The amount of data agencies generate each day is making it difficult for agencies to process. Agencies are looking to artificial intelligence (AI) solutions to help with challenges, such as the concept of data gravity, which is the idea the applications can move data to solutions. Matt Langan, Government Technology Insider’s podcast host, discussed the notion of data gravity and AI solutions with Kal Voruganti, Senior Fellow and Vice President at Equinix, and Scott Andersen, Distinguished Solution Architect at Verizon.
Matt Langan (ML): Thank you both for speaking with me today. Can you tell us what is data gravity and why does AI need to move to the edge?
Kal Voruganti (KV): Essentially, data gravity is the notion of large amounts of data being generated and that it’s expensive to move it. So, agencies want to bring compute to where the data is.
Scott Andersen (SA): The reality of data gravity right is that it sucks the applications to it. I have looked at and considered the concepts behind what edge computing could be, and data gravity is a driver. As we move towards this new world of 5G and this edge computing paradigm, it’s all about getting rid of some of the data and create an infrastructure that removes the data. Well, edge is the perfect place for an AI to sit and sort through the information that humans don’t need to see.
For example, if you have 16 hours of DVR data from your video security system of your front door, but all you care about is the 32 seconds when somebody actually breaks into your front door. You need to get rid of 15 hours 59 minutes and 28 seconds, that’s where AI provides value. The value proposition within the concept of data gravity is the sucking of applications and devices closer to the edge. The edge computing paradigm and adding AI to that allows us to begin this concept of let’s truly build applications. But, not in the way that we’ve done them in the past, where it was exorbitantly expensive, and people had to have massively redundant hardware.
With edge computing, we’ll literally be able to have cheaper arrays of hardware spread out, and the AI will simply keep the conversation going with the device. Every device that connects into the edge would have its own unique signature, and the AI would know who is having this conversation with the different signatures. AI is not limited by the human multitasking problem, which is that, at some point, when we multitask, humans stop listening. AI will continue to be able to listen to the multitask and multi-conversation process much longer, so that’s why data gravity will drive AI to the edge.
ML: Let’s talk about AI Anywhere. How is this solution helping?
SA: AI Anywhere is the whole reality of ubiquity. A key driver is the expansion of 5G. . 5G will give us a low latency network that is ubiquitous. It’s everywhere and individuals will no longer have to be worried about a dropped signal, weak signal strength, where you are, because between moving AI solutions to the edge and building out the 5G network, together this will give us this concept of having ubiquitous AI and will allow us to have it implemented virtually anywhere we go.
KV: I would like to use a couple of examples here. People are now building solutions called smart stores, where customers go in and look at items. Then, based on their facial features and how they’re actually reacting to the actual items on the shelf, stores want to figure out what ads to place on their smartphone. Similarly, there are “smart cafeterias” where a child takes a tray and places what they want to eat on it. As they walk through, they don’t need to stand in a line for the checkout cashier. There are smart apartment buildings, where they’re looking at the security for the apartment building, and quickly want to identify any type of security threats.
Essentially, in all of these examples, a lot of data is getting generated. Agencies have to have solutions that can process that data wherever it is getting generated. I think people are now looking at, and figuring out, , exactly where in the 5G world that training should happen and where the inference should happen.
There are two phases in AI. Historically, the inference will be in your smartphone, your Tesla, or in the smart store. Whereas, training will be doing on a more metro level location, because the training equipment needs a lot of power. It is building models and are usually offline tasks, but nowadays training is becoming real time. It requires a lot of power, so traditional closets in a store cannot support that kind of power drop. You will have a bifurcation and the inference will be closer to that edge. Whereas, training will be more in metro level data centers or in the cloud, because that requires a lot of power and a lot of computer hardware. AI Anywhere is the type of vision we’re seeing.
ML: Why do organizations need to share data or access external data?
KV: Let me give you a very personal experience to illustrate why agencies need to share data. I teach school on Sundays, and one of my students got diagnosed with a very rare form of cancer. They’re going to the best hospitals in the Bay Area, but the doctors are saying that they don’t have enough cases of this type of cancer. Now, her family is taking her to the cancer hospital in Houston and are hoping for more insights and help there.
Even those doctors are saying if there’s only a way that they could share this data about this particular type of cancer from all the patients across the world be able to better diagnose and better treat it. Why is this the case? Most hospitals and organizations are afraid to share their data, because they’re worried about what others will do in the privacy and security risks. The more data that agencies can bring together, the better AI models agencies can create. It’s not just sharing data of the same type, but also a variety of data that can be brought in to create better AI models.
SA: If you work in a software development organization, there’s a favorite little project that always occurs. They’re called skunkworks. What is a skunkworks project? Well, nobody knows. They just sprout up. The reality is all of these different little groups within a development shop begin to build solutions to a problem that they see, but they don’t share information, they don’t talk, they don’t say “hey what’s a better way to do this?” So, what agencies end up with is 35 or 40 different variations of a solution. If you think about rarity of information, it’s not just that the information is rare, but it’s that the systems are built for that information and the systems built to manage and improve that information are also rare since nobody needs it.
For example, if organizations are more open to sharing information, Coca-Cola and Pepsi could sit down and have a long conversation about “here’s the routes we take to deliver Coca-Cola to stores,” and Pepsi says, “here’s the routes we take.” Coke then says, “well we spend $10,000 a month on gas” and Pepsi says, “we’re only spending $7,000.” The two can get collaborate and realize Pepsi has a better system that will save Coke money. Sharing data in this scenario is a way for the two organizations to also help the environment and do something bigger and broader, which goes to the concepts of smart cities. If we get into this area where data is shared, we can begin to truly talk about entering the next phase of digital transformation, where the natural creation of data is digital only. If we think about only digital data, that is going to greatly improve the impact or greatly augment the impact of AI as we go forward.
To learn more about AI solutions at the edge, click here.
Click here to learn more about Verizon’s Professional and Managed Services.