Using AI as a data opportunity

- Whenever I talk to people about AI, I encounter people who feel like AI is a threat for a myriad of reasons. But one of the reasons is the task of starting work on an AI project like this is daunting because of the data problem. And I tend to look at it from the other direction, that this is our opportunity to rethink how we handle data to say, "You have all this great data in your company, but it's a mess because it's always been designed to work for the people who are working with it. This is your opportunity to consolidate the data, get it all into one place, structure in a way that it makes it more accessible to the people who need it and more manageable for the people who need to manage it." And then as a bonus, you can start building AI on top of it. But even if you didn't build AI on top of it, you would still have the benefit of having properly structured data and properly governed data now. - I see the world heading towards narrow AI, actually not general purpose AI and what we call as AGI. Artificial general intelligence is way, way far away right now, like I'm thinking more about I need to survive. I need to succeed, my company needs to succeed, my community needs to succeed. My city or government needs to succeed. So I need to think about doing what is possible in my realm, but with responsible AI in mind. And to that extent, my advice to people in this area would be to start with a smaller scope, a manageable scope with a good vision, a good target architecture in mind, a blueprint. You start with a small area, connect an AI application and ground it to your own enterprise data. It could be a simple SQL Server, it could be Cosmos DB, it could be OneLake, whatever it is, right? Wherever you're storing your data. SharePoint, for example, right? A lot of people want to use SharePoint and Teams. So the data could be in all of those places, but we build it in a simple and easy way. Start small, build an application within the controllable realm. Verify, validate that it works for you, train more people in it. Check the efficacy of the model itself and the outcome, the output. Build the right governance. Now you have a good play model, per se, right? Like a MVP to get started with. Now you can expand the scope, but keep in mind that security, safety, responsible AI never goes away. All the time, for every scope of item, you need to keep thinking about the data security, as well as the user security when they're using that AI application on top of your data. So maybe start with internal applications first because it's more controllable and usable. Then external applications, which are customer facing, and then to the whole world.