contexts. For example, AI would never be able to bring that level of subject matter expertise to the table. They just know their stuff and can absolutely understand hallucinations,

[00:27:24] deal with them appropriately and guarantee the validity of the output that has been generated. That's how the world changes from a translation perspective specifically. Well, thank you so much for joining me today and sharing your insights. Before I let you go,

[00:27:39] I'm going to ask you to leave everyone listening with one final gift. Yes, we are living in a world where people can listen to audio books at 3x speed or digest in 15 minutes on something

[00:27:49] like Blinkist or even go to AI for an even quicker summary. Despite all that, I am a little old school. We've got an Amazon wishlist of books that people like to read in a more traditional way

[00:28:00] that they recommend. Is there a book that you'd like to add to our wishlist and why? Yes, I can. I am a very avid reader, I can tell you, and I'm not a fan of artificial summarization at all. I normally read mostly fiction, sometimes nonfiction. I've just

[00:28:19] discovered David Graham for myself. I've read a couple of books of him and Killers of the Flower not the movie, the book has really positively blown me away. A, because I've learned a lot, I am intently curious, but German nonfiction does not necessarily spark imagination in most instances.

[00:28:42] He really does that. It was one of the few nonfiction books that I've ever read where I got angry and all emotional and whatnot, and it was incredibly informative. I've learned a lot. Definitely a book that I will recommend to everybody who doesn't run away fast enough.

[00:28:59] Love it. Well, I'll get that added straight to our Amazon wishlist. And for anyone listening just wants to find out more information about RWS, connect with yourself, ask your team a question, for anything RWS, is there anywhere in particular you'd like to point,

[00:29:13] everyone listening? We do have, well, rws.com, of course, everybody can follow us on social media. We do have a podcast as well called Globally Speaking, so plenty of opportunities to follow us. Oh, fantastic. I will add links to that.

[00:29:26] Your podcast, tell me a little about that. What's that about? It's essentially multiple conversations around all of what localization means to us and to our customers. We have a couple of really cool podcasts. We had one lately around essentially

[00:29:40] the combination of humans and AI. You may want to listen to that one with one of our incredibly talented colleagues called Marina Pancina. Definitely listen to that one. Excellent. Well, as I said, I'll send a link for that so people can listen nicely.

[00:29:55] And I think listening to you today, your passion for technology, your curiosity, and everything we talked about really shines through. And the big takeaways for me is AI content just isn't the same without human input. Ultimately, AI needs humans to be effective and

[00:30:10] humans need AI to be efficient in this digital-first content-rich world. Pure gold from my side. But thanks for talking to me today and sharing your insights. My pleasure. Thank you for having me.

[00:30:21] After my conversation with Maria, it got me thinking. It really is clear that AI is not here to replace us but augment our abilities, allowing us essentially to focus on things like creativity, context, cultural nuance, areas where humans excel. And the RWS Group's journey from

[00:30:42] initial AI adoption and its current genuine intelligence approach, for me, highlights the importance of human insight and the value of AI as a tool for efficiency. So as businesses continue to navigate the evolving landscape of AI, the key takeaway is the need for a balanced approach

[00:31:00] that actually combines the strength of both humans and machines. But what are your thoughts on the role of AI in your industry? There were so many positives in my conversation with Maria today. I'd love to hear how it's impacting you, your industry, and what you're seeing. Both good,

[00:31:18] bad, and indifferent. There's room for all arguments on here. So share your insights. Join the conversation by emailing me techblogwriteroutlook.com, LinkedIn, x, Instagram, at neilchughes. If you're not following me yet, why not? And if you're going to,

[00:31:34] don't just hit the follow button. Send me a quick DM. Tell me you listen to the podcast. For me, it's all about connecting meaningfully with each and every one of you. But that's it for today. So until next time, stay curious, keep exploring the possibilities

[00:31:49] of technology. And most importantly, I'll meet you same time, same place tomorrow with another guest. See you then.