Swarajya spoke to Senthil Nayagam, a content creator who works primarily in the hotly trending areas of generative artificial intelligence (AI) and deep fake space.
He runs a startup in Chennai, Tamil Nadu, called “Muonium AI.” He and his team have been behind viral AI content in the entertainment and political space this past year — most famously an AI avatar of the late Tamil icon M Karunanidhi, but also before that an AI rendering of the Kaavaalaa dance performance showcasing a switcheroo between actresses Simran and Tamannaah.
Over the course of our conversation, we picked Nayagam’s brain about the use of AI in politics and its implications, especially with the 2024 Lok Sabha election fast approaching. We also asked him about the regulation of AI, the ethical use of the technology, government funding for AI in India, and how AI can change cinema, among other things.
Here are the edited excerpts from the interview:
We are less than a month away from the start of the 2024 Indian general election. Political campaigners are probably hard at work, as they have been for the last several months, if not more, as tends to be the case before elections. But interestingly, this time they have had a new tool that they could turn to — AI. How do you think AI and deep fake technologies have found use in political campaigns in the run-up to the election?
Mostly, all political parties use whatever new trend comes in — a new song, new imagery, or new technology. So this time, they accidentally got into AI. As a capability, it was there for a couple of years. But it’s a lot easier today than ever to create content in that format.
So you want to get a meme, you want to recreate a person, or let’s say, resurrect a person, a politician, maybe create a parody or song — those are all possible. It’s a lot easier with AI. That’s one part.
The other part would be maybe personalised messages to voters. It won’t be a generic message. What if each voter can be addressed separately? If I say, “Hi, Karan, you are from this part. Please elect me.” So once somebody says “Karan” and he knows about you, your location, which candidate (is from there), you feel like somebody’s paying attention to you and you’re being made to feel important, right?
Most people feel their votes do not matter. That’s a challenge which people have had. But if people feel the leader knows me or the politician knows me in some way (that will change things). People realise it’s AI-generated in some way or technology-generated, but still, people will have a feeling that at least people notice me.
That’s significant because, back in the day, there were the traditional methods of, say, people coming door to door and talking to us. Now they can save time, and you can have someone at your doorstep, except that they aren’t really there. It’s a tool that they are able to use and still try to make that connection with the voter. How do those two methods compare, you think?
As a kid, I saw elections in Delhi in 1984 and 84 onwards — every five years, three years, state elections, and central elections will keep happening. We have seen different versions of it (election campaigning). Earlier, it was a pretty noisy one, where people were going (around), they would shout, they would give you pamphlets, they would knock on doors.
If it’s a small constituency, it’s practical to do it, but constituencies have become bigger, or the population has grown significantly — I think we are 3x or 4x of the population we had in the 80s. So, it’s not practical to go to every nook and corner.
Also, the election duration has kind of changed. It is longer, but from candidacy to the election part is actually smaller for now. And also, a lot of digital candidacy. Social media has come in, people can take photos, images, and everything.
So, the idea is: can you change people’s perception? That’s what people are looking at. Am I electable? Will people choose me over my competitor? That is how the system goes.
I have seen somebody build a VR (virtual reality) solution where, with AI-generated content, you can make the politician come knock (on) your door or somebody representing him will come and you can see him talk to you in a customised way.
So, technology has come a long way from there (door-to-door campaigning), but maybe this election will be more of a trial basis. Maybe in the next election, in one or two years, we will see a lot more happening in this space.
And another unique thing is that we have a huge cinema industry. So, creative people are involved in elections in some format, and some political parties have channelled them in. We have the capability of producing content. Also, all the digital content is a lot easier. With computers and a couple of instruments and software, we can do it. And with AI, even more possibilities are coming.
How do you think political parties and candidates have taken to AI in their campaigns? Have they been quick to get on board or have they been testing the waters, just trying it out? What’s their level of embracing the technology, in your experience?
The system is fundamentally changing. Content is not a single format, multiple formats are happening. So you are doing more of a feature parity. If your competitor is doing it, just to match up, you have to catch up, you have to do it. If you don’t do it, people will think you are lagging for whatever reason, or you don’t have a counter to it in some way.
2014 onwards, the BJP has been ahead in this part. I think the Modi campaign used virtual images of Modi at that point in time, on a 3D version of him on a smoke screen. They did a lot of interesting content way back in 2014. Now, other parties, even smaller parties, are considering that. That’s a unique thing.
Another challenge is that India is a large country. We have nearly 1.5 billion people, and nearly a billion people, about 96 crore, 97 crore people are going to vote this year. That’s bigger than probably all the other countries put together, in terms of democratic countries. China doesn’t have a voting system, per se. The US in comparison is about 300 million people. We are at least 3x of what America is.
Another challenge is, we have a lot of languages. So maybe state elections have been dominant, but when you have a national leader who goes to a lot of different places — India has about 22 official languages — the content also needs to be in a different way. So, there are challenges, and technologies helping in solving those as well.
So we’ll have two or three types of content. It will be national content, which will be on large events and physical events. Then there will be some kind of samples that can be used on television and social media. Totally personalised content. So these few areas are where probably a lot of new things are happening.
Physical campaigning will still happen. Each section of voters will be swayed. So somebody might have already chosen the party. Sometimes people choose, “I choose this candidate over the party.” So there are different schemes of thought going in.
And there are new voters, undecided voters. The idea is, who can sway the undecided voter to my favour, or can we make sure he doesn’t vote for my opposition party, the candidate opposite to me? That’s only two areas where probably a lot of work is going to happen. Because the decided voter is anyway decided, you can’t do much to change him anyway.
And the new voters will be 18 to 25, first-time voters in some way. This is the generation which has seen TikTok and Instagram Reels and Snapchat and YouTube. They have less patience. They get easily distracted. So they need a very concentrated effort, so that would be a few minutes of content. One-hour content won’t work for them.
Three minutes, two minutes, or 30 seconds is what probably would work and sometimes you have to remind them why to vote for a person because they also forget easily. That’s a market which probably a lot of people are trying to address today with technology.
I like how we’re looking at the possibilities that we have in terms of technology tools and how we can use them to the wide Indian market. It’s a fresh breath of air because generally there tends to be a strong sense of tech pessimism, not just in India but around the world as well, where people gravitate towards the negative possibilities of new technologies like AI. As an AI-based content creator, do you see AI impacting Indian democracy in a positive light? What can it do for Indian democracy?
News coverage can be negative because that’s easy to spread. Positive news generally doesn’t get much traction. That’s the reason why people go into that (negative) part.
See, India has been tech-positive for a very long time. I remember the 80s, there used to be videos of booth capturing happening. I remember seeing, I think, India Today, they had a journalist who was capturing live videos of booth capturing happening. So I’ve seen those kinds of videos.
And after that, I think we had T N Seshan (former politician and chief election commissioner). He regulated parties in a particular way. If you don’t do it in a proper way, we will either not let you contest or if you have a criminal case against you, a lot of disclosure has improved from there.
And then we moved towards the EVM or electronic voting machine. So we are tech-savvy in that sense. That’s not a Bluetooth device or a software, Android device, but it’s an electronic device manufactured by a Government of India PSU and we have been using it for decades now. It’s a simple system.
The election system, if you notice, from independence onwards, we have solved a lot of fundamental problems which probably a country like us — we were a newly independent country, we had poverty, we had people who were uneducated at some point of time — we had to solve a lot of these problems.
So, in the first part, we need to have a party symbol. A symbol will come in a particular way. It could be a bird or animal or whatever. So people choose a symbol. In the second part, when you’re doing paper voting, you can put a single mark, as some countries are doing it. Then we have it with a cross. Then you realise the ink can spread on the opposite side, we have this curve as well. Again, voting fraud might happen, so we invented the indelible ink, which we are also sharing with Africa.
So, in terms of technology, this is all indigenous technology. We are not importing it from some country. We understand democracy better than a lot of other parts of the world. People assume that some people in some states are very underdeveloped, and that they don’t understand politics. But grassroots politics, people understand.
Like Bihar, people assume things… Yes, corruption and other problems are there, but people are very political. They are very aware of what is happening. Somebody sitting in another state which is well-developed may not understand the local politics much. They know change is required, and for various reasons, the current system may not help them that much, but people are still positive.
Yes, I would prefer to have a higher voting percentage. That doesn’t happen for various reasons. But I think we are a vibrant democracy. And we are tech-savvy. When we were doing voting machines, other countries didn’t know voting machines or multiple technologies.
Another part was voting on multiple phases. Wherever there was a possibility of violence or any assistance required for it, sensitive zones were identified and we made sure everybody got to vote. We have several central police forces used for these purposes.
Another part, we have the voter ID. In the US, most states are passing laws that allow illegal people to vote. That’s how they are doing it. In India, like we are saying, get a voter ID if you can. The only thing which is probably pending would be, that there may be a lot of fake voter IDs possible or when people vote on behalf of dead people, whose names have not yet been removed. This might still happen.
Maybe this will be solved once the Supreme Court allows voter ID and Aadhaar to be linked in some way. Once biometrics come into play, maybe it will fundamentally change how it works. So, in terms of technology, I’m very happy with how elections are going.
We need more improvements, I am not denying that we need more improvement. Maybe the queues can be smaller in some way. We could plan better in some way. Or the candidate information, though it is public in a particular way, we can improve certain other ways of communicating with citizens. Maybe that will help us improve the elections even further.
Senthil, you are leading the charge as one of three signatories to the “Ethical AI Coalition Manifesto,” which outlines key principles for the responsible development and use of AI technologies. Can you speak to the motivations behind the manifesto? What made you come up with it?
The whole fear of AI… the fear of the unknown is always there. Generally, there are three things which people use against any technology: “FUD” — fear, uncertainty, and doubt. This can be against a technology, or it can be used against a particular candidate. What if a candidate doesn’t do this X, Y, Z thing? What if he is corrupt?
Can AI be misused? See, every technology can be misused. We have seen in 2001, on 9/11, they used aeroplanes as missiles. After that, at least we have improved security and processes in some way. After the shoe bomber incident, when we travel to the US or other countries, all our luggage, everything, has been X-rayed now, right? You have to remove your belt, metal detectors are being used more. So, for mitigation, there will be some techniques or technology which will come into play. That’s how that works.
Every tool can be misused. Every tool has a positive use case. A knife can be used for butter, it could be used for surgery, can it be used for stabbing? Yes, all three are possible. But you can’t do a blanket ban. People who are violent will find a way. A book can be used for beating people. A toothpick can be used, right? So if we go on banning things, we don’t have anything left with us.
We are a startup. Me, Varshul from Dubverse, and there’s the Indian Deepfaker. And it’s not only three people, there are more startups who have read it (the manifesto) and they are getting formal approval. Some startups need to get some permission from their teams or lawyers or investors, so they can sign it. Because if they sign it and then they can’t pursue a certain business, will it impact their path?
All I’m saying is, are we ethical? Are we decent? What things do we do, what we will not do? And we are voluntarily offering this path. We made sure we announced it before the elections were announced. Because we want to clear all the doubts. If I am publicly available, visible, giving all the media interviews, and I have a startup, my number or contact is public, what is the reason for me to create bad things? I will be implicated very easily in that part. It doesn’t make any sense or business sense to do it.
As a startup, I want to be known for good things. That’s what I work for. My startup is not an election startup which is doing election work. We are more into creating content. We are more into: Can we make use of AI to create movies? Can we dub movies in different languages? Can we improve the content? Or can a small team of creators make a movie? That’s what my startup is about.
Elections will finish in two months. But if the government passes one wrong law or bans everything on a blanket ban, it will impact not only me and fellow startups, it will impact hundreds of potential ideas and startups which might emerge later this year.
And AI is at a very nascent stage. It’s growing very fast and there are several more applications that will come. A lot more will happen. Just because some politicians, judges, or bureaucrats do not understand the implications and they pass some laws, rules, or restrictions, we will lose trillions of dollars in the future.
This is a very meaningful moment, is what I feel. So we felt a need to announce our understanding: hey, we will not do these things. Anything to do with CSAM or child sexual abuse material, we would never create that part. And the first point is to do with the election part: ‘Hey, we will not interfere in the elections in that way.’ We create content, positive content. We will not create any fake content. So we have put in seven-eight points.
This is not legal wording at all. We have made sure a common person who reads it understands what is being said. This is not some 100-page document which probably most people won’t read and will not care about. So, as the startups in this space, we say we have the capability of doing it, but our intention is to create good content. That’s why we are announcing it and this is the purpose for us to exist.
In this context, what did you make of the now-withdrawn advisory requiring explicit permission prior to the deployment of AI models, LLMs, and algorithms?
It’s a very vague document. The terminology is not determined. Can I make a parody of you? Can I make a meme of you? What is not included? Then each time, come to the government. Have you been to a government department and got permission? In my previous startup, I ended up trying a few things. We don’t get any feedback from the government, or they have a preferred source of technology.
I’m sitting in Chennai. If I have to go to the MeitY office in Delhi, get an appointment with the IAS officer, and showcase our capability, server, and everything, it will take weeks, or months. And what if they never acknowledge it, right? I can’t have to go to court and say, can I release my product or not? So this whole process kind of meddles with the whole purpose of doing it.
The startups are like, “Hey, if India is not a very large market for our services or products, why should we remain in India? So maybe we will become a backend like SaaS (software as a service) companies.”
We have so many problems with credit card operations. They have restricted a lot of things. So most SaaS companies are incorporated in the US. It’s easy to process payments, and the target market is the US. India is still a very tiny market, even for Freshworks or Zoho. It’s not a huge market for them. Though all the back-end operations are here. The only thing is because the differential is there — salaries are cheaper here, and operating costs are better.
Do we want to do the same thing for AI? The whole part is that, right? So AI is a new market, a new opportunity. And we prefer to stay in India until and unless the government forces us with some bad laws or something else to move away from India.
You’re developing an LLM yourself, and it appears that you’re basing it on Elon Musk’s Grok AI chatbot code, which was made open source. Could you speak about that?
We evaluate whatever new models or tools are launched anywhere. Anything which is open source, even if it’s a Chinese AI model, we will try that part. I don’t deny it. Because what is the state of the art? When you want to work in this space, you have to identify who the state of the art is, who is the new contender, and who is the new possible winner or emperor. We have to explore all the models. That’s how we do it.
So, on the LLMs, the Grok one, which Elon Musk has released, that’s a huge model, that needs several GPUs, and my preference is to run all these locally. Think of it: I want to buy some GPUs, we have to pay all the money in advance and wait for several months. That’s how it happens. And currently, with importing GPUs and hardware, there are a lot of challenges. So whatever is locally available is what I would prefer to use.
Running AI operations or anything on the server part, we are nearly 2x what it costs in the US. For various reasons, let’s not go too much into that part. So, running a Grok model would need about 140 GB of RAM, and GPU RAM, which means you have to have a H100 or H200 kind of server and that will be about Rs 30-40 lakh. That’s not an investment which I will make at this point of time and it’s not going to solve the problem. So, people are going to quantise it in some way, we will do that part.
Smaller models like maybe Llama 2 or Llama 3, which is a 7-billion (parameter) model, are more useful for us in some way. There are several other AI startups that are coming up in India, excluding the Krutrim part, who kind of meddled with the water in some way.
There was a huge valuation, but the product was bad in some way and they’re now tying up with Databricks. That’s a different story. But there are a lot of small individual AI developers, individual small companies, who are actually building their own AI models. Taking a reference model like this (Grok), they are building a data set for Indian content and training.
The whole purpose of me doing it in India is India’s large market for movies and everything. And we are not doing only English or Hindi. We have to do all the Indian languages. Can you cover all the 22 languages? That’s what I would work for. So I do more on the image, I would do more on audio, voice, and video.
With LLM, we are exploring how we can work on it. I will prefer an independently developed model which people in Hyderabad and Bangalore, other people are building, which are in principle trained on Indian content for now. That’s how I’ll prefer it. The Grok AI model is cool to look at and demo, but as of today, it doesn’t solve my use case. I’ll prefer to have a smaller model with a rag and I prefer to run it on local machines.
The next trend that we are seeing in AI is distributed AI. So one of a few companies controlling the whole internet, the whole AI model is not a model which I would prefer. I will have my own independence and own way of running my own models. We prefer that kind of model. Though we are exploring all the models.
My personal preference would be for smaller models which you can operate where the running cost is lower and which you can offer to other customers as well. ‘Hey, take a bunch of GPUs from us and we’ll set up a machine for you where you can run all your models at a low latency for your use case.’ That’s how I see the future.
In that sense, we are also talking a bit in the direction of sovereign AI. It’s been proposed in India, outside as well that we must have AI for the region that we are building it, localised AI.
Currently, AI needs a couple of things. You need to have a lot of GPUs and a lot of time to train them and big data sets. Now a trillion tokens is a normal number. That’s how it has come in with Nvidia launching exascale computing. So (the Indian) government made an announcement, about Rs 10,000 crore is what they want to invest in AI for the next five years. That’s about 2,000 crore per year.
And Yota is one data centre provider. They took nearly Rs 2,000 crore of delivery of GPUs last month. And I think they did a puja for the GPU. That’s what the news said. It’s a small Indian company. This is not a top-10 company in any way. Maybe a real estate-linked company, but they have raised some funding. And by this year’s end, they will have Rs 8,000 crore to 10,000 crore worth of GPUs just for operations in India.
India should be doing a lot more. So Rs 10,000 crore for five years is a very, very tiny number. We are not even (among) the top 100 companies, who would have been spending more than Rs 2,000 crore per year.
So if India wants to be a leader in this space, maybe Rs 1 lakh crore for a year or two, that’s a good number to get started. That is the only way you can supply GPUs to us startups for solving interesting problems. And then we can monetise it, then we can build our IP.
Think about this scenario: the India-US relationship is kind of slightly shaky in the AI part. See, the US election is also going to come in. What if they say ‘Indians might build deep fake content and make Biden lose’ — that is a story somebody pitches to the Biden government. They may deny access to all the AI models. It’s as simple as that, right? They can deny access to Gemini, ChatGPT, Claude, and three or four models which are operating from US companies. Currently, leaders (in this space) are all US companies.
So we need to build our own capacity. But should it be government capacity? The government probably should provide an interface, and infrastructure, and let people work on it. The current system is kind of slightly broken is what I feel.
I’m not saying against the government, maybe it’s election timing, they don’t want to announce too much on that part. Maybe after the elections, if the new government is formed, they should be generous on the AI part because this is going to have a huge impact on how civilisation develops.
Finally, I think you’re also venturing into AI film production. It’s probably a relatively new space in India, maybe even in the world. What’s your sense of how AI will change the space of cinema? How different is it from the CGI and all the other technologies that are being currently used?
The systems CGI, and VFX are how we used to have medical transcription. We used to have a bunch of people who would listen to content, they would convert it in some way, then there would be somebody who would do the coding part, and then they would deliver. So it was more of a BPU, KPU process, which is what VFX and CGA was. I’m not saying they were not smart, but the value-add, there was no motivation. It’s like a government contract.
Think of how NASA works. NASA in the 1950s and 60s and up to the 70s, they were trying to go to space, go to Venus and Mars. That’s what their aim was. I think, for the last 50 years, humans have not landed on the Moon. Think of it, before I was born, the last Moon trip happened. So somewhere, government organisations get into that part.
Something similar happens with VFX. Think of it, Jurassic Park, the first one, the 93 movie. They had a couple of minutes of CGI VFX thing, but the storytelling was what the important factor was. So that was a blockbuster. Now the whole movie (landscape) is VFX-CGI-driven. What happens is the cost keeps going up. The duration, overshooting will always happen.
And also, you need to have a lot of people doing it. Let’s take, for example, Avengers: Endgame. If you see the IMDB listing, the end credits had nearly 2,500 people working on it. And the movie cost nearly $300 million. So, if for the same quality of effort, fewer people are required, maybe the cost will come down, (especially) if AI can do a certain part of it.
Think about it. The Hanuman movie was made at a far lower cost than that other movie which was supposed to be based on Ramayana, right? So they made a Rs 500 crore investment. I don’t know. Was it a money laundering scheme or whatever? Why did that cost (that much), the quality was so bad? So people are going to question this part of it.
I’m saying, that interesting creators with good technology can create amazing content. That’s the whole point, right? We have 1.5 billion people. Each person has a story. We have 1.5 billion stories to be made. The current system of movies and workflow will not work. So, I guess, can we take the best of our stories? Can somebody fund it? Can we cover for it in some way? Can we distribute it in different models?
We also actually have fewer theatres. For our population, about 10,000 or 15,000 theatres are not enough. We need several more, but the problem is our real estate cost — because of our population density, the real estate cost is very high. So for the same price which I own a home in Chennai, I can buy equivalent or even better in Japan. Or you can buy a bigger lot in, let’s say, Texas or some other place. So we have to solve a lot of other problems as well. So we are solving it like movies could be a good way.
And I would like how K-pop and K-drama are. Koreans actually planned it and executed it. That global dominance, that’s the soft power part. India has to deliver the soft power in our own way.
Take the best of Indian stories and make them available in all the languages to all the countries. Our stories and content will bring us goodwill and probably more tourists, or Indians will be going to go to those countries and that will help us improve our stature in the world.
