EP 340 – Are We Ready to Chat with AI Avatars Instead of Humans? – Igor Hofman – D-ID

Read the best-effort transcript below (This technology is still not as good as they say it is…):

Michael Waitze 0:13
Hi. This is Michael waitze, and welcome back to the Asia Tech Podcast. We are joined today by Igor Hofman, a VP of Sales at D-ID. Igor, thank you so much for coming to the show. Before we get into the main stuff, let’s get a little bit of your background for some context.

Igor Hofman 0:28
Yeah, great. Hi Michael, thank you for having me. It’s a pleasure. Yeah. So I’m currently head of sales at the ID, as you mentioned, I’m actually in sales for the last 10 years more or less, did some interesting, diverse things. Started from, actually the military space, and was selling unmanned aircrafts and airplanes. Then, then turned into the video space, working for a company called Kaltura. We were doing, they were doing video platforms and offering solutions for cloud TV services, and join the ad right now.

Michael Waitze 1:05
So that’s it. Let’s let’s just jump right into this. Avatars, right in your experience, in your research, in your day to day sales activities. Do humans really want to interact with avatars? What’s the sense you get?

Igor Hofman 1:18
Yeah, very good question, and a big one. And I would say that there are multiple answers to that. And so if you ask it this way, I don’t think there’s a straight answer, fair enough, but the answer is that people want to interact with people, right? This is, this is, this is 100% so people feel comfortable interacting with people, speaking with people, seeing people, their brain is better, better suited for interactions with people. Avatars are representations of people. So in that sense, if you bring it this way, when someone is speaking to an avatar, and the avatar resembles enough to a person, it makes the effect. So the short answer is yes, the longer answer is it depends on the quality of the Avatar, on the experience that the customer, the person, has with the avatar, and then the experience is definitely better when it resembles as much as possible interaction with the person.

Michael Waitze 2:17
How fast is this stuff developing? Do you know what I mean? Because, yeah, two years ago, three years ago, four years ago. If we looked at what an avatar looks like online, it was kind of a little mechanical. Even the voices were a little bit mechanical. I feel like it’s moving really quickly now. And I’m just curious what the sense is on your side as well. Yeah,

Igor Hofman 2:33
so it’s moving very, very quickly. We saw, we saw the jump something like two years ago with, with everything happened with with large language models, GPT boosted two years ago, more or less, and ever since, it’s really, really growing. Also, there are multiple types of avatars. Tell me we specifically, at our company, we’re focusing on hyper realistic avatars, so we’re trying to make the avatars as close as possible to the human right. There are also other types of avatars. There are cartoon like avatars, there are generated avatars, and so on, so many, many different types, and they’re catering for different needs. And because sometimes you don’t want to speak with a human, maybe you want to speak with a cartoon. Maybe you want to interact with someone from the past that you know, that doesn’t live today. So so it’s not like you You fool yourself you’re speaking with right on that kind example. But yes to your question, it’s moving very, very fast. And we actually started a from photo based avatars, avatars that are that are being made from one still photo. Take one photo, when you animate the photo, you bring for life. This is a where we started, like something three years ago, more or less, when we entered into this generative AI space. Ever, ever, since things changed. So we also have, not only us, but everyone, have photo based avatars, video based avatars, short video based avatars, and there are multiple types. But everything’s progressing very, very quickly.

Michael Waitze 3:58
It’s just kind of amazing. Is this one of those things, you think that’s just going to be so pervasive that over time, it’s going to be hard to tell it’s going to be indistinguishable between like, whether I’m talking to a real person or whether I’m talking to an avatar. Does that make sense in certain situations?

Igor Hofman 4:14
Yes, I think, I think in the longer term, this is where it will end up, I think, for now, and we also see it no one tries. And I think also our recommendation is not to fool people, right? So we want, we want to bring the effect of having the human interaction while still keeping people honest and making them aware that there are still speaking with AI generated characters. I think it’s very important, because there are still some discrepancies. It’s not 100% yet, and if people feel that they’re being fooled, then they will be less ready to engage. So we’re actually maybe jumping ahead here. No but. We in the company, we’re very big believers of ethics and responsible use of AI. We’re kind of seeing ourselves as a flag holders of this, of this movement, of using AI in the right way. So we encourage our customers not to fool anyone, although maybe it’s possible in some in some places, but let people know they’re speaking with generative AI based avatars, and enjoy the conversation and enjoy the effects of it, but not kind of always be alert to Who am I speaking with? Speaking with a person. It’s an avatar. We’re trying to move away from whatever deepfakes And all those things that happened in the past, which were very negative in this space?

Michael Waitze 5:41
Yeah, I want to get into the ethical issues a little bit later, as you suggested, but I want to ask you kind of a funny question, too. You’ve been in sales, you said, for 10 years, and there is kind of a sales function that an avatar can’t like. It can W, triple, U, quadruple you, right? And if your customers have questions, they’re, you know, they’re naturally going to ask the salesperson a question. Is there a way for you to use this as well? Do you know what I mean? Take a photo of you. Maybe take a video of you, put your voice in there, and just have an FAQ. Let’s say it’s like standard questions that your customers would ask you. And they call they know it’s not you right, because you said this is the ethical issue. But the simple stuff that they may need to know. Are you already doing that kind of stuff?

Igor Hofman 6:23
Yes, 100% 100% I’m doing it because I’m working in this in this space. To do it, as you say, you have to eat your own dog food. We are doing it, but I would just take it and explain so one of the main markets that we are, we are focusing on right is or being verticals, because we’re more vertical oriented than industry specific is sales and marketing? Tell me actually, it starts from marketing and then it goes to sales, because the easiest way to use AI generated content is in marketing. Marketing content, advertising, targeted personalization and so on. So we’re definitely doing it, and not only us, but many, many companies in the world today, they’re creating marketing content using AI, using avatars, and then they can really, really make the content more accurate, more personalized. So imagine you can create 1000 different videos, and every video is different just by how you approach your target audience, or who is, who is the presenter that you use for specific audience. So maybe in certain way places, a woman be better than a man. In certain places, someone who is more older is better than someone who is more younger in terms of how the content is is absorbed and how people are engaging with it. So this is on the one hand, on the other hand when it comes to actually, like one on one interactions. So we’re not yet replacing completely salespeople, right? Although maybe this is kind of kind of direction that we can we can go, but we can replace the initial step. So the qualification step, for example, in a sales call, definitely can be done using an avatar. The Avatar can can do the Q A session, they can qualify the prospect, and then they can provide the initial information and pass it on for the more complex parts to the actual human. So we’re definitely doing that right right now, both from a content perspective and from an interactive perspective.

Michael Waitze 8:16
I was actually really impressed with the website. Was rooting around the website. And I liked how you had the, you know, how to have an agent. You had, you had an avatar agent actually doing the introduction. It was just kind of cool. And all the videos, at least, most of them that I looked at, had like a different person doing it with a different voice, and stuff like that. And it really makes it. I was just sitting there stunned, actually, but also tell you this, I didn’t know this, but my team is already using this technology from di D. I did not know that. I told him that I was having you on the show. And they were like, really, we use that thing every day. Here’s what they do. They write an introduction that a human does the same like one of my teammates does an introduction for one of the products that we have. He does the intro and then he has his avatar, Max go through a step by step explanation of it. And the way they use it is they’ve built max on their own. They have somebody else’s read a text, but then have that text use one of your voices, and then at the end, the real human jumps back on and says, that’s our product. Blah, blah, blah. Please click here if you want to use this thing. So it’s multiple ways that this thing can be used. Do you have other like, specific examples you could share as well?

Igor Hofman 9:27
Yes. First, I’m not surprised, although we are flattered, but I’m not surprised, because when we look at our customer base, we can’t have something like 40,000 people using our platform in multiple ways, but we actually have millions of people trying it on on a day, on a daily basis. Yeah, now only, only, only a very small portion of those customers are actually growing to a large scale use. And those are the bigger companies, the bigger enterprises of the world. But also, like every day, we discover that some huge enterprise, you know, some Coca Cola target, was. Walmart, whatever, are using us for a commercial they did or for a corporate training piece. They don’t even talk to us. Maybe, maybe I didn’t say it, but we’re following a plg strategy, product led growth. Okay? What it means is that we’re actually allowing people to use our product on a self serve basis, just going to our website and starting to use without talking to anyone, without having sold anything, and only when they reach a certain point, there’s a certain scale of operation, only then they’re being approached by a salesperson to try and see if they can, they can, they can go into an enterprise plan. It’s a sales dynamic. It’s a marketing and sales dynamic which we’re operating. What this allows is a very high exposure for to people of our product. So this is why I’m not surprised and in terms of how people are using it. So I would just take it from a customer journey perspective, the way company is looking at it, as they see, they see the different touch points with their audience and their customers along the way, from the beginning when no one knows about their brand, all the way to the actual, the actual closing of a deal, or even even selling a recurring deal and upsetting and so on. Throughout this, this, this journey, there are multiple touch points where you can use the avatars to interact with the person. So we start, we start with like, the initial advertisement, right? So someone wants to create an advertisement, which is creating a content, they’re creating an avatar, incorporating the avatar into the content, and just sending it out there at the one time commercial. Then the second thing that they do is trying to focus of specific audience, right? So they’re sending, sending, as I mentioned, like personalized videos or different different types of videos with different avatars incorporated into those videos to try and catch the attention of those of those audiences. Right? Then when the audience come back, then what they can do is they can actually have an avatar on their website, communicating with people who are actually going to the website and answering questions or guiding them to the website. The next thing is actually having a more personalized interaction. So you can in those we see a bit less because it’s just the beginning of it, but it grows where actually people are opening a live chat with an avatar and allowing them to speak with their customers and actually kind of qualifying or selling something or offering something, and then the last piece of it is a customer service. So we see today more and more people who are actually replacing their chat bots or enhancing their chat bots with an avatar based chat bot, which, in fact, what you see on our website, but not from a marketing cell perspective, more from a service perspective, right? So it actually file tickets or support requests, and then the avatar can speak about what’s being done and how they’re being supported and so on. And the last piece of it is customer retention. Customer retention is around like if someone wants to cancel their subscription, so the avatar can actually help with understanding why, and maybe even even keep this customer for for longer period. I’m telling it in a very generic way, because we see a lot a lot of different industries using us, from from banking and insurance to retail and to telcos everywhere, and also small medium businesses like like yourself, for example, they want to have create several touch points with their customers, or customer base, or listeners. Just create a few content pieces, and then they create an avatar. The Avatar speaks with their audiences, and so on. So we see it from all the spectrum.

Michael Waitze 13:31
I am sure that you don’t have stats off the top of your head, and I’m not interested in the exact statistics, but I will say this, and I’m curious about your opinion on this as well. I know that, like, for instance, when I post something on LinkedIn, if I just post text, let’s say I get like, 1000 views. I’m making up the numbers, yeah, if I put a photo there, some kind of image, I probably get 2000 views. If I put a video there, though, right? Any kind of video, particularly a video with a human in it, I get way more interaction and engagement. And I’m just curious if that’s part of this too, right? So I when I see a chatbot, it just kind of annoys me, because I know it’s a machine, and I feel like I’m interacting with the machine. And I think everybody feels this way, but I think if it was with an avatar, even if, even if I know, like you said, because you’re not trying to trick people, you’re saying, Here’s a human representation of this information, essentially talk to them. If that’s what I was doing, I feel like I would be way more engaged. Is that what you guys are seeing as well?

Igor Hofman 14:27
Yes. So it actually has two, two levels, tell me and what is mentioning? The first level is the first one, which is why video content is better than non video content. Okay, this is very, very evident. We see it all around. I don’t have specifically, I just want to we know a few things so we know that people reacting to video content in a much better way than non video content. We see it all over. It can be educational, content can be promotional. It can be just, just information. But really see it, I think that there’s none. Here that like engagement and memory of the content is 90% higher when it’s video versus when it’s just text, makes sense. I’m not sure. I’m not sure about images. And that makes sense even when the when the person doesn’t know, even when the person is aware. This is an avatar, yep, still having, having the video there, having the person they’re having the face there make something does something to the brain. So he gets more engaged. He kind of, kind of even primal when someone sees the teacher they’re watching and trying to see what they what she has to say. But if they doesn’t, don’t see it, just move away to something else. And so this is short, and we see that. And the second, second thing is about, is about the actual, the actual interaction with, with the avatar and the same the same thing, same thing happens there and when, when people, when people talk to chatbots. So there is something already ingrained in them. They know it’s AI based or not even AI does it pre scripted. There are, like, multiple choices, decision trees and so on. When someone appears in front of them, they start conversation, something changes, and they actually have a more natural conversation and a bit less annoying about the conversation Exactly. It doesn’t it doesn’t happen anytime. It really depends, and this is why it needs that. We always recommend our customers to be very careful about where, where is the highest impact? Because if you are, if you’re, if you’re like, driving in your car, and you want to have customer service or support for someone, you don’t need an avatar. You need someone to look at. It’s just going to be destruction for you. But if you are now in a place of comfort, you’re in front of your computer or in front of your TV, and you and you want to have a more, I would say meaningful conversation, but it’s more of a insightful conversation, then the visual makes a lot of sense. So we see a transformation from chatbots towards AI assistance. This is like the code name here, but essentially it’s chatbots that are based on two things. One is large language models makes the conversation more more fluent, more engaging, not decision choice that they used to be before, and like picking up one of five choices and so on. And second is having the avatar as the interface for this conversation to actually have this the conversation more natural. So do you

Michael Waitze 17:15
want to talk a little bit about Nui, these natural user interfaces at scale, right? And I’m super curious. Again. I don’t know the answer this too, but the math behind the development of taking a photo and turning it into a video, or even taking a video and then turning it into this interactive thing has to be super complicated. Anyway, talk to me a little bit about Nui so I can learn more.

Igor Hofman 17:35
Very complicated. We are taking it a little bit for granted in this conversation, because we are very several years just to give you, give you an idea. So we in our company, not only ours, companies in our category, we employ the most smart people, yeah, the planet. This computational PhDs

Michael Waitze 17:56
must be just like falling out the windows in your office. You know what I mean? They must be everywhere,

Igor Hofman 18:00
pulling out our windows very carefully. We’re not talking about AI deep learning researchers that are coming from the academia and applying this to the to business world. So the algorithms are built in a very, very accurate and sophisticated way. But let’s, let’s put aside for a second what we believe. I’ll talk about Nui, but I want to start by explaining where we actually came from. This is, this is our long term vision. And so we believe that the real or the new way to interact with anything digital, and talking about services, appliances, and pretty much anything would be different than it is today. So today, where you were used to something called GUI, graphical user interface. This is and again, we’re used to it for the last 20 years, but it not always been this way. No before that people remember it. It was command line. Command lines. And then we started with something more graphical. Some people are used to having buttons and links everywhere, but actually, if you think about it, what’s the easiest way? What’s the easiest way to communicate and talk about it before talking to people?

Michael Waitze 19:11
I’m much can I just interrupt you for one second? Because keep the thought. I hate typing into chat, because sometimes the thought is too complicated to explain, and there needs to be a little bit of back and forth, like, here’s the thing, I think here’s the answer, then I have to go back and forth and do this. It’s much easier. Sorry, go ahead.

Igor Hofman 19:28
Yeah. So this is exactly that it is. It is not natural. It’s very efficient, because this is the way we used to speak with with machines up until now. So the easiest way is to speak to someone. If I would say, I would bet that anyone would ask, What do you prefer to go on a website? Or if you had immediately appeared a person in front of you, you can just ask them having a personal assistance for everyone. If this was possible, this would be ideal. It’s not possible, but we’re not. Double the position of our planet. Also, not everyone can be can have assistance, and it’s very expensive and it’s not really, really feasible. So people want to have those type of assistance, which is more natural. Now, there was something in between. A few years ago the voice movement started. We all know Alexa and Google Home and all the theory voice assistance, they are, they are like half step towards where we need to be, but still not yet fully adopted. And I think the reason for it is because it was what. There are several reasons. One is because the actual conversation was not yet ready, or was not in the level that people were expecting. But second, something, something else was missing. We believe what was missing is the actual visual, the face, yeah, the face of the person. And what we believe will happen is that in talking about long term, they were changing this from graphical user interface into natural user interface. Natural is a general world word, but what we mean by it is that eventually you’ll go on a website and the first thing you see is a is an AI assistant or a person asking you, why are you here for how can I help you? Yeah. And then immediately you start a conversation. If you want to have it in this way, great. If you want to keep on driving your car and do it in a voice way, also fine. But the first engagement is a very personal, very natural engagement where you see a face of someone, preferably the someone is someone who you most relate to. It can be someone in your as you mentioned. It can really tailor this to be in your age group, in your ethnic group, or whatever, not to fool you, just to make you feel more comfortable. Yeah, exactly people, when they see people who are more resemble to them exactly, or who they don’t divided by, or authority wise, or so on. So this is how we see the future. We see it with natural user interface based by AI assistance that looks like people and that speaks and interacts like people. So

Michael Waitze 21:54
are we going to get to the point where, like, I’d love to have a personal assistant today, but you’re right. I can’t afford one yet. Yeah, almost there, but I can’t afford it yet. But if I had an avatar that I could just say answer this email with this thing, please make an appointment for me to do this thing, or all these little things that just take up so much of my time, and that that person you know resembled like one of my friends, or was it part of my peer group, I would do that all day, right? Do you suggest that that’s happened sooner rather than later?

Igor Hofman 22:25
So I’m suggesting it’s already happening today. It’s just not yet widespread as as regular communication with large language models, because even even today, you can already have your own. So it’s not really full assistant, it’s still, it’s still companionship, right? And it moves towards assistance, yeah? Because right now, you can have an avatar using the ID or or just just speaking with with the chat GPT, or with Claude, and they can be your companion. You can actually talk to them, and they can talk back. And it’s more driven by by you, not by them. So you ask them a question, you get an answer. You want them to create you want GPT to create you a PowerPoint presentation or an Excel sheet that they can do those things for you already, but it’s still very, very reactive. What we are doing now, and it’s already being built with our top our top customers, is an AI assistant that is not only informative, but also capable of performing multiple tasks. And this is possible already today. It’s just being built by the big companies, so you will see it out there in several months. Right? Talking about is like, not only so you have an AI assistant, you can ask them a question, they can give you an answer. Then you can say, let’s say, for example, it’s a bank, so you want to do something, some transaction in the bank. So the AI, the interface, is the person you’re speaking with, but behind the scenes, there’s large language model. It also interacts with the bank systems, so they know how to make authentication. They know how to make transaction itself and everything happening behind the scenes, and you get the response back, which is a visual response, transaction is made, or your loan is approved, or whatever, and everything, everything is done. So those things are currently in motion. They’re requiring some deep integrations of multiple systems, but it’s already possible today.

Michael Waitze 24:12
So this brings up another issue, and again, you may not have the answer to this, but I’m just thinking kind of on my foot here, if I’m working with a bank, and I’m dealing with an avatar, which I’m perfectly happy to do. And I say, Can you please move $10,000 from this account to that account? And blah, blah, blah, do whatever. And I’m like, Oh, actually, can you make it five grand? But they already moved the 10 doing me like, who? If it’s an employee, like, I know their name. I can know their employee number. I know the person responsible. In this case, like, how do you handle that other issue of who’s actually accountable for this? Do you know what I mean?

Igor Hofman 24:47
Yeah, it’s a good question. I think this is why it will not happen tomorrow. And maybe banking is not not the first one where this will be applied for. It’s all about the process, defining process, what we’re doing. Your customers is we’re actually designing the solution and designing the journey that the customer or the interactor is having with with the system. So everything you mentioned here is logics. You just need to define them. So, so we have people who are responsible for customer experience. This is a new it’s a new term now, CX customer experience. Yeah, we you see many places now they have Head of Customer Experience, Head of Customer experience using AI and so on. They are responsible for designing how the experience will look like. And those are one of those things. But it’s very simple to create, like safeguards and boundaries and limitations around what’s possible not possible with the AI right at the end, eventually, if you look at it, from a few years from now, everything would be possible, but to get there, resign those processes properly. So

Michael Waitze 25:49
when chatgpt first came out, you know, we talked about all the parameters. First it was, who knows how many, and now it’s trillions of parameters. Most big companies though, to run a proper, large language model that’s specific, so a smaller, lighter version of an LLM don’t need that many parameters, and in some senses, can really customize it for their own uses. I can see cases, and I can see Avatar cases for this as well, where internal like company learning is based on all these kind of internal company wikis that are constantly being updated. And if I have a question about some company policy, or my insurance policy or anything, I can literally go onto a website or even internally and just go, can you tell me what my policy covers? Does it cover this thing or that thing, or the other thing, or I want to learn about this part of this business on the weekend, so that when I go in on Monday I’m just more informed. I presume that these lighter, smaller llms paired with avatars and then fine tune just for specific purposes are already being used, yeah? Yeah.

Igor Hofman 26:47
So you don’t have to have a smaller llms. So the way it works today is something called LLM, grounding, grounding LLM, or maybe you heard the term vector database. Yeah. What it means that you take, you take a large language model, it can be GPT, can be Claude, it can be Mistral, whatever, even if it’s very big, train on a lot of parameters, and you kind of direct into the right direction. You say, you also, you know everything you know, but now you also, you also have this specific vector on which you are the specialist, and then, and then, the way to do it, this is like me simplifying it a lot, but, but the way to do it afterwards is that you define, or the company defines, what would be the default versus what would be like the escalation path. For example, we’re working with an insurance company. We we made it clear that this specific AI assistant is responsible. They know everything about life insurance, specifically about like, five or 10 different policies. This is what they’re expert in. Now the company can decide if they want it to be completely grounded, which means if the question or the task is around those policies, they can answer if it’s outside of those of this space, they just say, I have no idea. Don’t ask me about those things. I have no idea about it. You can ask my colleague This is being fully grounded. Then you can be you can be hybrid. Hybrid means that if you don’t know the answer, then you can have some kind of a relay. If you can say, I don’t know the answer to it. But according to the to the internet, this is the answer. So you can kind of go away from the vector outside, or you can be completely ungrounded, and then they always answer, regardless of the answer is the right one with from the vector or something coming outside. So this is possible to do today. It’s also to do actually, even when you create an agent on on the ID, so you just upload the information the database, and then you can decide if it’s ungrounded or hybrid or grounded. Okay,

Michael Waitze 28:48
a couple more questions, and then maybe I’ll let you go. This is so interesting, though. I just want to keep you here forever. No, does it? Would it be helpful for and I think we talked about this before you started recording, right? Like I wanted you to move your camera angle so I could see you better. I like to have a nice microphone so I sound good. I don’t want to sound like this because it’s harder to hear me. But will it mean that when I’m interacting with an avatar, if I have slightly better equipment, that most people are going to end up being equipped with a slightly better camera, a better microphone, so that the avatar can actually sense them in a way that’s better. Does that make sense? Yes,

Igor Hofman 29:26
it’s a bigger it’s a bigger topic. Yeah. And actually, it’s a good topic for us, for us to kind of kind of conclude. And the way this the path and the place where this is going the avatars is into a place where the avatar is more human, meaning, more emotional or more responsive, and it comes from two directions. So one direction is how the avatar itself reacts to the one he’s speaking with, right? So I. We are moving into a place where we can do expressions, facial expressions and voice intonations. I was gonna ask, yeah, so it’s something that we’re working on right now. It’s still not 100% there yet, but we’re working on it. I believe by the end of the year we’ll release our first expressive avatar. Right now, we can do some expressions, but on an ad hoc basis. So this is us, but it’s all going direction. And so expressing emotions and being more human, it comes from two places, from how the data looks like, like the expressions they have, how they sound, the intonations they have, and also what they say, the context itself. So this is on the output part. On the input part, there is something called sentiment analysis. Yeah, we already doing it from a contextual perspective, so which means that, based on, on how, on what’s, what’s, your sentiment of the text that you are, you’re talking to the avatar with. So if you’re talking to an avatar, you say something, everything you say, turns into text. Then this text is being analyzed. And then, and then we can decide, or the LLM can decide, if the response is to calm you down. Going forward, there will be more cues that will be gathered. For example, as you mentioned, cameras, voice recorders, all those things will be captured so the avatar will eventually know based on how, on how they sense you from the camera, if you are happy or not happy from your face. Also like facial recognition and facial analysis, the same ways for your voice, right? So going forward, the avatar would be completely, completely expressive and also analytical. So they know. They would know based on all the cues they’re gathering, they will give the response both contextual but also emotional, yeah,

Michael Waitze 31:48
because that’s key, right? I mean, look, there are certain phrases in every language I could say to you, get out of here, which just means, like, you know, I’m not taking you seriously. Or I could say, Get out of here, which means leave the room kind of thing, right? So very simple phrases, the context can be different and the sentiment can be different. And if you different. And if you can hear it, as opposed to just reading it in the text, right, you can get a better sense of what that person is thinking. I can’t wait to see what this stuff is going to look like. Do you want to comment just at the end about the ethics of all this stuff? Because you mentioned in the middle, and I just want people to understand that you are not as a firm. You want to be flag holders on this. You want to tell people that it’s not trying to be trickery, that it’s just trying to for productivity and for helping. I just want to make sure that that’s clear to people. If you just want to say one more time,

Igor Hofman 32:29
yes. So there are also several, several elements, please. So on the one, on the one hand, again, from a from a statement perspective, we believe in the ethical and responsible use of AI in general, which means that we don’t want to see misuse, and we also want to see very authentic use of what we can offer. So which means that we are encouraging people, and not encouraging only, but we’re also enforcing it and having watermarks and making sure people understand that this is an AI generated content and moderation controls. We have them on what people, the avatars, can actually say, and so that they cannot be, they cannot be processed or designed in a way that they can say things are offensive, not only offensive, also also political and stuff like that. There are some, some ground rules that we’re enforcing on our side. The other other side of it is actually, actually going into authentication of AI content, making sure that every content is out there it’s clear if it’s an AI based content or not, and also if it was created in a proper way. This is what I can say from a title perspective. So we’re currently working with some of the bigger these companies in the world. One of them is Microsoft. We’re in partnership with them on that. We actually have an entire large plan about the responsible AI and the use of AI in general, with avatars there. Awesome. This is one of the things making sure that people that it is, that it’s a content is out there. Not only we, people know it’s AI based, also they know it was created in the proper way throughout all the protocols. And there’s one more side of it, which is social impact and doing good with AI, right, which we are, we are very big believers of it, and we’re doing a lot of pro bono around it, not only pro bono, but in general, projects around bringing awareness, helping people, and with the use of avatars. So there’s a lot, a lot to it, but altogether, we believe that the right way to use it is leveraging the positives by kind of limiting the negatives and trying trying to see how it’s done for the right way, perfect.

Michael Waitze 34:45
Okay. Igor Hofman of VP of Sales at D-ID, thank you so much for doing this. That was really incredible.

Igor Hofman 34:51
Thank you very much. It was a pleasure.