Anisha Sreenivasan: Hi Moyukh.
Moyukh Goswami: Hello Anisha.
Anisha Sreenivasan: OK. So today, we’re delving into the forefront of innovation with Gen AI, the cutting-edge trend reshaping our technological landscape. We will be exploring pertinent questions surrounding Gen AI, alongside some engaging inquiries to delve into its multifaceted aspects and potential implications. So, can you describe what generative AI is in layman’s terms, and what makes it different from other forms of AI?
MG: So generative AI is one of the most powerful AI systems that we have today and has the capability of generating content. Previous generations of AI used to make predictions but could not generate content like Gen AI does. This capability to generate content, be it text, pictures, music, or anything else, makes Gen AI much more powerful. That’s the difference.
AS: OK, now coming to language models, can you describe LLM in layman’s terms and explain its significance in the context of Gen AI?
MG: Yeah. So, the LLM is the underlying program of Gen AI or generative AI, built on top of LLM. Think of it as a powerful program or the heart that does the work. It’s called the large language model (LLM) because it is built with an enormous amount of data. We created a model and then fed it with so much data. It started learning from this data and gained the capability to generate responses for given prompts or questions.
AS: Okay, now coming to its real-world applications. How has Gen AI been successfully utilized in real-world business scenarios, and looking ahead, how do you foresee its potential applications evolving in the future?
MG: Oh, still, it is evolving, and the full potentials are being speculated. There are so many things it’s going to touch in the future. I don’t think there will be anything left. Until now, we used to talk about digital transformation, how it will touch the business, change the business, the whole together, making everything digital. The next step will be the Gen AI transformation, where everything will be converted, and will be using generative AI. That’s my thought process. So going forward, like in business where Nuvepro is, uh, Gen AI, can be used very highly effectively, what is in the place of like self-learning. Today what we have is content and then the suffixed content. People go and read about it in a book or a page or a course. So many of the content providers are not very customized for each user, so generative AI has the capability where to put those different content into the Gen AI and then you can create your own content. So, these are the things I wanted to learn and to generate content for me and it can do that for you. So that can immensely improve the way that people will learn. Uh, the way the same topic, if two people are learning it, the content can be customized based on how the user can learn it. So, this is like it’s going to touch each and everything that is going to come in the future, beat on the way you watch a movie, the way you watch music, the way you read the news, the way you read the content, everything.
AS: Yes, got it. Moving on to data privacy and AI, ensuring data privacy is crucial. What advice do you have for enterprises that need to maintain data privacy while also leveraging AI technologies?
MG: The biggest challenge today revolves around data privacy. As discussed earlier, LLM requires a vast amount of data to train, which is how Gen AI models learn. Many Gen AI models, such as ChatGPT, Claude, or Titan, are hosted on the cloud, where companies collect user data to improve the models. Once this data is used for training, it’s essentially consumed. This raises concerns for enterprises, as their data is being utilized without their control. There are two paths forward: hosting Gen AI models internally within the organization to maintain data privacy or opting for enterprise solutions offered by platforms like OpenAI, although there’s still skepticism about data privacy even with such solutions. In critical situations where data security is paramount, it’s advisable to implement AI models within the organization.
AS: Looking at it from another angle, considering the anxiety employees may have about job displacement due to automation, what advice would you, as a lifelong developer, offer to those in the field?
MG: Both our jobs are safe, right? Gen AI generates content, but does that mean jobs in content writing are in danger? No. It’s essential to draw parallels with historical examples, like the textile industry during the Industrial Revolution. Initially, people were concerned about job loss due to machines weaving clothes, but in reality, millions of jobs were created in other areas. But that’s not how it turned out. Today, the textile industry likely employs more people than in the days of handweaving clothes. I think it’s because now more clothes are being made and worn by people. So, in terms of content development, while we’re automating some parts, there’s still a lot to do. There’s always work to be done and room for innovation. Just look at how vehicles have evolved from being pretty basic to smart vehicles nowadays, right?
And now, even everyday items like bulbs and TVs are becoming smarter. However, there are still many other things that haven’t quite caught up. So, developers will still have plenty of work. Things might change, but I don’t believe anyone will lose their job entirely. We’ll just be doing more things.
AS: Okay. So, generative AI is also said to “hallucinate.” Is this good or bad? And will it affect people’s trust in these models?
MG: For me, I see “hallucination” more as creativity than anything else.
AS: Okay.
MG: Exactly. As long as everything remains factual, this “hallucination” shouldn’t really affect creativity, right?
AS: Hmm. Yes, indeed.
MG: So, when we talk about painting, for instance, if it’s entirely realistic, it’s essentially a photograph.
MG: But that doesn’t capture the essence of imagination, does it? That’s where the human touch comes in. So, I view “hallucination” as a form of imagination or creativity.
AS: Yes, it’s like a manifestation of creativity.
MG: However, there are scenarios where factual accuracy is crucial, like in news reporting or medical contexts. In such cases, relying on hallucination or creativity could be risky. So, while creativity can be beneficial in certain contexts, it may not be suitable everywhere. It’s a nuanced balance between the two.
AS: Yes, I see your point.
AS: Okay. Now, moving on to cybersecurity protocols. With the rise of generative AI-driven cyber threats, how do you strengthen cybersecurity measures to defend against potential malicious users, such as deep fakes or other AI-driven attacks?
MG: This involves two aspects. Firstly, there’s the aspect of attacking human behaviour, like phishing emails, where attackers trick users into revealing sensitive information.
AS: Ah, yes, phishing emails are a common tactic.
MG: Indeed, phishing attacks have become increasingly sophisticated, often leveraging generative AI to create convincing content tailored to the victim. This makes it easier for attackers to deceive users.
AS: So, generative AI enables attackers to create more convincing phishing emails, posing a greater risk to users.
MG: Precisely. On the technological front, generative AI also poses risks as attackers can exploit its capabilities to devise new methods of breaching security systems. So, companies are now utilising generative AI to identify different patterns when it comes to protecting my email, network infrastructure, or applications. They analyze these patterns to determine if they appear legitimate or suspicious.
AS: Agreed. Okay, shifting focus to another aspect, you mentioned how biases in AI models reflect biases in the training data or the people training it. How can we address or minimize this bias during the development phase?
MG: My perspective on bias is rather unique. Consider the traditional notion that doctors are typically male and nurses are usually female. So yes, I’m saying that I need to stay one step ahead now too.
MG: Visualize a doctor. It always needs to be male, and nurses always need to be female.
AS: Yes.
MG: That’s self-bias, right? So, when I say so.
AS: Yes.
MG: It’s society’s fault. That’s just how it is. Our minds are full of biases, and so are our documents. It’s everywhere. LLM learns from all the stuff we give it – documents, data, everything. Since there’s bias in our content, LLM picks it up too. And then that bias goes into the models. Unless society changes, the content won’t change, and neither will the models. Just fixing the models won’t fix society’s bias. But it’s a start. You’re basically saying, “This is how my society behaves, so my model should behave the same way as humans.” But you want to change its behaviour. It’s like asking it to be biased one moment and unbiased the next, but humans can still be biased.
AS: I get it.
AS: OK, let’s shift gears back to LLMs. Specifically, the future of LLMs. They’re competing based on billions of parameters. What’s your perspective on their future?
MG: Well, in terms of parameters, it’s a massive scale. Think about it like the early days of computers. I recall seeing a photo of Bill Gates working on a computer that filled an entire room. Nowadays, computers are much more compact yet more powerful. However, LLMs still operate with billions of parameters. Currently, it requires a massive amount of computing energy and huge GPUs. But in the future, it needs to become more compact and energy-efficient while becoming more intelligent. We can’t continue relying on massive GPU setups consuming hundreds of watts each. We need to emulate the efficiency of the human brain, which operates on much less power but accomplishes so much. Our goal should be to create AI that operates on minimal power yet provides human-like responses, fitting into small devices like watches or phones.
AS: Alright, what’s one fun thing you’d like to see Gen AI do?
MG: There are many ideas!
AS: Like replacing yourself and engaging in some other work.
MG: Yeah, I love that. I could do better things, right?
AS: Yeah. (laughs)
AS: That’s like, you do my work, and I’ll go do something else.
MG: Yeah, exactly. There are so many generative things you can do. For example, if you don’t know how to sing, you could create a song with your voice. And see, probably I would have never sung a song, but today, I could do it.
AS: Yeah.
MG: I have a friend who took just two photos of me and created a video of me dancing. It looked pretty cool, but also a bit scary.
AS: Huh. Yeah, those who cannot sing we can make them sing.
MG: Yeah, we can make them sing, make them dance, do so many things. It’s pretty fun and fascinating.
AS: I think that could even be incorporated into films, right? In cinemas, people who cannot dance, we could make them dance.
MG: It’s like a product for music, isn’t it? The need for playback singers might disappear completely because now, like Rajnikanth, we can have a song created in his voice only for his movie. So we don’t need a playback singer to sing for him. Well, so.
AS: That sounds promising. By the way, there’s a movie coming up in Malayalam with Mammooty in the lead.
MG: Oh? Tell me more.
AS: So they’ve recreated his younger self for the film. He’s in his seventies now, but they’ve made him appear as if he’s in his 30s.
MG: That’s impressive and a bit eerie too, isn’t it?
AS: Absolutely, but it’s quite fascinating.
MG: It makes you nostalgic, doesn’t it? Who wouldn’t want to relive their youth?
AS: Indeed, he doesn’t look his age at all.
MG: Yeah, Mammooty’s name has been familiar to me since my childhood. My next-door neighbour was from Kerala, a Malayali, so we used to watch a lot of Mammooty movies together.
AS: Okay, that’s interesting. so now for the next question. Actually, what are your predictions for the capabilities of Gen AI in the next two years and then in the next ten years?
MG: This is quite disruptive and it emerged quite suddenly. It had been in development for many years, but with the advent of new generations of computing and research breakthroughs, it really started gaining attention around 2021. We initially thought significant progress would happen within a year, but while there have been improvements, they haven’t been as groundbreaking as expected in 2021 or 2022. However, by 2023, we started to see more substantial advancements. Looking ahead, in ten years, I envision Gen AI to be significantly more compact, akin to our brains, capable of being integrated into everyday devices like watches. It should serve as a personal assistant, guiding us based on what we see and hear.
AS: I think it could be beneficial for people with Alzheimer’s, though.
MG: Absolutely, it could be useful for them. But imagine if suddenly you have someone who talks to you exactly the way you want, whenever you want.
AS: Yes, that could be a bit concerning.
MG: Exactly. You might start relying on it more and more instead of talking to real humans.
AS: Hmm, true. But then we might forget how to interact with real people.
MG: That’s my concern too. If we have a personal assistant who’s always there for us, we might start feeling like it’s better than real friends. That could lead to societal changes, which is a bit scary.
AS: It’s like two sides of a coin. It has its pros and cons. Bringing people together and also bringing all the issues (laughs) that look better.
AS: Yeah. The next question is like that, Moyukh. When do you think the common man will come face to face with Gen AI, and what might interaction look like?
MG: Umm. That’s an unfortunate thing, that’s the first thing that people come across. The common people get to know Gen AI with the wrong thing due to deep fakes. People may not have heard of Gen AI, but they have started seeing deep fakes in photos, videos, and news. So these are the first things that people are encountering.
AS: Umm. People who don’t know about AI will actually believe what’s happening in front of them is true.
MG: Yeah, that’s true. Fake videos are circulating on WhatsApp, and people believe them to be real. Yeah, and people will see. But there are so many good things also. So many are there. For example, we are using ChatGPT right now. It’s… We have where you think people are using this. This are talking about this tutor. People are building personalized tutors for things. ChatGPT is one thing that is like it is, across genders, across ages. Now school kids have started using it.
AS: Yes, for projects, yeah.
MG: Yeah, for projects, everything. So I don’t know.
MG: My daughter keeps on using it. I don’t know whether to say or not to use it, or tell her to. It’s like in my generation, someone stopping me from using Google. “Don’t use Google” or “Yeah, this generation is like, it’s okay, we will use ChatGPT.”
AS: Hmm, yeah, that’s right. OK, so another thing is related to Gen AI music. How do you feel about rock music created by AI?
MG: It’s a mixed bag. Sometimes it’s good, but not so much. I find it intriguing when AI recreates songs from classic singers like Kishore Kumar or The Beatles. It’s not just about rock music; it applies to any genre. However, I worry that AI-generated music might dilute the original artists’ work. While the first few songs might be enjoyable, there’s a risk that subsequent ones might feel synthetic. It’s akin to the debate over synthetic foods versus natural ones. People might initially be curious about synthetic songs, including rock music, but ultimately, I believe they will gravitate towards authentic ones.
AS: People will always prefer the authentic ones in the end.
MG: Exactly.
AS: Okay. So finally, do you think there could be a Terminator-like outcome with Gen AI, and what precautions can be taken to prevent such a scenario?
MG: Terminator, you know, that movie was one of my all-time favourites from my childhood. It came out in ’84 when I was in 12th grade. I was a huge fan, and it felt like a dream come true to see those stories of machines dominating. Then there was “The Matrix,” where machines took over humanity. But honestly, those scenarios seem pretty far-fetched today.
AS: Hmm, I see what you mean.
MG: But you know, what I find more relatable to our current situation is the movie “Her.”
AS: Oh, okay. Tell me more
MG: It’s about emotional connection, you know? The protagonist falls for an AI he’s been talking to on a dating app, only to realize later that it’s not a real person. It’s kind of sad when you think about it.
AS: That’s unfortunate.
MG: Yeah, and then you realize that AI can talk to so many people simultaneously. I remember this one time I was chatting with someone, and they mentioned they were talking to about 3 million people! (laughs)
AS: (Laughs)
MG: It’s crazy, right? AI has this ability to manipulate human emotions, and that’s something we really need to consider carefully.
MG: Absolutely. AI has this remarkable ability to learn quickly and manipulate people’s emotions very effectively. So, you see, there are chatbots being developed specifically for that purpose. They might seem harmless at first glance, but they can have a significant impact on society. You don’t always need brute force like in the Terminator movies to bring down a society. Emotional manipulation can be just as devastating.
AS: That’s scary.
MG: Exactly. It’s a concerning scenario where AI companions become so close that human relationships start to lose their significance. Everyone gets absorbed in their AI interactions, neglecting real-life connections.
AS: Yeah, it’s indeed a frightening prospect.
MG: Imagine struggling to connect with a human after getting accustomed to conversing with AI companions.
AS: Yes, not even knowing who lives next door anymore?
MG: It’s quite concerning, isn’t it?
AS: Indeed.
MG: Back in the day, people used to engage in more conversations before the era of smartphones.
AS: Yes, exactly. Nowadays, everyone seems glued to their phones.
MG: And now, instead of conversing with each other, people are increasingly absorbed in their devices.
AS: It’s like a similar scenario to what we were just discussing.
MG: But if people start relying solely on AI for companionship, it’ll lead to a more isolated existence, with everyone essentially talking to themselves.
AS: Indeed, it’s a very scary scenario to contemplate. The Terminator analogy may seem extreme, but it underscores the fact that AI has the potential to exert control over humanity in subtle yet significant ways.
MG: Well, I hope I haven’t instilled too much fear during this interview, but it’s crucial to discuss these possibilities.
AS: No worries. It was very nice talking to you, Moyukh.
MG: Likewise, it’s been a stimulating conversation, delving into the realm of Gen AI and its transformative potential. It’s truly revolutionary, similar to the shifts brought about by the Industrial Revolution. Automation has transformed various segments of society, and now, with AI’s advancements, it’s poised to revolutionize the realm of knowledge and cognition.
MG: Let’s wait and see how it unfolds. There’s a mix of excitement and apprehension about what the future holds.
AS: Thank you so much for your time and insights.
MG: Thank you and have a good day.
AS: Goodbye. Take care and have a wonderful day.
