Podcast: Play in new window | Download
Subscribe: Spotify | RSS | More
Amarjot Singh talks with Jason Barnard about building adaptive AI.
Amarjot Singh, founder and CEO of Skylark Labs, talks with Jason Barnard about how machines can be taught to learn like humans, the complexities of building true autonomy, and why long-term memory systems are critical to progress.
Amarjot Singh shares his journey with Jason Barnard from academic research to launching a commercial AI company in the defense sector. He shares the realities of navigating long sales cycles, hiring for highly specialized roles, and creating a mission-driven team culture.
They also touch on the continuous evolution of AI memory systems and the strategic partnerships needed to thrive in defense innovation. Amarjot Singh envisions a future just 5 to 10 years away — one where digital beings don’t just exist, but learn, adapt, and coexist alongside humans.
What you’ll learn from Amarjot Singh
- 00:00 Amarjot Singh and Jason Barnard
- 01:48 What is the Gemini Result for Amarjot Singh, According to Jason Barnard?
- 02:09 Why is the Deep Research That Jason Barnard Made for Amarjot Singh So Long?
- 03:08 What Happens When Jason Barnard Asks Gemini to Access Information Beyond Its Hippocampus Memory?
- 03:47 How Should You Feel About Google Using Academic Data to Shape Your Online Identity in AI?
- 04:32 How Should You Convince Investors to See You as Both Academically Credible and Profitable?
- 04:55 How Did Amarjot Singh Move From Academic Research to Building a Commercial AI Company for Defense?
- 05:07 Why is Cambridge a Great Place for Building Fundamental Technologies?
- 05:14 Why Should You Aim to Simplify How Machines Make Sense of the World Like the Brain and Children Do?
- 06:08 What is the Difference Between How AI and a Child React to Situations They Have Never Encountered?
- 06:22 What is Missing in Traditional AI Systems That Amarjot Singh is Trying to Address in His Company?
- 07:20 Why is Adaptive AI Particularly Valuable in Defense Systems?
- 07:57 How Does the Approach of Amarjot Singh to AI Differ From Simply Replicating the Human Brain?
- 08:04 What Learning Capabilities Are Being Replicated in AI to Make it More Efficient and Adaptive?
- 09:13 What is the Difference Between Short and Long Answers When it Comes to Retaining Information?
- 09:28 What Happens When You Add New Content to Your Website?
- 10:17 How Does Credibility Affect the Confidence in Information of a Machine?
- 10:43 Why is the Credibility of Information Important?
- 11:38 Why is Memory Considered Continuous Instead of Being Short-Term and Long-Term?
- 12:18 Why is Frequency and Recency Important When Encoding Information?
- 12:39 What Factors Influence How We Retain Knowledge?
- 13:11 How Does the Knowledge Graph Determine the Importance Between Pieces of Information?
- 14:43 How Do You Deal With the Long Sales Cycles in the Defense Industry?
- 16:06 How Do You Find People Smart and Skilled Enough to Handle the Highly Complex Work of Your Business?
- 17:01 What Impact Does Giving People the Freedom to Choose Their Work Have On Your Company?
- 18:16 How Do You Protect Your Intellectual Property When Working Across Different Countries?
- 19:43 How Much Has AI Development Been Influenced by the Brain?
- 20:58 How Fast Do You Think AI Technology Will Evolve in the Future?
This episode was recorded live on video May 13th 2025
Links to pieces of content relevant to this topic:
https://www.youtube.com/watch?v=xD6RR3Om8bc
https://www.theverge.com/2018/6/6/17433482/ai-automated-surveillance-drones-spot-violent-behavior-crowds
https://www.innovatorsunder35.com/the-list/amarjot-singh/
Amarjot Singh
Transcript from Amarjot Singh with Jason Barnard on Fastlane Founders And Legacy. Building Adaptive AI
[00:00:00] Amarjot Singh: We do a lot of difficult deployments. We go to the desert, we go into the sea. Sometime it’s fun, sometime it’s not. But you want to be able to have anchors you can rely on and keep on doing what you need to do when nothing is working. And that is, I think the only way to do that is if you are, what you’re doing is meaningful work.
[00:00:21] Jason Barnard: Right. And that’s a really good kind of point as well is when you’re doing something and it doesn’t appear to be working, how do you keep your team motivated to keep trying and keep finding new ways?
[00:00:32] Narrator: Fastlane Founders and Legacy with Jason Barnard. Each week, Jason sits down with successful entrepreneurs, CEOs and executives, and get them to share how they mastered the delicate balance between rapid growth and enduring success in the business world.
[00:00:48] Narrator: How can we quickly build a profitable business that stands at test of time and becomes their legacy? A legacy we’re proud of. Fastlane Founders and Legacy with Jason Barnard.
[00:01:01] Jason Barnard: Hi everybody and welcome to another Fastlane Founders. I’m Jason Barnard and I’m here with a quick hello and we’re good to go.
[00:01:09] Jason Barnard: Welcome to the show, Amarjot Singh.
[00:01:14] Amarjot Singh: Glad to have me, Jason..
[00:01:16] Jason Barnard: I literally just realized that your surname is perfect for that introductory song because I’m singing.
[00:01:26] Amarjot Singh: That’s great.
[00:01:26] Jason Barnard: And that’s not a very funny joke, but I had to make it because as I sing, I thought that’s good. Anyway, let’s get onto the topic, which is going to be about building Adaptive AI.
[00:01:37] Jason Barnard: And I just asked you, are you basically building AI that learns like humans replicating our brains? And you said something along those lines. So we’ll be getting into that in a moment. But before we go onto that, I wanted to show you. The Gemini result for your name, which is pretty good. I asked it to tell me about you.
[00:01:57] Jason Barnard: And it comes up with a reasonably good answer. I quite like that, and I think your academic background is what it picks up on, and then it’s moved on to your corporate background, which is now. But then I thought, let’s use deep research, which I’m really enjoying and I’ve made a video because it’s really long.
[00:02:15] Jason Barnard: It’s really long because it knows so much about you and there’s so much information about you. So the idea here is I asked deep research on Gemini to build a plan to tell me about you. And it built a plan and then it executed and you see that table, that’s a lot of information about you and your academic work.
[00:02:33] Jason Barnard: It goes through an entire history of your life, which I believe is pretty accurate with that second table there. It’s got seven sections all about your life as an academic, the work you’re doing, the work you’re doing with Skylark, and it literally doesn’t stop.
[00:02:49] Amarjot Singh: Am I crazy?
[00:02:50] Jason Barnard: This is something. Yeah. We all need to pay attention to and we all need to take care of because if I want to know about you, I can ask Gemini and it will tell me a quick overview like that.
[00:03:01] Jason Barnard: And it needs to get it from its called hippocampus in the brain.
[00:03:07] Amarjot Singh: Yep. Hippocampus.
[00:03:08] Jason Barnard: And then if I asked it to dig down and figure out more and go into its long-term memory that it doesn’t necessarily have in that hippocampus memory, it then does this and it’s a due diligence using AI, which I love. And at Kalicube, it’s exactly what we do for our clients. We make sure that it is accurate and convincing for your audience. And I think for you, it already is because of your academic career. Because Google and the other AI have got so much solid information about academia and they trust those academic institutions.
[00:03:45] Jason Barnard: How do you feel about that?
[00:03:47] Amarjot Singh: I feel like that’s great. I’m actually overwhelmed that once you showed me that table and the seven sections, that there is actually so much information out there about me. Now I wonder if all of that is accurate or if I wanted to tweak it in a certain way so that I come off as more appealing to the investors or other entrepreneurs or our clients. And that’s the thing I’m thinking about.
[00:04:13] Jason Barnard: That’s a great point because the whole thing about pivoting. You’ve pivot from academic research to building a commercial AI company is really, important because how you frame it across your entire digital ecosystem will greatly influence how the machine frames it.
[00:04:32] Jason Barnard: So as you said, if you want to convince investors, you need to convince them that your academic qualifications are there, but that you can actually make money for them. And it’s up to you to frame it. It’s up to you to spread that message so that the machines will repeat it. And that’s what we do at Kalicube.
[00:04:47] Jason Barnard: But I’m now interested in how did you move from academic research to building a commercial AI company, particularly in defense?
[00:04:54] Amarjot Singh: I did my PhD in the UK.
[00:05:00] Jason Barnard: Cambridge.
[00:05:01] Amarjot Singh: Yeah, Cambridge. I went to Cambridge.
[00:05:03] Jason Barnard: That came from deep research. Sorry.
[00:05:06] Amarjot Singh: No. That was great. I had a great experience and Cambridge is all about building fundamental technologies, and that’s what I did. I started the human brain. I was trying to figure out how the brain is able to make sense of the world. How kids make sense of the world, but they’re able to do it really quickly within seconds. While machine required a lot of data, it requires a lot of compute. So I was interested in simplifying how machines could do the same thing. And I was able to an extent, able to do that. But beyond that, I was trying to figure out what are the practical applications of this and the kind of things I wanted to do with that tech.
[00:05:46] Amarjot Singh: It’s hard to do in academia because it requires a lot of capital. It requires a lot of real world data and it requires a lot of real world testing. The simplest solution I had was to start a company so that I could do what I wanted to do.
[00:06:04] Jason Barnard: I did that too. I create the company so I could do what I wanted to do, but that makes me think the problem that AI has is reacting or being able to react appropriately to situations it has never seen before. Whereas, a child is able to do that. And the fundamental difference is what you’ve figured out.
[00:06:22] Amarjot Singh: Yeah. So AI traditionally learns from large data sets, large compute, while I’m able to tell you whatever I’m able to tell you right now, and you’re able to learn it within seconds. So our brain has the ability to do long-term learning and reasoning, and also short-term learning and reasoning.
[00:06:39] Amarjot Singh: And that is a thing which is missing in traditional AI systems, which we are trying to build in our company. And if you are able to do that, then you’ll be able to put this AI onto machines, which are able to go operate in the real world and learn if needed, leading to truly autonomous systems, which we don’t have today, because still we need to update them continuously. And that is the gap we are trying to bridge.
[00:07:06] Jason Barnard: Right, and in defense in particular, the situations the AI is seeing are more often new. And also require instantaneous decisions that need to be right.
[00:07:19] Amarjot Singh: Yeah. Yeah. Because in defense, if you have a system which is working in the battlefield or a system is on top of a drone, you cannot bring the machine back, update it and send it back.
[00:07:30] Amarjot Singh: You need to take decisions then and there, and that is where Adaptive AI will be valuable.
[00:07:38] Jason Barnard: Okay. So let’s come back to the brain idea. We all talk about AI as being some kind of replication of the human brain or an attempt to replicate the human brain. Is your particular approach much closer to that reality than everybody else’s and why?
[00:07:57] Amarjot Singh: I think I would say that we are trying to learn from the principles of the brain rather than trying to replicate everything. So functionally, we want to learn at different time steps. We want to be able to learn quickly. We want to be able to learn with very little compute, and that’s what we are trying to replicate.
[00:08:17] Amarjot Singh: And I think we have been able to do that to some extent because our AI is able to run on edge devices like CPUs, Raspberry PIs, things which are really low compute, really low power, and they’re able to learn new things instantaneously. And then obviously, they’re able to retain it over a period of time, like humans retain knowledge.
[00:08:38] Amarjot Singh: So if you’re able to fundamentally achieve those three functions, I would say that at least you’re able to receive the world like humans do and operate in that rather autonomously than having a human in the group.
[00:08:54] Jason Barnard: So when you talk about retaining information, would that be the difference between this very short answer?
[00:09:02] Jason Barnard: There’s very long research answer where it went out onto the web and found new information to supplement the incredibly short answer that it initially gave me.
[00:09:13] Amarjot Singh: I think the short answer is basically a summary of the long answer. Both of them are long-term knowledge, which basically it has pulled from whatever places where this long-term knowledge exists. But if I add something new right now onto my website, it may or may not appear right away, in this search. And that’ll you, as you mentioned, would depend on if that information is relevant. If it is relevant, then Google will crawl it and it’ll retain there for a period of time. And over time, if the credibility is there, it’ll permanently keep it, otherwise it’ll get rid of that.
[00:09:54] Amarjot Singh: And that’s what we try to do. The short term knowledge is looking at something relevant and then crawling it and bringing it in front of the user. And if the credibility of that source remains over a period of time, we will solidify it. We’ll keep it in a long term knowledge as of us get rid of that.
[00:10:17] Jason Barnard: I really like the idea of the credibility remains over a period of time because people often forget or don’t realize that the period of time that a machine sees a piece of information from a perceived credible source or multiple credible sources and becomes confident because of that credibility and because of the time, the more likely it is to retain it long term, and it’s that long term memory that we’re aiming to hit.
[00:10:43] Amarjot Singh: Yeah, that’s very important because information changes over a period of time. The credibility of information or the importance of information changes a period of time. And that is something which also these systems or the current AI systems don’t really tap into because whatever information they have learned is static and the world is changing.
[00:11:06] Amarjot Singh: So if you’re operating with the same tool in a changing world, you’re going to be left behind. And that’s where humans come into play, where they have to update the system. So it keeps in pace with the changing world and that’s what this short-term, long-term memory would fix.
[00:11:24] Jason Barnard: So what I’m now thinking as, you talk is the concept of long-term, short-term memory is incredibly important to the work that we do. Would there be a midterm memory or is that a silly idea?
[00:11:38] Amarjot Singh: No. So memory is continuous, it’s not short term or long term. We just say it short term, long term because it’s easy to program into machines, but memory is continuous.
[00:11:49] Amarjot Singh: So there are some concepts, some experience you had when you were a child. Maybe you remember the street you were playing on, or maybe you got hit by a bicycle and you remember that. And those are events which happen once in your lifetime, right? But you remember them and that’s long term. But maybe you spoke to someone an hour ago, and they were a person from a credit card company. They were supposed to just tell you something about your credit card. As soon as you put the phone down, you forget about that. So traditionally, we encode information, which is based on frequency. It’s frequent, it’s recent, and it has some sort of importance factor to it.
[00:12:27] Amarjot Singh: So we understand frequency and recency really well. If there is a, let’s say links to your website and you have a lot of links, there’s a lot of high frequency. So that’s a trusted source. And if it is some information has come recently, we can say that, okay, that’s more important. And then obviously there is some sort of importance factor that it’s Reuters or it’s TechCrunch, or it’s Google, so they’re like trusted sources.
[00:12:53] Amarjot Singh: But beyond that, there are other factors as well, which kind of control as to how we retain knowledge, which we don’t really know much about. Hence, the example to just give you.
[00:13:03] Jason Barnard: Sorry. Yeah, those are the three big ones.
[00:13:06] Amarjot Singh: Yeah, those are the three ones. Yeah.
[00:13:07] Jason Barnard: Frequency, recency, and importance. Because in Knowledge Graphs I talk about close, strong, and long relationships.
[00:13:17] Jason Barnard: Those are the ones that you retain or that’ll become the most influential. Would that be important?
[00:13:26] Amarjot Singh: So, short and long. We know it’s frequency and recency. How important works, we don’t really know. Why do we forget the name of a person who we just spoke to because they were just from a credit card company and we don’t really care about, or when you’re playing cards and you have a hand.
[00:13:41] Amarjot Singh: You put the cards down and you get the next hand and you forget about that hand right away, right? It’s a very frequent, it’s a very recent event and you play cards frequently, but why do you forget about that right away? So there is something going on in the importance factor, which tells you that’s not important.
[00:13:57] Amarjot Singh: But you do remember a memory from a childhood. It’s happened once. Maybe someone hit you, but you remember it because importance is really driving that. All the frequency and recency over there is really low.
[00:14:11] Jason Barnard: In fact, it’s perceived importance.
[00:14:14] Amarjot Singh: Depends on like how the brain perceives it. What is important for me can be very different for you.
[00:14:22] Jason Barnard: Alright. No. This is in my brain going off on different things. Let’s come back to business. How do you deal with the long sales cycles in the defense industry? I imagine they take decisions over multiple years. They’re big numbers, but you have to wait.
[00:14:38] Jason Barnard: Two years for them to actually make the decision. It goes up the hierarchy.
[00:14:43] Amarjot Singh: I think their ways of cutting them shorter, compressing the timelines. The best way to do this, when you’re a young company, you don’t want to go and grab the bull by the horns or by yourself. So you rely work with a trusted partner or a defense prime, and they already have contracts secured from the government. So you work with them and in the process you build your name and then after two years or so, you are in the process to go and get those contracts yourself. So in the process, you have learned how the process works, what kind of security clearances you need, how much money goes into doing all that stuff.
[00:15:18] Amarjot Singh: And that’s how we try to secure things. Unless there are certain provisions for small businesses where we get an advantage, that’s the only time we do things direct. Otherwise, we will partner with bigger people. And that also adds to the credibility because someone big is willing to work with us.
[00:15:39] Jason Barnard: Okay. that’s a really good answer and I’ve been doing it all wrong at Kalicube so that’s brilliant. When you’re building a team, obviously you need super specialized expertise, and it’s a problem we have at Kalicube for the production aspect of what we do, and it’s not obviously as specialized as what you are doing.
[00:15:56] Jason Barnard: How do you deal with that? How do you find the people who can actually do the kind of incredibly advanced brain work that you require?
[00:16:06] Amarjot Singh: I think people have to be. So in order for you to add smart people to your team, we have to figure out what they’re driven by. One is obviously they’re driven by how much money you give them.
[00:16:16] Amarjot Singh: And as a small business, we cannot compete with the likes of Google or Facebook or whoever that is. So the other thing which drives them is, the work they’re doing has a potential of making a great impact. If you’re able to weave that into the work they’re gonna do and they have actually seen you do impactful work, they would likely want to come and join and work with you rather than going and working with a bigger company.
[00:16:45] Amarjot Singh: The other thing which we really focus on is autonomy. We want people to do what they want to do, what they’re good at, and we give them a free hand and then they come up with either excellent ideas or they fail miserably. Both are okay, but until the point you don’t micromanage people, you make them in charge of what they want to do, and that has a potential of really making a big impact in the real world.
[00:17:09] Amarjot Singh: People would come and want to work for you.
[00:17:12] Jason Barnard: Okay. Fun, autonomy, or doing something you enjoy rather than fun. I talk about it as fun, but doing what you enjoy is usually important.
[00:17:21] Amarjot Singh: Yeah. If it’s meaningful, I think people will do it. Even though if it’s difficult and work is fun sometimes, especially for us and not all of the time, because we do a lot of difficult deployments.
[00:17:33] Amarjot Singh: We go to the desert, we go into the sea. Sometime it’s fun, sometime it’s not, but you want to be able to have anchors you can rely on and keep on doing what you need to do when nothing is working. That is, I think the only way to do that is if what you’re doing is meaningful work.
[00:17:50] Jason Barnard: And that’s a really good kind of point as well, is when you’re doing something and it doesn’t appear to be working, how do you keep your team motivated to keep trying and keep finding new ways?
[00:18:01] Amarjot Singh: Yeah.
[00:18:02] Jason Barnard: Okay. Next question is how do you protect your IP? Your intellectual property across countries working for the military? That’s a huge question. You must have a lot of IP, very difficult to protect.
[00:18:16] Amarjot Singh: Yeah, so I think one is that we have obviously patents, which we file, and that’s one way of doing it.
[00:18:24] Amarjot Singh: But it’s very hard to protect a patent once someone infringes it. You don’t want to go through a very long battle. So the best way to do that is compartmentalize your company. So people who are doing machine learning. They give encrypted codes to people who are doing production and goes back and forth.
[00:18:41] Amarjot Singh: So you want to be able to create as much silos as you can in your company and only share raw code if it is essential. And the other way of doing that is that people don’t really get to blame each other. They don’t really get to say that, Hey, this was someone’s code and I changed something because I thought it was a good thing to do, or a bad thing to do, which always creates chaos and friction between companies.
[00:19:02] Amarjot Singh: So if you can create silos, they’re encrypted and you can create firewalls, and if you’re able to build a product using that. I have found that to be the best way of, keeping, things secure.
[00:19:18] Jason Barnard: Brilliant. Next question then is you are using the brain as a way to improve AI rather than trying to duplicate or replicate the brain.
[00:19:32] Jason Barnard: To what extent have you succeeded. To what extent or where do you think it’s going? Where do you think you are going? We’re all going next. How fast is this going to evolve?
[00:19:43] Amarjot Singh: Yeah, I think that’s a really important question. I think the fundamental pillar, which we are building is the pillar of adaptability.
[00:19:51] Amarjot Singh: We want machines to adapt. If you are able to build that, which we have done for, certain applications, trying to journalize that you get to journal intelligence, you get to actually build intelligence, which works for most of the applications, have fixed knowledge and have the ability to adapt. Now, if you’re able to embody that into different kind of machines, be it humanoids, robotic dogs, drones, towers, which we do, you have embodied intelligence or embodied journal intelligence.
[00:20:26] Amarjot Singh: And if you actually have that, eventually you will have digital citizens, digital humans, digital beings, whatever you want to call it. Which would have the ability to live among ourself and operate and do most of the things which we do today, giving us the freedom to do what we actually want to do. And that’ll be a world where humans and digital beings will coexist, and that is where it’s headed.
[00:20:54] Jason Barnard: How long do we need to wait?
[00:20:58] Amarjot Singh: Probably around five to 10 years. It’s not that far away.
[00:21:03] Jason Barnard: Wow. I think we can end there. That’s the most delightful ending to an episode I think I’ve ever heard. Thank you so much.
[00:21:09] Amarjot Singh: I appreciate it.
[00:21:10] Jason Barnard: Amarjot, that was brilliant. It was a really interesting conversation mixing AI, machine learning and business and I really enjoyed it.
[00:21:18] Jason Barnard: I got a lot out of it. Thank you so much. You get the outro song. A quick goodbye to end the show. Thank you, Amarjot.
[00:21:27] Amarjot Singh: I appreciate it, Jason. Thank you for having me.
[00:21:30] Jason Barnard: Thank you. Delightful.
[00:21:31] Narrator: Your corporate and personal brands are what Google and AI say they are. We can give you back control. Kalicube.
The event is 100% free:
Organized by Kalicube.
