In a recent podcast, Imteaz speaks with Kaustav Bhattacharya, an AI and data science leader with over 20 years of experience, about the potential for generative AI to assist with and accelerate learning, particularly for neurodiverse populations. Kaustav emphasizes the importance of keeping up with the rapid pace of AI and data science change and the need for lifelong learning.
One of the challenges with generative AI is the potential for bias in training data sets. Kaustav acknowledges this issue and emphasizes the need for caution. However, he also believes that generative AI has significant potential to assist with learning, particularly for neurodiverse populations. For example, using platforms like ChatGPT to complete homework and assignments can accelerate education and research.
Kaustav also emphasizes the importance of cultural nuance in personalized experiences, particularly in areas like healthcare, where the impact of bias can be significant. He recommends the book “The Culture Map” by Erin Meyer, which provides a deep appreciation of global cultures and can be a helpful tool in creating culturally nuanced, less biased AI algorithms in the future.
The conversation also touches on personalized consumer experiences based on their data profiles. Kaustav emphasizes the need to consider cultural nuance when building customized experiences, particularly in areas like healthcare.
Throughout the podcast, Kaustav emphasizes the importance of lifelong learning and the need to keep up with the rapid pace of AI and data science change. He recommends keeping an open mind, staying curious, reaching out to experts, and reading widely.
In conclusion, the potential for generative AI to assist with learning, particularly for neurodiverse populations, is significant. However, the issue of bias in training data sets must be carefully considered and addressed. Cultural nuance is also an essential factor to consider when building personalized experiences. As Kaustav emphasizes, staying curious and learning is essential to keep up with the rapid pace of AI and data science change.
Guest’s social handles:
LinkedIn – Kaustav Bhattacharya
Twitter – @jupiterorbit
Hosted by: Imteaz Ahamed
Podcast Transcript: Imteaz and Kaustav
Imteaz: Hi everyone, welcome to Applied Intelligence. I’m Imtiaz Ahmed, your host. On Applied Intelligence, we go through some of my contacts and friends in terms of the artificial intelligence in the tech industry, and we learn from them and see all the things that they have to bring to your development and growth. So today, on my first episode actually, I have a dear friend, Kostav Bhattacharya. Kostav’s currently the CTO at Invica, part of the Havas Group. And I’ve had the pleasure of previously working with Kostab previously in my career back at Racket, and it’s wonderful to have him today. So welcome Kostab.
Kaustav: Thanks, Imteaz, it’s wonderful to be here. Thanks for inviting me to your inaugural episode.
Imteaz: Yeah, so, you know, we’re gonna learn as we go. This is the first time I’m actually hosting a podcast I’ve actually been on quite a few podcasts myself. I’ve never actually hosted one before so be easy on me anything else
Kaustav: No worries, let’s go with the flow.
Imteaz: Very cool. So in terms of getting to know you and for the audience to get to know who you are, what’s your story and how did you kind of get to where you are today?
Kaustav: Sure, so if you cast your mind back to the early 80s and a seven-year-old child sitting in front of a very early 8-bit microcomputer called the BBC Micro, that’s how I started my life in the career I’m in today. The real story is, back when I was seven… An uncle was coming to visit my family in the UK, here in London where I’m based. And that uncle was about to join his master’s programme in computer science. And coincidentally, the week that he was arriving in the UK, we invited him to stay with us for two weeks before he moved up to his university here, get acclimatised to the lovely British weather. He was coming from India to the UK so it was probably quite a shock for his system with the weather over here. Coincidentally in that first week he was with us my dad decided to buy me my first computer and it was one of these early 8-bit micros where you switch it on and it goes beep and a blank screen appears with a cursor flashing and that’s it. You kind of got to flick through the manual and figure it out. So I knew nothing about this stuff, but my uncle happened to be here. He’d never heard of a BBC microcomputer before, but he flicked through the manual. Within a couple of days, he figured out how to program in a language called BBC Basic, and he actually programmed Space Invaders for me in a couple of days. And you know what? That blew me away. And ultimately for me, that uncle was my role model when I was seven. and it set my entire future career from the age of seven onwards. It inspired me to get into software development. It led me down the path of launching in my early teenage years an online, well it wasn’t even online, it was a magazine, it was a computer magazine, an enthusiast magazine that we launched on We used to have these things called 3.5-inch floppy disks that you’d stick into your computer and you’d store and save data to and transfer between desktops and laptops. And I created this magazine.
Imteaz: Just on three and a half inch floppy disk. My daughter thought I 3D printed the save icon when I showed her a floppy disk. No context of what physical storage is anymore.
Kaustav: Exactly. So, you know, it led me to launching that magazine on floppy disk format, and then later on, it became an online website in the early days of the web, and so on and so forth. And it led me into my first career out of university in software development. And so, you know, for me, it’s so important to have role models early on in your career. and that’s the kind of thing I like to do with my kids as well. I encourage with my friends and my family if they have young children to really just set them up for life with really strong role models because it worked for me and hopefully it works for others. I think you know once I got into my career I followed quite a traditional path in software development. Although my background is in artificial intelligence and computer science academically back back in the late 90s, early 2000s, the AI scene wasn’t really anywhere. And so I went into web dev in the good old days in the wild west of the web and loved it. Went onto mobile development, continued the web development and progressed in that trajectory. And I think at some point along that road, I felt like I was reaching a technical glass ceiling in… in the type of work that I was doing and the seniority and the change that I could affect. And so that led me to look more broadly around how do I become a more rounded technologist. And that encouraged me to pursue training and softer skills, MBA like training and learn about things like corporate finance, about change management, about team motivation, team building. and all of those things that actually make up a much more holistic whole person and not just be a really good software developer. And for me that was learning a toolbox, you know, various tricks in the toolbox that can help me drive a conversation or drive a strong negotiation and convince a client that they should be going down a certain path, learning the art of presentation and standing on stage and having a conversation, the kind of conversation that we’re having right now. right now. And so you know that was a pivotal point in my career. And I guess the two other areas that are really important to me is as a technologist it’s so important to continue to learn. You know lifelong learning has been a mantra for me and will continue to be till the day I retire and probably beyond. And you know without learning and staying at the forefront and not just reading about it, but actually being a user of it, learning it and figuring out how it works. If I didn’t do that, I couldn’t be honest with myself and I couldn’t command the respect of the teams that I manage as well. And that sort of leads me to the final point about where I’ve got to in my career, which is the whole notion of the art of the possible and not getting bogged down in dogma. I’m a very big advocate of testing and learning. just because something has been done in a certain way for years or decades doesn’t necessarily mean you have to continue doing it that way. I know there’s the old phrase, don’t fix it if it ain’t broke, but there’s so much new technology coming on board every single month, every single year, every single week now. We’ve seen that happening with generative AI that I think is really important to continue to be experimental and figure out the art of the possible on a continual basis.
Imteaz: Very, very cool. Going back to the beginning of your story, I started my journey with a 386. I’m a little bit younger than you, Constance. I remember my dad at the time in Australia dumping $3,000 on a 386 back in the day, and I was just obsessed with it, right? Like, learning the basic prompts of just printing stuff, Word docs, et cetera. What was MS, no, it was called Word Perfect at the time, I think. It just takes you back to what computers used to be and what they are now. And then even in terms of how I found myself working in the e-commerce, digital marketing, and tech space now, it’s all happened because of happenstance and circumstance and being at the right place at the right time more than being technically qualified to do all of those things. So, you know, coming back to your point on lifelong learning. The ability to unlearn right now, I think is so pivotal and important because just because you’ve always done something in a particular way doesn’t mean that you have to continue that way. Like on your point about if it ain’t broke, don’t fix it. Yeah, lots of things work right now, but it’s not necessarily the most efficient way to do things anymore. And it’s not necessarily. the fastest way to do things anymore. So, especially with generative AI, especially when it comes to things like creative production, content production, any business process that has any form of static process attached to it. Even if you have a profitable business model now, it is an opportunity to… re-evaluate all of this stuff and see what you really want to focus on more than anything else. Which is a great segue to my first question to you, which is, what are the pivotal moments that you think in the last three to five years that have enabled generative AI to get where it is today?
Kaustav: Yeah, gosh, where do I start with this? It’s such an amazingly fast-moving area. I think the first thing to say is that I think, especially in the last year and a bit, the availability of generative AI tools in a democratized way has really changed the equation. Before this era, you had to have specialist knowledge. In fact, hell, you had to have a PhD in AI to figure out a lot of this stuff. And companies were paying crazy, silly money to hire these specialists who could help them figure out how to move forward with AI in general, and more recently with generative AI. But with the advent of platforms like StableDiffusion, Dali, Mid Journey, ChatGPT, a lot of this technology has been brought into the public domain through interfaces that are dead simple and easy for anybody and everybody to understand and use. So if you’re an artist or a part of an art collective and you wanted to use Dali or Mid Journey to generate AI-based imagery, you can do it literally through a Discord prompt or some other kind of interface and start to figure out the nuances of how these platforms work by adjusting your prompting. and getting different outputs and so on and learning the subtle art of prompt engineering as we call it today. So I think that’s been a really amazing enabler to get, I don’t know whether I’d call it the masses yet, but certainly a greater degree of the public mindset kind of focus towards the usage of these tools and what they can empower. I think the other area is kind of similar to this, is alongside the democratization of generative AI where anybody could use it, for developers as well, the software engineers who have been working in this domain for years, again, you needed a lot of specialist knowledge to make sense out of the software and build product and solutions with it. But now we’re starting to see The next level of abstraction where developers with non-specific generative AI experience. So take your typical web developer or full stack developer who knows front end really well or who knows back end software server side development really well. A lot of these people are now able to create tangible solutions out of generative AI libraries and modules. that they can stitch together and really hyper-accelerate their development. And in fact, using things like large language models, they’re now able to really accelerate and empower themselves to create solutions based on the back off of generative AI. I think, you know, those are two of the key areas that have really transformed the space in the last couple of years.
Imteaz: Let’s just dig deeper a little bit on developers specifically using generative AI. Can you give the audience a bit of background in terms of things like copilot and the efficiency that it’s unlocking for developers in terms of building out stuff much faster and having something to aid them to do that?
Kaustav: Yeah, so let’s talk about Copilot. And I think there’s a couple of other really exciting areas as well, other than that. But starting off with Copilot, so Copilot’s been quite revolutionary in the way developers are able to accelerate their path to completing a project or starting a project. Now, most developers use a integrated development environment or an IDE, which is their software, it’s the interface write code in. And many of these IDE’s have had the ability to complete a single line of code based upon what you’re typing. And that’s nothing particularly new. It’s been around for a while. But what Copilot has really allowed developers to do is not only complete a single line but actually complete entire blocks of code literally within seconds. And so using the training set that Copilot is feeding off of, it’s been fed with millions and millions of examples of similar code that other developers have done, as this has been sourced from various different places on the internet. And using this knowledge and training, the algorithm is able to figure out that maybe you’re trying to write a block of code that would validate the content of a form. and therefore not only give you the regular expression to check whether the email is in the right format but actually write the code for that entire form from the name to the email address to the address to the telephone number, etc., and just output that block of code for you. And then more broadly, as you start to build up a larger file of code or multiple files of code, Copilot can start to look across the interconnections and the association between code blocks and start to suggest ways of tying it together. I mean I’ve been playing around with it for a couple of weeks and I’ve really found it amazingly useful in my workflow and I’ve been using it as a way to get my own team excited about Copilot and starting to experiment and using it. And yes, there are some potential challenges around the ethics of how it’s been trained and are you potentially ripping off somebody else’s work, and whether it’s accurate or always right. I think all of that needs to be taken into account. And my advice to the developers I work with is don’t blindly take the output as production-ready code. use it as an accelerator to get you from 0 to 20% quickly, and then use your own knowledge and experience to properly analyze the code and tweak it as and when necessary. Never ever take generated suggested code from copilots or any type of code completion software, literally. I think the other area outside of copilot, that’s super interesting, is some of the emerging space, literally within the last couple of months, where libraries are coming out and orchestration platforms are coming out that are allowing software developers, not necessarily AI software developers, but just generalist software developers, to chain together different processes that trigger something that triggers another thing that triggers the final part of the journey in an autonomous way. So a really good example of that is a rapidly growing library called Langchain. We’ve been using Langchain where I work recently over the last couple of months to great effect and we can do some really amazing stuff now. Perhaps I can share one example of a project that we’re working on.
Imteaz: Yeah, let’s go.
Kaustav: We were trying to figure out… So, the industry has been talking about chatbots now for what seems like over a decade. And chatbots have been, to a greater or lesser extent, semi-successful, but quite often quite miserably bad. You often create a lot more frustration than good out of most chatbots. There’s a few exceptions to that. And… What we’re really looking at is not chat bots, but chat UX, so to speak. Whether it’s a bot interaction on WhatsApp or on a website, or whether it’s an FAQ or a knowledge base that you’re querying or a simple search, a free tech search, how could we make that better? And we started with our own knowledge base. So at Inviqa, we have our career progression framework. that’s publicly published as a URL. Anybody in the world can see how we do it. Privately within our sort of virtual firewall, we also have our developer tools and ways of working and processes. That’s kind of unique to us in the way we do things. And so we keep that within our private GitHub repository. And so I started thinking, right, wouldn’t it be amazing if a new starter who joined Invica could come along and after their onboarding, just go to a search page and just literally type in anything they want about career progression or the tools that we use or the processes that we use. So for example, as a rookie software developer, six months out of university, you’ve just joined Invica, you’re a couple of months into your role. and you’re loving it and now you’re starting to wonder, yeah so in my first 90 days my boss has set me some goals and objectives so how do I really double down on progressing my career and learning what I need to learn, asking the questions I need to ask, meeting the people I need to meet in order to progress and get that promotion next year. Imagine if you could type We’ve actually managed to do that. And what we’ve done is right now, because our career progression framework is public information, there’s nothing private or proprietary about it. We’ve trained OpenAI’s, well, we haven’t actually trained OpenAI. We’ve pointed and we’ve constrained OpenAI’s large language model on our career progression ladder. And we’ve discovered a couple of things which is really interesting. Firstly, one of the… challenges in the industry is what we call hallucination. Quite often these large language models will just make stuff up if it doesn’t know what the actual answer is. We’ve figured out a way of actually getting the LLM to say I don’t know as the answer if it doesn’t know. So as an example, you could ask the Q&A chatbot that we’ve created how do I progress from a junior developer to a senior developer. and it will give you a really clear answer. But then the next question you could ask is, how do I make pizza? The answer literally will be, I don’t know, because that’s got nothing to do with career progression, and it’s got nothing to do with our tools and our processes. Now, OpenAI’s large language model, ChatGPT, does know the answer to how do you make pizza, but we’ve figured out a way of constraining its scope to the subject matter that we’re focusing it on. So we’re starting to apply this proof of concept that we’ve built on multiple different scenarios for things like employee enablement, increasing the enjoyment and engagement of employees within a company, or applying it to potentially some of our clients’ data and figuring out how to help clients use large language models to sift through massive amounts of data and do useful things with it. So it might be figuring out financial projections for your company or analyzing your press releases and providing an interface for the for the press media to extract the right type of information about your company or your services or your products. And that’s a difficult thing because I think one of the things that companies are really worried about right now is privacy. And I think privacy is not a solved problem. And there are various solutions emerging. So having private large language models within your firewall. So there is a platform called GPT4ALL. And we’re building POCs with GPT4ALL right now. And it’s great because you’re literally running a large language model within the virtual firewall of your company. that it’s more secure than sending your private data and proprietary data up to the big God mind in the cloud. So it’s early days and it has issues and problems. It’s not as sophisticated as the larger large language models. It’s also quite resource intensive. So one of our developers had it up and running on his laptop and hey, guess what? It took nearly 30 seconds to get an answer to a question. I’m not surprised. Most of these large language models are running on huge server farms up in the cloud. So there’s a few problems to crack there, but the fact that you can do this even is pretty amazing.
Imteaz: And I think over time, this will improve. Like just coming back to the chatbots, you know, when a company has a customer service chatbot and that chatbot takes anything more than three seconds to respond, I get upset. I’m like, you’re a chatbot. Like you should be, you know, they try and pretend that, you know, it’s actually writing something by having the dot, dot, dots. But then when it takes too long, and then the handoff to a physical human takes even longer. you know, from a customer experience point of view, that’s not ideal. So if we can have specific LLMs for specific use cases that work instantaneously as, you know, current consumers want everything instantly, that’s gonna be an amazing customer experience. But also, you know, from an internal point of view, having like a career coach or having very specific use cases sitting inside of your own firewall. that answers those specific questions for internal use cases so quickly. That’s insane. Right.
Kaustav: Yeah, absolutely. And I think, you know, chat UX, as I like to call it, rather than chat bot, is a large part of the solution, which is, as you were saying, you know, some of the times we’re trying to simulate a real-life human interaction and those dots are bouncing up and down to give an impression that somebody is responding from the other end. And then when you do get the answer, you’re like, oh, it’s a bot. But I think when you look at the human-like qualities of the… structure of the English sentence that’s being returned by these large language models, they are damn good, right, in most cases. And I think that builds a greater level of trust, potentially, with the end user. I think, you know, ultimately, when I think back to, what was it, 2005 or 2006 when Amazon entered the sort of the cloud marketplace and launched AWS, and I was a super early adopter of AWS. I remember spinning up EC2 clusters in like 2006 or 2007. And you know back in those days most companies were like, oh holy shit we’re not going to put our stuff into the public cloud. You know like that’s never going to happen. And some industries stayed with that notion for a long long time. So financial services for example and other highly regulated areas and they had very good reasons for that. But now fast forward to 2023 and I would say 80 to 90 percent, even more probably, have most of the infrastructure in the public cloud and where necessary in virtual private clouds. So a bit like cloud hosting and the mind shift that happened with cloud hosting, I think we’ll see probably a similar thing playing out with generative AI. And we still got to figure this out, right? Legislation around the world under different jurisdictions are still being figured out. So it’s not just legislation but is something that’s a fast-moving space.
Imteaz: Very cool. I’ve got two questions and let’s see which one’s the better one to go with first. One is, for a non-technical or non-developer business professional, what’s the best way to get started in generative AI? What’s the best way to learn the proper or what’s the best way to get started in terms of what are the best prompts to use for specific use cases? And then the second question is more around the education space. and the impact that it’s going to have or is already having on primary, secondary, and further education fields, right? Like what is, how do we ensure kids and students learn properly and learn the how and not just the, learn the what and the how and learn that critical thinking that is so important, you know, for future careers, rather than just, you know, learn how to prompt the… properly in copy-paste answers. So, you know, I think those two questions are kind of related in terms of, if you’re in your teens right now at school, what do you do in this space? Versus if you’re, you know, well established in your career, how do you kind of have a reset and think about how do I apply all of this stuff to my day-to-day?
Kaustav: Yeah, for sure. I think from a sort of non-technical business perspective, there’s a lot of ways you could learn. I’ll just share with you what I’ve been observing and what I’ve been doing at the company I’m working for right now. And I think there’s no substitute from doing, from playing. And I think going back to what I said originally at the top of our conversation, A lot of the generative AI tools are so democratized, so easy to access. Some of them are still free. Some of them you need to put a credit card behind. But ultimately, anybody, literally anybody could start up a premium chat GPT account and start experimenting with prompt engineering and using it and figuring it out. And in fact, as a prime example of that, some of our internal departments are now putting it to good use. So in terms of our marketing department and in terms of our outreach program, we are using various forms of generative AI to do research for us, for example, to go and research a client or a topic. So for a client, for example, we have now automated the process of researching the three, the top three burning things that our potential target vertical industries might be worried about today or problems they’re trying to solve and then automate the association between those three topic areas to the people and the stakeholders that are concerned with those things and then extrapolate the specific name of the person and the contact details through That’s a thing that we’re doing right now with generative AI, and it’s incredibly powerful for non-technical business focused people to be able to do something like that. The other thing is, there’s a lot of really great learning material coming out from various leading institutions now. Whilst on the surface, some of them might sound a bit scary to non-technical people. I would really encourage people from like say, project management, client services, other type of departments, you know, product people to go and look at some of this really valuable material. One really great example of that is what Andrew Ung is doing now with Deep Learning, his company called Deep Learning. You can go to the website that he’s running called deeplearning.ai and there is a series of short courses that he’s launched around how to do prompt engineering really well. once you’ve mastered that, how do you start creating product using prompt engineering and beyond into various different subject areas. And Andrew’s courses I’ve personally found really accessible and whilst they are somewhat technical in nature, in fact some of the courses are very technical in nature, some of the earlier courses around prompt engineering and how do you create products and services out of it, you as a non-technical person, as a business could get a lot out of that. In fact, that’s what spurred a lot of our more business-focused departments to start experimenting and get savvy and more comfortable with this technology. So, yeah.
Imteaz: I think when you speak about Andrew Ng, this weekend I did exactly what you just said. I had a look over his deep learning courses, the introductory ones at least, and yeah, very accessible for a non-technical person to get started in this space.
Kaustav: Yeah, this is true. And there’s undoubtedly many, many more areas that you could draw inspiration from. But getting to your other question around the use of generative AI in education, I think, personally, this is a really passionate subject of mine. I’m a father of two young children. One is a teenager, one is a pre-teen, and they are both very aware of generative AI and platforms like ChatGPT. Their school is even more aware of it to the point that they’ve banned all access. And I believe if they were to even type in the URL, it just shuts the browser down. And I don’t think that’s unique to my children’s school. I think a lot of schools are doing that to first figure out themselves, the implication of this kind of technology and what it means. and how it could potentially be harnessed. And personally, I feel that platforms like ChatGPT that are used by, are already being used by children, teenagers, or even younger to complete homework and assignments is a real boon, just like it is for a software development expert, like a software engineer. It is a way to accelerate what you learn anyway in school, which is how to do research properly, how to cite the source of your research so that you’re not blatantly plagiarising. And I don’t see a problem with doing preliminary early-stage research to try and find an answer to an assignment question. If the student is taught that this is just a starter for 10, it’s a way to get you to an answer quicker, but then after you’ve got the answer, you still need to refer back to what you know and the subject matter that you’re particularly focusing on and then build around the answer. You can’t just copy and paste and by the way, you can’t you haven’t been able to just copy and paste anyway since education was a thing right? You
Kaustav: know if you’re writing your university thesis You’ll be failed if you just blatantly cut and paste from anywhere years and decades before chat GPT even existed. You need proper methodology to go and learn and research and store your notes and process and use that information to deploy your knowledge to the answer to the question that’s being asked. either an exam or an essay or a paper or a research piece, whatever it might be. So I think this is where schools are grasping with how chat GPT in particular and other similar technologies are already starting to affect education and we’ve all seen those memes going around on the internet where some parent or teacher has taken picture of an essay where it says, I’m sorry I’m only a large language model and I don’t have an opinion about this and the kid has literally copied
Imteaz: Yes. No, but like, I think back to my schooling, you know, when I was in primary school, in Carter was just, you know, basically starting out and before on Carter, we’d have to literally reference books, right. And then, you know, later on in high school Wikipedia launches and the teachers are like, Oh, don’t reference Wikipedia, because it’s all made up stuff. And now everyone looks at Wikipedia as one of the first things, what did they do at research, right? So I think it’s the natural evolution of information becoming way more accessible, way more targeted, way more nuanced to the specific use case that you’re looking for. It’s a matter of as educators, as people teaching other people, how do you leverage these tools appropriately and still learn to critically think about whatever problem you’re trying to solve. or whichever area that you’re specializing in, because ultimately the machines will only do what you tell them to do. Yes, they’re getting smart enough to kind of resolve and solve problems for themselves, but ultimately we need to take our own responsibility for our own learning. And I think that’s the biggest thing that we can teach to children, people, et cetera, in general, because otherwise, you know, if we rely on machines to do everything, they don’t necessarily have all of the interests that we want to maintain as humans.
Kaustav: I think the other area that’s super interesting is in the area of neurodiversity and education. There is such a broad spectrum whether you have dyslexia or dyspraxia or other form of neurodiversity. I think there’s huge potential for generative AI technology to really enable people with very diverse ways of thinking and learning to actually get over some of the blockers that they have in their learning. So, you know, using chat GPT, for example, to ask for an explanation in an alternative way from the perspective of a different character or person or job role or function or whatever it might be. And I think that’s a potential area of research that needs to be done to see how this can be really assistive to a neurodiverse population of students, for example.
Imteaz: building on that I went to Turkey recently and in one of the old Ottoman schools they’re called madrasas they had a statement on the wall that said here no bird will be taught how to swim and no fish will be taught how to fly so you know taking on your point about neurodiversity I think using generative AI within the education space to create bespoke learning agendas for people that have difficulties or very specific learning needs, you know, with constrained resources from a teacher point of view and the number of, you know, teachers to students. Using generative AI to customize learning plans for people that don’t have, you know, all the cognitive ability that their cohort might have would be an amazing use. of unlocking potential of people that don’t necessarily have
Imteaz: Or don’t necessarily get the amount of attention that they need from their teachers as well.
Kaustav: Yeah, totally agree. The potential is huge. There’s a lot of research to do.
Imteaz: Yeah, and I saw a TED talk by Sal Khan as well where he’s building, I can’t remember the name of it, but basically an AI tutor tailored towards your kids as well through Khan Academy. I think it’s called Khan Amigo. But similarly, you know, every child, every person has a different learning experience and journey. And I think we should leverage AI to kind of build that customized journey rather than just one size fits all, which is very expensive to do, but not anymore.
Kaustav: Yeah. Totally.
Imteaz: Okay, to tie all of this together. And one question that I love asking my friends and asking people, in general, is, you know, what do they read that really inspires them and motivates them from a book point of view? And specifically, what type of books or what book, in general, do you gift to your friends? And yeah, tell me.
Kaustav: Yeah, well I could talk about this for hours but I won’t. I’ll give you maybe one or two examples. So the first book that comes to mind and I very rarely read an entire book in one setting but I read this book in one setting because it captured me so much and I happened to be on an 11-hour flight which also helped in reading that entire book in one setting
Kaustav: But truly it was super interesting. So the book is called The Culture Map, decoding how people think, lead and get things done across cultures, by an author called Erin Mayer. I don’t know if you’ve heard of that book, but I found it fascinating. It’s a relatively recent book, I think it came out in 2016 roughly, and I read it I think back in 2018. And I just found it so inspirational. It helped me open my eyes to the amazing diversity. that culture is and the mindsets of people from different cultures and the way they think or don’t think, the way that words and body language come across the divide of different cultures and it inspired me and encouraged me to read the entire book in one setting because literally every page, every chapter I was reading, I was learning something new and I was having… aha moments and I was scribbling and writing down notes in the margins and folding the corners of the pages where I felt that I could use the knowledge from that book in my day-to-day work or even day-to-day life and I refer back to that book constantly even to this day, you know, five or six years later having read the book. I find that a fascinating book so I would highly recommend that to anybody and that’s got nothing much in particular to do with AI per se. But in a in a tangential way, it kind of does, because it shows that cultures are so diverse around the world. And the way a lot of the generative AI technology today is developing, and some of the precursor to generative AI has been developing over the years, has been steeped in a lot of bias in terms of the training data set. the people who have created it and their backgrounds. And even to this day, I read recently, we saw sort of 10 years back how certain image recognition algorithms couldn’t distinguish the difference between gorillas and people of a certain skin tone and background. And so they put a filter, a hard filter to stop Google Photos or Apple’s equivalent to tag those kind of pictures. Fast forward 10 years, according to a recent article that I read in the FT, they still haven’t solved that problem and some of those manual hard filters are still in place because they haven’t managed to weed out what is perhaps the unconscious bias that’s built into these algorithms. from that book’s perspective, getting a real deep appreciation of global cultures is so important and one of the many factors that goes into creating more culturally nuanced, less biased AI algorithms in the future. Yeah, so that’s one of the books that comes to mind. The other books that I’ve found inspirational recently are more to do, again actually surprisingly, with culture. memoirs of an author called Satnam Sanghera. He is a British journalist and author and has recently published a book called Empire Land and Empire Land is about the modern-day lasting effects of the British Empire on British society and before he released that book he published his own memoirs about his family and dealing with schizophrenia in his family and the typical story of a Punjabi family coming over from the Punjab to the UK and trying to deal with all of the nuances of society, of foreign land, racism and then on top of that dealing with schizophrenia and the family and so on. It was a fascinating read. And the other book I’m looking forward to by the same author, Satnam Sanghera, is I think imminently about to come out and again it’s around the theme of the effect of empire on the world and in particular British society but from the perspective of a child. There’s lots of literature out there that’s adult-friendly that talks about the effect of the British Empire or other empires through history. on current day culture, but there are very few books that are aimed at children and that’s a huge area that is often missed out of education completely, at least here in the UK, I can’t speak of any other country, but you know kids here learn about World War II, but they really, really learn about the immense sacrifices that African nations and Asian nations and India and other countries the Second World War effort. And there are a myriad other things I could talk about related to that. So those are some of the books that I’m fascinated by and really interested with recently.
Imteaz: Coming back to your point on the culture map and the impact and influence that it could have on generative AI, I think when I look into my crystal wall in the future, there’s an opportunity for creating highly personalized experiences for each and every consumer based on the data that they share back to big tech, right? So whether that’s… My Apple health data, my email data from Google, all of my business-related stuff from Microsoft, my health tracking stuff with Fitbit, my home, smart home stuff with Nest, etc. All of this builds an enormously rich data profile. and my travel history through Star Alliance and OneWorld and all of this stuff, right? It builds an incredibly rich profile of who I am as a person based on all of the data signals that I’ve left throughout all of these experiences. In terms of connecting all of those dots and personalizing marketing communications, personalizing government interactions that I have with government agencies, personalizing healthcare interactions that I have with the pharmacy that I go to or the doctor that I go to. How do you see all of that piecing together? And being
Imteaz: hyper-personalized based on my culture and the data that I’ve decided to share with said companies.
Kaustav: Yeah, I think it’s such an amazing question that, and you could write an entire documentary about that question in itself, but I think one of the watch-outs I have is when you look at some of the recent things that have come out in the news over the last 10 years, take medicine, for example, and the bias towards using male subjects as your primary subject to research. and the effect that had on, for example, pain control using analgesics or other types of pain control. Recent research in the last few years has shown actually that the painkillers that we have used over the counter or through prescription from your doctor actually work in very, very different ways on the male body and the female body and for decades and decades. hardly any research was done on pain control in female body. And I think that’s one example of inherent bias that’s built into data sets. Now take the medical example even further. So things like fitness data, we both come from a cultural background where diabetes, for example, is prevalent in some Asian cultures. And… it would be interesting to see how fitness data is used in a much more culturally nuanced way so that if the product has been predominantly created by an American team in Silicon Valley by certain people from a certain class and background and earning, cultural background, skin colour, religion and so on, are they creating those products with a more worldly view? and a more nuanced view on a diverse data set to train these solutions on. Because if they’re not, then we’ll just continue to perpetuate these unconscious biases and problems. So I think that’s where culture is such an important facet of the greater whole of these solutions that we’re talking about right now.
Imteaz: I saw a funny TikTok of a guy who was summarizing 2023 nutrition tips from TikTok. And he was basically going through different people giving different advice. So it started from being, you should be keto to you should be vegan to you should be, you should do intermittent fasting. And then, and he was basically trying each one of them. and towards the end of it, all he could literally eat was ice. So I think the nuance of, you know, what might work for a certain person based on their genetics, based on their cultural upbringing, based on the foods that they’ve eaten their entire life, could be, you know, further influenced by, you know, the data that’s available for them. So for example, you know, I personally think one of the reasons, and I’m sure there’s empirical data to prove what I’m going to say, but I’m sure that gluten as a thing being introduced to the subcontinent was more recent over the last 100 or so years through the British colonization. I think that’s something that wasn’t necessarily in the subcontinental diet prior to 100 years ago. And I think the overconsumption of gluten and sugar has probably prompted the… diabetes epidemic that’s happening across the subcontinent and people from the subcontinent that have gone on to live in Western countries right now. So I personally am trying to deal with my I have diabetes, young-onset diabetes. I’m trying to deal with that through trying to stay away from number one gluten as well as sugar. But if I followed all of the advice that comes from doctors… slash TikTok slash Google, then it would be more around, just control your sugar intake and don’t necessarily worry about everything else. But I found eating rice doesn’t impact my blood sugar as much as eating bread does at the same caloric intake as well.
Imteaz: So, having the advice tailored to cultural nuance using AI or using doctors that know what they’re doing and have that nuance available to them is super cool. I think as it becomes easier, it’s going to be better for health outcomes globally.
Kaustav: I totally agree. I think there’s a great potential with this technology to actually advise and direct people in the right direction rather than just blindly following memes or social media. And obviously, there’s no substitution to talking to a true expert, your physician or your general practitioner or a specialist around health matters. But the reality is, we all like to self-prescribe and self-diagnose. And that’s nothing new that’s been happening for decades. But the challenge that that poses is being magnified tenfold by the advent of things like generative AI and adjacent technologies that are seemingly turning us all into medical experts or experts in whatever field we care to be in. So yeah, I think that the potential is there. It needs to, you know, it’s multiple layers. of regulation, legislation, getting the biases out of the system as much as we can. And it’s about education, it’s about empowering people to know how to use these tools properly for the right purpose. And that’s very easy for us to sit here and debate it, but I’m sure it’s one hell of a hard problem to solve that will take many, many years to get right.
Imteaz: Very cool. Karstaf, this has been a wonderful chat. Before we wrap up, is there anything else you wanna share in terms of your story?
Kaustav: Um. Oh no, I wasn’t prepared for that question. Um, yeah, I mean, I guess… I guess the key thing is keep learning. It’s always stood me in good stead. And like I said before, I’m a lifelong learner and we all learn in different ways. But I think the key important thing is never stop learning. No matter what, no matter what profession you’re in, no matter what you do, whether it’s professional or it’s a physical job or it’s a knowledge economy job. whatever it may be that you’re doing as your career, maybe you’re running your own company and you’re an entrepreneur, never stop learning because the day you stop learning is the day that you get out-competed by the nearest person next to you who is learning when you’re not. And for me, that’s been a mantra all throughout my career.
Imteaz: super cool, and make the time to learn. Especially when people are so busy, when you prioritize the doing over learning all the time and you don’t take that time for yourself, it’s very hard to keep ahead.
Kaustav: Yep, for sure.
Imteaz: How can people reach out to you if they wanna learn more about you or learn more about AI, et cetera?
Kaustav: Yeah, for sure. So hit me up on LinkedIn, cost of Bhattacharya. And I’ll be happy to respond to people there. Or ping me on Twitter. Yes, I still do use Twitter. My Twitter ID is at Jupiter Orbit, and I’ll be happy to have an open or a DM conversation on there. Either way, yeah, reach out to me. I’d be happy to continue the conversation.
Imteaz: Super cool. So all of your book recommendations we’ll put in the show notes as well. Thank you so much for sharing your time, your knowledge, your experience with us today. And I look forward to catching up one day again when you’re over here in the US or I’m over in the UK.
Kaustav: Amazing MTLs, great catching up, and thanks for having me on your show. Speak soon
Imteaz: Thanks a lot. Take care, bye.