Conviction & Code Utilising Foundational Frameworks to Build New Economic Ecosystems

All right, we'll uh start right away with uh with Aiden. So, Mubala, obviously we think about it as this uh global player in deploying assets, but what are you doing in your in your AI department right now to bring that into the private equity portfolio?

We're doing everything a lot.

Yeah. Um on the private equity side, we've basically reimagined our entire investment life cycle with AI at the center. So from screening, due diligence to investment approval to post investment monitoring and on top of that we're actually uh AI enabling our corporate units and we're actually working with inception with Ashish to actually do proof of value. So we're actually commercializing some of our products. So the AI enabled solutions we're building, we're actually commercializing and selling to our port codes to accelerate AI adoption.

Yeah. And can you just talk a little bit about the the product itself? So what so how how does that help uh you know uh the investment committee or portfolio managers to go and and and pitch for u or pitch the internal teams on on their next investments. So what so what are you doing there?

What's the value created?

So look, we're we're already seeing uh significant value creation. we're seeing on the investment life cycle uh scaling effects of between 20 to 30%.

Yeah.

So that's a lot if you scale it across two or 300 deal teams.

Um and then what we're also seeing because we're moving into the next frontier part which is um we're creating a digital twin for our investment committee. So what this is doing is actually using agents to simulate uh the investment decision discussion. So putting deep insights into the context of a human dialogue. So you know it's only an augmentation of the decision process but it really adds another perspective and also maybe even more valuable. AI what we're seeing is providing very strong contrarian views.

Yeah.

So it counterbalances and actually challenges the thinking and the process of making the decision. So Ashish, you're you're you're you're working with Aiden on this and um you know when you create these digital twins or you want to you want to come up with this kind of simulated uh committee um first of all how do you go about creating that product and then does it does it also introduce a kind of confirmation bias in a in a sense right like you're basically replicating what's already there rather than adding um a whole other data set of of information to it. So um yeah uh so excellent question and I won't talk about the product because I think Eden has described it in in full capacity there. I think the partnership that we have in Mubadala is a great example where we as builders work with the subject matter experts at Moubadala to actually ensure when we build it we build it with conviction. uh so why I'm trying to say that conviction is essential here in most other enterprises that we've seen there's a lot of experimentation but at moubadla we focused on and will agree to this point on setting that foundation first and they built a data platform where they ensure that the data comes in comes in clean so that we can spin off use cases to the ones that exactly as what Aiden talked about from investments to procurement to legal that goes in the right way and and the at the center of all is definitely trust.

They ensure that the data goes in is audited and there's the right compliance mechanisms even whether whether it's on processing or inferencing to get this across the line. So if I have one takeaway from this is I think to get the same level of conviction customers such as Mubala has on driving this forward because that drives change. We take that into our products and know what we want to build. And I was that my offline conversation with Trevor is what I was saying is we are focused on what we know what we need to build and there are other solutions that we can just bring into this one single platform to offer the client in this case an investment end to end all the way from sourcing to an IC memo built completely agentic

and and just on the on the origins of that and you know to to to do the pun of inception do you go to Mubada and tell them we can we can create this for you or do they come to you and say we need we need uh this application or this platform.

So there is I don't think it's a one-sizefits-all in mobile. I think it's just a perfect DNA how we work together. We've we've reached that stage because we've been engaged for more than a year right now. We we have a sense of the wave when it's coming and we're prepared for the investment case. It's completely driven through their leadership.

His excellency Kaldun, he exactly knows what he wants. He drives it through the head of AI enablement where we have conversations and then we just draw it on a on a on a whiteboard and just make that into a reality there.

All right.

Can I add a little bit to that?

Sure. Please. Please go ahead.

This is where the magic's happening.

The magic. What?

This is where the magic's happening. We are co-creating. So, they're playing position with technical partners. We've got a lot of the business domain knowledge.

It's it's really working very well with the partnership model.

All right. Well, Trevor, I guess this is exactly what you do in in your work. So, when you when you hear about this uh you know, them working together, the the the customer and the and the software provider in a sense, where where do you come in on on on making sure that data set is is actually, you know, the the the right data for them to be using?

So, scale sort of has two two businesses. The first is uh we've worked closely with the model builders over the last decade and uh and have provided human data that actually allows these models to grow. And so we are big believers that data is the the differentiator in a lot of these models. And I think there's this, you know, we were talking before, but there's certainly a theme that many enterprises and even governments have tried pilots and found them, you know, they they leave something to be desired. And often that missing ingredient is a really in-depth understanding of the data, whether it's the sort of ontology of the system or it's the quality of the data.

And so that is always where we start. Um and I think that uh you know part of the way that we work is we sort of understand you know iteratively figure out what the customer what what the intention is what's the big vision then you really got to dive into the systems and see where you're starting from. And once you have that starting point and you can assess kind of the quality of the data and the quality of a a subsequent model then you figure out what are the additional inputs that you need to sort of bring into the system which is where our kind of experience on the on the large language model side really plays in.

Yeah. and Lake, you know, when you know, and and we we're it's really great to have you here because you have two hats. You're, you know, you're the academic and you're also the investor in in in a VC. Uh when when you hear this happening, uh what do you think about the actual applications that are coming out of it? Uh what what is the what is kind of the the sweet spot of the value that's coming out?

Yeah. So, so when we think about what actually have pushed AI in the last 10 years is really the three things. one is the data and second is compute and third is algorithm and then when we're looking to the data how it's driving all the innovation of AI the key thing is do we have high quality data as Trevor mentioned so high quality data means that the data is available to train the models but also for application companies it means that do they have the ownership of the data do they have the right information and do they also very importantly is can they ensure the data is protected the privacy is protected and also meantime there is continuity of how you collect data right so as we think about you know there's a lot of companies say we have for example we're hospital that we have so much preparatory data so we should be uh have a lead time compared to our competitors however if there is no data flywheel which means not only the data is usable for your model but also you don't have a product solution let you continue to refresh that collection and annotate that collection and that will be very difficult. The best way we have seen so far is that people who have embedded the data collection process into their existing workflow. And so one example actually I want to share with people is that most people think that you know data advantage is only applied to tech uh forward companies uh large company like Uber Airbnb uh but I'm actually serving on the board of a traditional industry diamond industry.

So I serve on the board of GIA. We we are the largest diamond grading industry in the world. So what's interesting is that we think about diamond grading is is computer vision problem. It's so much harder than look at X-rays, right? So if you look at a diamond inclusion and you look try to look at 200 diamonds in a day and you're going to go blind and it's not consistent.

The traditional uh work process is that we have a one grader two graders grading at the same time. If they agree then you got a certificate. If they disagree you escalate it to a senior grader and that's a very natural process to you replace one human with AI agent in this case right so we see so much success with that

on that note is is there a kind of a cross cross-pollination of like this industrial sector knowledge that's happening into finance or is everybody kind of creating on their own and like really this is an open question. So digital twins exist in uh in manufacturing processes. uh there's a there's a whole bunch of lessons already out there and um you know it seems that everybody everybody else coming in is like okay I got to do this for finance I got to do this for tech I got to do this for government services or are they are they learning across

uh so we I think that the the question is that where we're when we think about different applications is that we look at a different application are they by isolation or they actually also collect other information from others so one of the example we're seeing is as we're moving from large language model to multi modality to uh robotics. So we're seeing the information actually collected a lot of them are not on just original training data set but also onset data collection as well. Okay.

So in this case you're not looking at one single providers you're looking at different signals from different providers and that's why also AI governance becomes so important. Yeah, if I could add kind of two points. The first is when we you know across totally different domains we are seeing similar components. So take the example of you know document extraction. These are sort of capabilities and tools that as we build more and more become more repurposable to sort of accelerate I think builds in new domains.

I think the second point on the on the data side is particularly to a a financial audience you know 95% success in a chatbot is is like a very high mark in in a financial sense. that could mean lawsuits and and major significant losses. And so to get that final 5% of reliability, which is something we think about a lot. Um, as as Lake alluded to, it's not just the data that's in the system. It's the new data that gets created from from the use of the product.

And that's a big distinction, I think, from software is when when you have users that are using it, you're actually creating new very valuable data for that specific institution. Um, Aiden, this is something you kind of alluded to in your in your first uh response, but the the platforms that you're you're you're uh co-creating with um uh with Inception, what are you going to do with that afterwards? Right? Because like for you, you're going to use it for for from for white label it, sell it somewhere else. Is that is that the plan?

Absolutely. I mean, that's the best proof of value, right?

Yeah.

Really, it's a really good tangible example of proof of value. So, we're working on um actually still thinking about working on how we could commoditize and commercialize our investment life cycle solution, but we're already commoditizing and selling uh we've already sold I think to three companies in our portcover which augments the um board decisions and actually allows them to kind of um analyze and extract out information from all the unstructured content. Um but I think look the goal here is to to really prove the value of the tool and secondly I think this is very important the main driver actually I'd say primary driver is to accelerate diffusion um if you think about these revolutions we've been in through industrial revolution you have maybe 200 years to diffuse technology digital 50 I think AI is probably two to five years so what we're doing with this commercialization channel through our accelerate impact with inception is to push these products into our port codes.

Yeah.

So it helps them accelerate up the journey as well.

Yeah. So is that is that is that why you actually went with them to be

No, absolutely. I think uh as an anchor partner to actually build this inhouse to see and make mistakes but fail fast pivot and adapt into the right solution is exactly what we did in terms of having an wonderful lab as mobile is offering and then as Aiden rightfully said we've actually rolled it out to a number of port codes where we actually seeing the not just uh the value that it's been creating but is also giving insights back after the investment hold calls on how the enterprise is operating and providing insights there.

Yeah, Trevor, one of the things that you know I spoke about earlier is that uh kind of this like hype versus reality. So there was I think a McKenzie study recently that said in a survey at least for the region 11% of companies feel like they're getting any type of bottom line benefit from from deploying AI in their companies. Um it's not magic in a sense, right? And and this is the dirty work that's involved. Is this what uh people I guess are missing in a sense of when they want to want to apply and work on a co-creation type model like this?

Yeah, I think um it's it's it's a obviously a common topic right now. I think there the promise of AI is there. I think the timelines have been unrealistic and this is a this is not new to AI. This is a sort of common uh problem that we have in assessing technologies. I think what we do see though is if you're willing to engage and if you're willing as I mentioned before to sort of iterate towards these solutions, we're seeing extraordinary impact.

Um, and that comes from u I think a pretty significant mindset shift away from software where you narrow in on the problem, you get really detailed about what that solution needs to look like and then you hand it off to somebody to go build it or to buy it. And I think with with AI, if you want to find success and you want to find really high impact, it is, hey, this is the north star and we're going to sort of iterate our way to that. Learn from users, learn from what's possible, learn from where the data is, but there's a lot of we call it data schle um of kind of figuring out all the little nitty-gritty hard stuff to get you from pilot to real impact. Yeah,

if if I can just chime in really quickly on to what exactly what Trevor said. One of the reasons why it's still a hype and is is the process how you actually stratify it in in mobile of for example there's a department of AI enablement most enterprises it's considered as an IT budget and eventually comes the ROI is not seen and then it just dies off into what we call as death by pilots but it actually doesn't see I think the focus should be you create an enablement department or a transformation department which has a budget which ensures that the CFO sees the needle moving in terms of ROI and that's the only way and choose the most complicated workflow which will add value and that's the only way you can start enablement there. Yeah, Aiden, I want to get to get to you a little bit about the uh you know, the other question is the frictions, right? So on the on the data, you know, the AI data uh or AI infrastructure itself, we don't have enough compute in this region. Uh you can't send a lot of your information out.

We're going to get to you on how you're going to build that out, but um how critical is it at this moment like as you as you're building this up? Like do you need it now? Do you need it yesterday?

Need it yesterday.

Yeah.

So look, I think this is a very important point. I mean we we've um been thinking about look we're looking to the next frontier now right

I think we've gone up this very accelerated delivery path together with inception if we want to be on the next frontier which you always have to be right you're constantly having to move to the next leg you know if we can't we can't move our data because it's highly sensitive so it has to be resident in the UAE we don't have the uh high performance compute here which is being built out rapidly with um Uh it's 200 megawatts coming on stream right with Stargate. Um and then also the other part of this is the next frontier models.

Yeah.

Because these models don't reside here yet. So we can't really take that next jump until all of this is in place which hopefully will happen next year.

Yeah. So is it is it on target and uh

no no uh next time Aiden and I should not be on the same panel because he takes away all the talking points that I have including all the product build. Anyway, uh I I know I'm conscious of time. Absolutely. I think it's a journey in this AI transformation. We start off we're showing real life use cases creating impact which we've done but eventually you want access to the frontier models bring that into the country we're now converting I think thanks to the leadership we have moving Abu Dhabi from just being capital of capital to capital of compute and the capital and that's the part where sovereignity is not just about the data it's about the model it's about the infrastructure it's actually also the human capital where we're getting a lot of individuals local individuals working on this data sets to build frontier solutions Yeah.

Yeah. But on on that point cuz like you're you're you know your parent company is G42. Do do you I'm sure you have some sort of u outlook or forecast on like what is that pent up demand like that that 200 megawatt how quickly uh will it be uh I don't know used up or or or leased whatever the

as Aiden had rightfully said they everyone wants it yesterday. They all have a pipeline of use cases ready to deploy provided data residency processing inferencing is done in the country and that's the battle of how quick I have a colleague of mine from KZna who's actually right now there in the middle of the desert actually building this at scale to ensure we can satisfy the customer demand here. So, so Lake, I'll give you I'll give you the last word on that on that, you know, uh resource allocation problem. Uh what's going to run out first, the the computer, the data at this point? Uh what do what do you see it?

I think the first thing right now is the energy. Uh so because right now all the foundation models are very compute intensive and so one area I think both on the research and in future investment that we're watch closely is energy efficient compute. So energy efficient computer if you think about it and actually exists today which is our brain we have a very small footprint we use almost no energy when we think retrieve information. So the latest research uh if you want to look in this is called free energy principle and active learning uh in active inference. So those two areas focus on one is learning from human brain how we work and the other is learning from for example like um colony of ants or bees because each of the ants are not sophisticated but work together is great.

So ants and brains and AI all together. Thank you so much to the panel leash Trevor and Aiden. Uh thank you to the audience for listening.

Thank you.