Identity-Powered AI: How Identity can Become AI's Saviour

We can get started. Hey guys, my name is Santosh. I'm a product manager at Indy Kite from Toronto, Canada, currently live in Oslo. So what's the expectations in this session?

First, I'll explain what Indy Kite does at a high level, and then we'll tie it to identity powered AI and what that actually means. So indica, in my words, is we are a data platform focused on identities and business data where the last mile, the last kilometer of the data use utilization problem. So what that means in plain terms is that we try to make your data accessible and structured and reliable for the right purpose. So imagine I'm a a business analyst and I want to get data to make a dashboard. How do I know what data I need and how do I know if that data's reliable?

And this is what we mean by the last kilometer, last mile. Now, when we look into identity powered ai, a lot of buzzwords there. So let's simplify it. Identity means looking from the perspective of either a human, a machine, an AI agent, an IOT device, and then centering that around AI applications, AI use cases, and we'll talk about how we think about that in detail. But before that, let's talk about the AI wave. Everyone knows about chat, GPT, everyone knows about the innovation that they're doing, but it is putting a lot of pressure on enterprises.

They need to stay competitive, they need to innovate, and they need to innovate with ai. However, the technology or the solutions they have deployed doesn't cut it. And what I mean there is from the data point of view, do they have the right data? Do they have trustful data that they can feed to their AI applications? So this is where we think identity fits well. And this is what in Kai is making a big bet on. So there's no such thing as AI revolution without reliable and secure data. There's visibility, there's intelligence, there's action. We indy kite focus on the intelligent point of view.

And we strongly believe in order to have intelligence, you need knowledge. And knowledge really means there's visibility. Do you have visibility of your data ecosystem? Do you have visibility on the reliability of the data? So data provenance, where it came from, which identity it came from, which then ultimately enables action. And action can be anything from getting data into an application, doing a get request, doing some type of query action driven.

Also, please feel free to ask questions. I I was just gonna ask One myself. Yeah. Is anyone in the room thinking about AI initiatives right now for their own business? By a show of hands, yes.

Okay, good. I would love to but it depends on Yeah, Yeah, yeah. Compliance. Compliance is an interesting one. Co.

Yeah, we'll get to, that's, If you don't mind me asking, what's, what use case do you have in mind or what's your vision? Well, a slightly in a slightly fantasized version are, or this world, wouldn't it be great to like replace an analyst with an ai instead of going to a person and asking for their opinion, you just click a button and it would generate a complete, let's say, leadership compass about your, the whole product market. Yeah. But would you trust it?

Well, that, that's why it has never happened before. And hopefully it'll never happen too. But small steps in that direction, yeah, be really Helpful.

Yeah, I agree. I agree.

So again, if something's confusing, you don't know what I'm talking about, let's double click on it. So feel free to just raise your hand. So the identity centric perspective is missing, as I said before, from a systemic point of view, identities are the first point of interaction. Identities are not human. It is devices, AI agents, data products, data sets, you name it. Something that we realized, and it became more apparent actually in this conference is that we're not real. The IM space is not talking about IM data.

They're, they live in silo systems, but they're not consolidated into one view. Meaning they can be one identity in a CRM system such as Salesforce and on one identity. And then IDP both indicate the same person but in a different context or relationship. Why? Like why? Why are we not talking about that? And it becomes, it became very obvious here today or the past three days. And then the final and the most important is the lack of trust. As you said, if an AI produces an output, how can we trust it? Especially in an enterprise setting, you're gonna assume that you can trust it.

How can we enable that? A good example is chat. GPT. You asked J chat gt a question, it gives you an output and in fine print it says, do not trust this. But we still do. So that's an interesting case from system of records to a system of intelligence systems of records.

This, what this really means is, oh, there's a thing, hereSo records, what that really means is relational databases, rows on the table. That's what system of records mean. System intelligence is taking those data and adding relationships or adding context to it. So they're, they're just not rows, but they're rows that are connected to each other. And then systems of engagement are the applications the top of the system of intelligence. So these can be web applications, this can be mobile applications, this can be chats, it can be AI agents, you name it.

And if we look back to your action intelligence visibility, that was in the previous slide as well. Intelligence depends on visibility. Visibility equals knowledge. And then action depends on the quality of intelligence. You can kind of translate to us as humans, we need knowledge in order to build intelligence. Intelligence then leads to action. So How do we build this layer of intelligence?

We, we think using a knowledge graph is the best way to do that. And really a knowledge graph is a collection of nodes and most importantly relationships. In this example, we have digital IDs. So this can be a digital twin. In this case we have Alice and Rachel and they have properties and they have relationships between the properties. This tells a story in so many different perspectives. This is a relational database cannot do that to you. A system of records cannot do that to you. But what this then enables is more contextual queries. It then enables use cases that we cannot imagine.

If you think about system of records or, or having siloed systems. Well isn't it just a graph database? It is just a graph database. So where does the identity come into this? Great question. Great question. It's in my slides, so I'll just, I'll get there. Actually I can answer this right now 'cause I don't think I've answered it well. So identities comes from the notes. So we believe that a node, in this case Rachel or Alice is a digital twin, which then equates to an identity. Identity can be verified and should be verified. We believe that trust comes from identities.

And that's just a really simple basic sense. That's just a verified identity. Authenticated identity. If we can understand where the data comes from for which identity, we can be more, we can trust that more or have a more of level of trust or a level of assurance towards it. Does that make sense to you?

Well, it absolutely does make sense, but why not apply the same logic to devices and cars which they own? Can cars have identities? Of course.

Cars, any data producer. So when I say data producer, anything that can produce data, us identities, a car, an IOT device, they have an identity and we're trying to capture that. And those identities can be verified and authenticated.

Yeah, So like 12, 13 years ago, I think here in EIC, we've been discussing identity, relationship management, things like that. And this looks very much like that. And are you aware of that concept? I am not aware of that concept. So that is, but if you don't mind, can you give a quick 30 seconds of what that means?

Yeah, so it, it's, we talked a lot about it relationship management and it's about the, you know, not only the humans but the nose and softwares and processes and everything has got the identity. And it was about, you know, talking about we got to actually create a relationship graph so that we can act on those. And I think the access is given based on, on, on the relationship to something. So everything Ev everything which consumes or produces data can be considered as identity and through the relationship to to objects, you would have access.

So it's not about can I get access to something because I should too. I can get access to something because I have relationship to it. And we had, I think that, I think you mean the interrelation concepts by CANTARA or something like that?

Yes, exactly. And there are some real life scenarios here in Germany. I have seen as well in some of the companies, and I think one of the companies providing such concepts already, one of the IM vendors. Yep. 'cause one of our capabilities or one of our offerings or add-on services is knowledge based access, access control, which is really attribute and relationship based access control as you were mentioning. So that is one of the use cases or services that we had on top of it.

So may maybe to add to it and a bit of background about indite because it, it might make sense and fall into places. So the concept of identity relationship management was kind of invented by, by for back in the day. Right. A bit of background, Andres was one of the founders of ForgeRock four years ago. He left ForgeRock founded in Nikai. Why? Because he wanted to reinvent the platform based up on a relationship, relational or not a, a graph database based on an IM data.

So that wasn't possible because it would change the entire technology that was based on what happened after, well we all know, right? But he took that and he was thinking about it and expanded it. So there's a component missing, which is really any kind of data, let's get will be gear tied back to an identity. Identity. It can be anything. A car can be, can be human. And if we look at the data ecosystem, it could be CIM data, MDM data, contractual data, contractual data could be transactional data, realtime information that all ties back.

And based on that information, you build the digital twin, you are able to put the data in context and do have decisions, authorization decisions, content decisions pretty much enable also other applications, right? It can be a signal provider to provide four signals if a translation transaction doesn't make sense. So this is where we are, where, where the solution is coming from. Right. I hope this kind of makes sense. The the concept is the same people's relationship. Exactly.

And, and this is about identity. It's about, but it is about the relationship, the relationship to an individual, but relationship to a data point as well.

But the, the concept is absolutely the same. You're absolutely right. And this is the evolution I would say It's Actually quite, it's promising to us that you see that and kind of immediately recognize what we're going for, which is good. But we also should like, like we say, Indy Kite sort of walks a world between digital identity and data management. And T is one of our brilliant data management people. So quite new to the identity space. So that's where Sebastian, well I was gonna say Sebastian and I will jump in, but Sebastian's managing phone on his own.

So, but yeah. Great questions. Thank you. Yep. So more Questions or comments? It can be a discussion, right?

Yeah, but I think if you go through the slides, I, we touch on these topics as well. So does that make sense to everyone? Does Sebastian make your brains light up a little bit more?

Sorry, still clear. It's still fair.

Super, super. Yeah. So what identity powered AI means?

So, so as Fraser was saying there's data management and there is a IAM, it's a combination of both. So one is addressing siloed IM data or any siloed source system. 'cause we live in a siloed world. People like to to talk to their own system. They don't, consolidating is a hard problem. Second is structured high quality data products. And this is about how do we consume data in a reliable way. And the last and the most important is data trust. How can we trust or rely the data that we're using? And that's what identity powered AI means. So let's, let's double click on addressing silo data.

The data management point of view. So this is how do we capture the data flow from from various source systems into our platform. And then out of the platform we have to impose data management principles such as lineage data, provenance, data quality, you name it. And then we have this semantic data layer. This is what we like to call our identity knowledge graph, which is just a graph with a bit more power to it. We have opinions on how the data model or the pathology should look like. And then we have data as a reference.

And this is a cool topic because it, what this is telling us is that we don't actually need to store your data. Rather we can reference the data that lives in these systems. So in terms of privacy point of view, we don't actually need to store PII data. We can just point to PII data that lives in a system under our firewall for example. So when we address silo data, what it means is that we're getting data into, into our identity knowledge graph. Once we have data within our identity knowledge graph, the question becomes how do we capture it and make it more context driven.

So we have an identity knowledge graph that is rich with relationships, rich with business data, rich with identity data. How do we curate a data product that is focused on a specific business problem? And what that means is context. Because trust is very context driven, it's very use case driven. Trust is not aligned across everything. It's very driven for particular problems. So how do we capture that? And we capture that through our, our notion called data products. So data products is about capturing context through relationships and nodes. Lemme simplify that.

It's capturing context based on information. How we get that information is not really too important. But fundamentally it is relationships. Secondly is managing control and access to the data product. So who has access to the data product and what data does that consumer have access to? Any questions here? Use your data with trust.

As I said, trust is context driven. How do we capture trust?

We, there's three principles we look at. One is the graph topology. So where does the data live in the context of a graph, the context of the ecosystem, the context of the system. Secondly is data veracity. So this is actually looking at the data itself and looking at data quality principles. So for example, this is like data freshness or data consistency. So how does the physical data points actually look like? Lastly as I mentioned, is context, which is just relationships. How strong are the relationships around it?

How, what is the metadata associated with those relationships as well? And what that enables is data reliability. Can I rely on this data for my business use cases? So I didn't really touch on identity principles, but from our identity knowledge graph point of view, it is enrooted in our data model identities and the metadata associated with identity. So a digital twin has strict metadata that we enforce. So when we, when we have a verified identity, we can do token introspect, claim some information and get it into our graph automatically.

Secondly is enabling AI with structured high quality, reliable data products. So we imagine a world where an AI model or A LLM will use our data products to be a rag. It would come and ask us, is this, what additional data do I need to give a high quality response, a truthful response? And then beyond project thinking, and this is the flexibility of having a graph at the core, a centralized graph. Because you don't have, you don't have to define your schema or your data model from the start.

You don't know how your data will look tomorrow, five days from now, two years from now, it's a hard problem. But having a graph we can be flexible in that schema and grow as you have more business problems. So it's just thinking as a foundational layer is where we are. I think that is my last slide. I might have a totally stupid question but I'm still of trying, There is no such thing as stupid Questions. So like one of the biggest selling points of large language models now is basically they, they, you could just give it like say A PDF file.

You don't have to explain the data model or anything it, it would supposedly understand everything on its own. Where would all this additional context come into play? How would your accompany A PDF file with this identity context or any other graph information? Would you have to create a new kind of interface for an existing LLM or would you have to force like open AI for example, to learn about your data structures and kind of implement a special connector to your solution? How would it all click together?

Yeah, so there's two types of data. There's unstructured data and there's structured data. A-P-D-F-A file, a image is considered unstructured data. Sure. Which is why you put it into a vectorized database and then you index it, whatever. But when we talk about our graph, we're focusing directly with structured data. So that's the difference. Fundamentally that's the difference. So we work with structured data, PDFs or unstructured data. When you have structured data, a knowledge graph or a graph based data model becomes crucial because you then you have relationships that add more context.

Whereas in a structured, you have a vector DB that's doing a lot of that magic for you. And you saw your second part of that question I missed Anyway, the original stupid question was if an LLM, if the existing AI can supposedly work with unstructured data easily, like why would you spend more effort and fit all this additional information? Like what would be the benefit for it?

Would it, Because how would you access the structured data? Are are, are you implying that unstructured data is more important than structured data Or just why would you use structured data? If you can use unstructured Data because they serve different, the my, this is my opinion, it's not, maybe not truthful, but they serve different purposes today. Relational databases, store structured data, structured data is how applications are built. This is how we solve problems today.

Unstructured data is, it's a hard problem because we don't actually know how things exist together and that's when the vectorized databases become very smart. But they both, we they, they co-exist together. It's just not one is more important than the other if that Makes sense. So how would you make it work? Like can you go and say hey change a pt, do a select asterisk from my data whatever, or would you have to create some kind of a proprietary connector or do you have a standard for it?

Would you just work with through sql what would be the, the actual interface between Perfect them and you? Yep, that's a great question. So our interface is data products as I was thing which is basically a rest API layer. So you communicate just to rest API and the schema of that API is something that as a consumer you can define when you define that data product, which then is much more tailored to your use case. So it's through an API layer is is what we expose. Okay. So just to understand, I correct me if I, I may be totally wrong.

What I got from the, the presentation as the impression is that you are building a graph database in, into, and the each node is identity based and you the cus your customers could start utilizing that and you know, because LRM has limited context, you may want to do the rag and for the to do the rag, you are querying that graph database to actually extract the, the relevant data which is linked to the relationship so that it's going to be even more relevant than just doing the vector daily such.

And you are taking that result and push it to the, the, the LLM context so that the LLM can actually give you more sensible answer. Something like that. Exactly.

That's, that's the AI use case that we see as prevalent Ish. Do we have any questions from the online audience? Not yet. No. We have an online audience.

Hey guys, Can you maybe like show us or at least like tell us about like a practical business case for such a solution and why would it be better than quote unquote feeding it a PDF file? To be truthful, we haven't had a use case to work with unstructured data. So it's hard for me to explain. But the reason why thinking of graphs and identity centric is about the trust point of view. And the reason why that's a bit different when unstructured data is because when you have structured data, you know what identity it came from, which then results to trust, trust then results to knowledge.

Knowledge then translates to intelligence. That's kind of the, the value chain in our brains. Yeah.

That we, I Think that's the point, right? If you're feeding an AI engine data, you want to know that that data is So kind of going back to that car example. So you would be able for example, to work with an automotive company and make sure that the car telemetry data they would feed to your product would be properly not just kind of authenticated but actually provide all this whole lineage and quality information and that would supposedly improve the whatever F1 racing statistics analysis, stuff like that, right? Yes. Okay.

Do you actually have some working projects or existing customers using that? So we're working with a financial service company in the Nordics, similar use case getting data, but we use the data as a reference concept. So we're not storing PII data, we're pointing to it, but they're using us as an aggregator to build that trustful operational layer. And then we also have a manufacturing company as a similar use case, but it's more focused on authorization. So do they have, can they have access to the right data? 'cause they have a data sharing use case.

Great, thanks. So if you do not have any further questions from the audience, I think it's just the right, oh with, okay. Did you say that you are just walking with a structure data right now? As of right now we only work with structured data, but I do see a future where we will need to support unstructured data as well.

Yeah, because I I, I think even for the documented repository, the identity relationship managed, you know, data can be actually pulled off from the, you know, the document repository and pushing it to a, to the rag might be really powerful. Yes, I agree with you. I agree with you. Probably even more than structured data. Yeah. But a combination of unstructured and structured oh my May maybe one more comment. So unstructured. Unstructured, yes. But also where the data is coming from, right? We're at the moment we are looking at internal data ecosystems, right?

But we are also are able to pull an external data sets, for example the device location from a network provider, right? So you're not the, so you're determining the position not only in IP for example, but on the actual network position once you want to do an authorization decision for example. So when there's an open API, you can enrich the decisions that you make. For example, from another authorization perspective with the wider data ecosystem, which is also quite powerful if we think of G-S-M-A-O-S-D-U, all those initiatives, PSD two open banking whatsoever.

So we, we, we need to wide the scope a bit in the data that we pull for any kind of decision that we make. Yep. Like best way to think about it, we are a thin layer.

The, as I said in the start, the last kilomet, the last mile of your data utilization strategy. We are a thin layer that connects all the dots together with then enables authorization, fine grain authorization, then enables rich trusted data use cases. Well thank you very much. That was like extremely thought provoking. If a little bit confusing at times, sometimes I would totally feel I am not getting something, but it's out there just to grasp. Yep.

I believe I will be following up with some more questions through the usual channels and I would suggest everyone else who is interested in the subject to do the same. Feel free to reach out to the Indy Kai team, myself, LASA, whoever we are more than welcome to answer questions and learn more. This is a hard problem to solve and we're kinda the first players to try to tackle this, so.

Okay, great. And again, we just reached the official end of our trek. Thank you very much to all those Yeah.