Focus on Sustainability in Addressing Gen AI Ethics and Risks: Prakash Arunkundrum, Logitech

Great Dr. Gleb, very nice to see you again. It's been a real pleasure to be talking to you. So I'm Prakash Arunkundrum, president of Logitech for Business. Logitech for Business essentially is a division of Logitech that makes products that gets it to customers, enterprises large and small, on everything from video cameras to Logitech mice and keyboards and headsets and webcams and everything in between if you happen to be somewhere on the go with our other accessories and peripherals. So all of those things as part of our Logitech for Business, and I've been at Logitech now for 10 years in various different roles. Until recently I was the chief operating officer for Logitech. I also ran some of our operations on product development, manufacturing, supply chain, and also led our sustainability efforts for the company and actually driving carbon reduction across our entire value chain. So that's a little bit about me, but super excited to be here. Dr. Gleb, so today we're meeting to talk about generative AI adoption.

Yeah, I mean this is by and far should I say the most transformative time that we as a species are going through, and therefore there's been a healthy level of questions, curiosity, and all kinds of things around innovation that has been happening within Logitech. Certainly, as for some of our products and product teams, like our video conferencing equipment already deploys some level of machine learning, and now with advanced models they're doing more and more with it. So the product teams have been super excited. They were excited to be able to get on GitHub as soon as it was available and do more with software and how we actually develop software. Our marketers internally started exploring how you can make better icons, how you can make video better, how do you create some questions on how do you handle certain things when you have avatars presenting for you. So we made some policies around product images being the key focus and how you get pre-product images and not dwell into some of the areas around people and personalities that may actually be included in videos. Our employees wanted to use latest AI tools. I personally use at least three or four different tools every day inside of our network. So we created that so we can protect our IP but also use the best of what AI provides. And I'll say on average our employees are super excited. As a tech company, this is the most transformative thing that's happening and we're super excited internally. I'll pass there. I'll talk about external maybe next in a minute, but internally super excited, we're seeing broad adoption in operations, in marketing, software engineering, product teams thinking about AI in the context of products. Yeah, it's a fun time to be in the midst of this.

Yeah. I'll say first of all, maybe I'll start with leadership and then the people in the team. Leadership had very little resistance to AI, which stems from our history as a tech company. We understand that we have to innovate to keep up with what's happening, otherwise you risk being left behind. So there was very little by way of resistance from the leadership team. And as I said earlier, the biggest topic top of mind for us was how do you make sure you still protect privacy content that's personal to Logitech, that we've made innovations possible, how do we make sure we protect that. So there was a little bit of not caution but real thinking around how do you utilize some of these AI models and what data do you actually allow the AI models to see that is Logitech proprietary. So that was one place from a leadership perspective that we spent some time on. We've actually been using AI in our video conferencing well before AI was a buzzword. We had a thing called RightSight and RightSound that was announced way before 2022 when ChatGPT became big, and it has been part of our products. It's been one of those things that we've been doing. We have had a machine learning team that looks at computer vision for at least eight years now. So it's been one of the things that we've always kind of done in the context of machine learning, but as these models got more involved and more capable, it certainly opened up a new possibility from a creativity perspective. We didn't get, to be honest, a lot of job loss type discussions at all. The bigger questions were how do you prevent bias in the models, how do you make sure you have the right data that you actually can make these models understand. As you know, depending on the input data you're going to run it through a transformer, a large language model transformer, and it's going to generate some output. So the input data is super critical. And if I zoom out for a minute, Logitech actually plays in that world of input, right? If you really think about where we sit in the ecosystem, we generate audio, video, and input from mice and keyboards. So we are kind of very aware of our role in the ecosystem, and that brought up even more awareness around having this Swiss neutrality which is what has always been known for, and really understanding the privacy aspects and the bias aspects of AI. So that was probably the biggest area of focus. And we've not really received too much internal resistance. There's been more wanting that we've been able to like everyone's wanted to try the new latest thing that comes out, and we've been very deliberate about how we allow that to happen. So we set up an AI governance board. They actually approve certain models from being used internally, and we have policies on responsible AI that we put in place. We also signed up for the European responsible AI act just to bring that top of mind in terms of our employees and product teams as we think about it.

Yeah. So I'll start with the most basic of basic examples, which is I'm looking at you at a video camera and the video camera has to recognize who I am and therefore represent me in the best of myself. And to do that you need the models to actually understand things like lighting conditions, and also equally things like tone and color and complexion and things of that sort. And the same thing applies on the audio side. So we spent considerable amount of time making sure that the data models that we actually use to train our RightSight and RightSound are based on as wide a set of parameters as possible, as diverse a group of people as you can. Audio, as you can imagine, there's a huge difference in tones, accents, pitches if you're male or female. So really thinking about it from that perspective and actually making sure that that is top of mind. So that's two things: audio and video. And then actually having access to the data, who gets access to that, make sure that that is pretty clearly marketed and the legal team. We have a chief information security officer who is in charge of our privacy and security policy. So she's really top of mind on making sure that we subject this as an additional gate in our security and privacy processes. So that's kind of how we thought about it.

Yeah, actually as a matter of fact recently in our annual investor day a couple of weeks ago, we actually shared our AI strategy as part of our mission and it goes back to this central theme. People want more precision, more intent, and more smarts. And if you think about what we do as a company, we make simple, smart, and sustainable products. I mean that has been our core approach. If I had to distill it to three things, that's basically what we do. And the domains where we apply them is audio, video, and input. So if you really think about us in the ecosystem, Logitech and the ecosystem, it's our job to make sure that the data that you get to these cloud models — we are on a video call and the cloud model that's actually running this is getting data from us, is taking out noise, doing all of these things — but it needs that last millimeter fidelity of audio, video, and input data. So we really defined that as our core sort of place in the ecosystem. We're not going to become a company that is only doing language models for a living. That is not Logitech. Logitech makes products that customers use for work and play. Our mission statement is to extend human potential in work and play, which means to make the human even more capable. And that's the second side of this AI topic that we really saw, which is how do you make us more relevant in the context of the human. So making the human more important, making audio, video, and inputs more important is really where we went back and reemphasized our core values.

Yeah. So this is like the most important thing, which is the trust and credibility with customers, especially in using AI-driven features and functions. So I'll start with a few. I just mentioned AI principles. We signed up for the responsible AI act, which was one of the things that the European Union put in place. That's to hold us transparent about what we're going to do and also be accountable, right? That's the first one. The second one as it relates to things like private data, even before AI was a thing, I think we were as a company really mindful about what private information do we actually store on users' behalf. And I'll say that is almost like a separate thing. If you had AI on one end, you had privacy on the other end, kind of like yin and yang of the world. You need more data to make the models better and you need data to be more private. So we don't store any personal information as a policy, and this is one of the reasons it was super easy for us to sign up for the responsible act. We didn't have to think too much about it. So that's the first one: responsible AI act. The second one is I think working with the ecosystem partners: Microsoft, Zoom, Google, a lot of the big tech players that we connect to, really making sure that we are connecting with them with the view of what makes our part of the ecosystem clear, like what do they do, what do we do, how are they protecting this information and what role do we play. So that's the second one. Enterprise customers really think of you know if you are using a mail server from one of the Microsoft or Google or someone else and you have calendar information, these reside with the cloud providers, we don't have access to it, but we are making sure that the devices can provide that additional functionality in service of Microsoft, Zoom, Google, etc. So that's the second part. Yeah.

Yeah, I mean this is a huge topic. As a company, we've really focused on sustainable practices for a really long time. We know that we exist in the context of an ecosystem and we exist in the context — even if you think about it from a sustainability perspective — in the context of suppliers and customers and partners. So the first thing I'll say is that has been top of mind for us is to clarify our own position, right? Where are we, what is important for us, what does bias look like, and what do we stand for? And that's best shown in the kind of people we bring. You'll see us talk about the kind of environment and equality rules that we have as a core value. If you read our sustainability impact report, that probably comes through, which is that we really want to make sure that we are paying attention to that. So it starts from that place, which is sustainable practices. If you have sustainable practices, you get to start to talk about fairness and equity, right? So that's the first framing of this question. The second framing of this question is how do you make sure you have the right safeguards in place to understand who and what data is actually used to train these models. So we at least have, as we say, a creator and a reviewer, two roles: somebody who creates a model, somebody who creates the data, and a reviewer who's not the same person actually reviewing what the data is and who is being used. So that's a very — it's one of those product development philosophies that as a hardware company we've had because when you make a hardware product, you have to make sure you have quality criteria. So it's very natural to Logitech that you have hardware and you have quality, you need to make that really important. So that's really the two things that we've at the core: sustainable practices and making sure we are very mindful of who gets access to the data, and we're asking questions of is this complete and comprehensive. And there's a lot of cases that we'll have to still do as a society to handle reducing bias. I can assure you that there is more to do here. We are by no means perfect. I would not tell you we're perfect yet. We have work to do, but at the same time, understanding that we have work to do is perhaps the most important thing to actually acknowledge up front, and we've tried to do that quite a bit. So yeah.

Yeah. Well, maybe it's useful to kind of talk about just what are Logitech's products and how do they intersect AI, just for your users to kind of have that in their minds. So when you think about video bars, I'll start with that and then I'll give you some other examples. With our video bars, we have a product called Logitech Sight, which is essentially a bar in the front of the room connected to a center-of-the-room camera, and we can now connect two center-of-the-room cameras. So there are six cameras and six plus microphones — actually 12 microphones — listening and viewing and bringing, as I call it, a little bit of Silicon Valley in terms of who should you frame, when should you frame, how do you use machine learning, and a little bit of Hollywood: if I'm looking, I'm a movie director, I frame you, and then I release the frame, I go and frame this other person. And we really built this with the idea that we want equitable meeting so that the remote participant feels like they are already there in the meeting room even though they may be miles away. So that's one example of where AI at the edge meets cloud in a conference room and really being able to deliver this. I'm happy to report we got the Times invention of the year award last year. So that's really congratulations. That's awesome. Thank you. That was really good. So that's one example. The second one which is a very practical one as well is we have a two-way AI headset. What that means is our headset has algorithms in it that will not only cancel noise on my side but it'll also cancel noise on your side by actually filtering out only the stream — the wavelength of sound that actually emanates from human voices. So it'll cut out all other sounds that are not yours. So that's another example of an AI-based technology that's as harmless as harmless can be because we are really not recognizing you but recognizing frequency ranges of people's voices and then being able to apply smarts at the edge to compute and do different things. The last example I'll give you is mice and keyboards. We have the ability in mice and keyboards now to really allow you to quickly launch your favorite AI tool in context. You can repurpose your documents you're working on, really quickly do that. And recently we came out with a new thing with our Streamlabs capability on the gaming side which is using agentic AI. So when you're playing a game and doing one of those games where — if you game a lot, you want a sidekick to help you play better. So we have an agent now that'll be your sidekick. It'll tell you, hey, somebody's coming from that right side, you may want to turn to the right and look at it. And if you played a first-person shooter game or something like that, it'll tell you how accurate your plan was, like what can you do to improve. Almost like a coach that's living with you through the game. So very cool exciting things coming out on the product side. And that's the role that we see ourselves: really augmenting human workflows which are human centric. And as I say, our goal is to make technology just disappear and fade away into the background so that you don't know it exists. It just works and it's like magic. So that's the goal.