Insight by Iron Bow Technologies and Dell Technologies

AI success means starting by asking, What is the problem?

Artificial intelligence is a powerful technology, but it’s not a magic salve you can apply to a process to make it better. If anything, AI – and the related...

Artificial intelligence is a powerful technology, but it’s not a magic salve you can apply to a process to make it better. If anything, AI – and the related machine learning, robotic process automation and even data analytics technologies – requires more attention than ever to an eternal basic of information technology deployment.

Specifically, success in AI and the other data-driven technologies starts with a clear and defined notion of the expected outcomes. That means the business or process owners are those most able to initiate successful AI projects because they most intimately understand the business challenges they face and the mission outcomes they must deliver.

“It’s the people who are living the problem and understand what’s going on day to day,” said Jim Smid, the chief technology officer of Iron Bow Technologies. “The IT manager, the CIO – they don’t understand what, really, those problems are. And so they don’t really understand what the outcomes need to be.”

Therefore, Smid added, the choice of specific tools and technologies should come closer to the end of a project design than it typically does. Problem definition and selection of data sources that will enable algorithms to produce expected outcomes must come earlier.

In fact, choosing a tool too soon can limit the power of a project. For example, choosing to automate a time-intensive process with an RPA tool might mean you overlook the opportunity for a more comprehensive solution involving API’s and programming. AI can come from a variety of tools and solutions.

“RPA might be a good thing for prototyping things,” Smid said. “That also might be a good thing long term. You don’t really know until you understand a little bit more about the data, and the process and what’s all going to be involved.”

Smid cited one client. “We do a lot with the VA. We’re not talking to the IT folks, we’re talking to the clinicians about what the day in their life of looks like.”

Having a view of the problem from that standpoint can enable the IT people better understand the totality of technologies that might be brought to bear on a problem. For example, the clinical need for reliable and easy-to-use remote sessions raises not only automation questions related to data gathering about patients and building trend analyses, but also about technologies such as 5G to reduce latency or the need for more intuitive user portal design. Perhaps better image processing with improved cameras can produce more and higher-quality data for applying AI to pathologies with visible symptons.

Smid cautions against the common mistake of thinking too grandiosely. In the defense domain, for example, often what operators really want is solutions to day-to-day problems that sap people’s time and attention, rather than an algorithm to bring world peace.

For example, the predictive analytical approach to scheduled parts replacement that has taken hold in the industrial sector can adapt to military logistics and sustainment. Smid cited the now common practice of replacing disk drives before they fail.

“That was game changing,’ Smid said. “To be able to do that with your helicopters, your tanks, your jeeps – all that predictive maintenance is a critical piece. It’s where a lot of resources are going to get applied now, because there’s such a big impact.”

Smid also advises agencies to think about the computing architectures best suited for the AI and related applications they do deploy. He said the trend for internet-of-things data pushes processing – the running of the AI applications – to take place at the edge, where the data is gathered. Often that’s where users actually need outcomes, and it saves the costs and complexities of shipping data to and from data centers or commercial clouds.

Shape

Definitions of Artificial Intelligence and Machine Learning

Most IoT projects we talk to people about, it's not about the tools themselves. I can start with a tool but not everything is a nail. So I don't always need a hammer. It's important to really look at the business owner, or the mission owner.

Shape

Automation Use Cases

The reason for pushing compute out to the edges? Moving data can be expensive. I don't want to move all the data all the time; I want to move the data that's important. Or maybe I just want to move my outcomes to a centralized location. So managing that data, and making sure you're crunching it in the right places, are really important.

Listen to the full show:

Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.

Related Stories