LEADERSHIP VOICES
Sponsored Content:
STELLAR AI: AN ASTROPHYSICIST’S GUIDE TO GETTING STARTED
Of the 7.5 billion people on Earth, fewer than 600 have ever blasted through the atmosphere to make it into outer space and only a dozen set foot on the moon. Yet, through the study of astrophysics, we can calculate with certainty which elements make up the planets and the stars.
That’s not magic; it’s math.
To Jamie Milne, it’s also a metaphor for artificial intelligence. “Astrophysics is like extreme AI,” he says, leaning forward on a sofa at World Wide Technology’s Washington, D.C., Innovation Center.
“There’s no reaching out and touching it. You’re basically making observations and trying to figure out the universe from that data. It’s actually kind of similar in large organizations: They give us a set of data they think can explain their world and we use that data to understand what they’re trying to do and make it work for them.”
Milne studied astrophysics in his native United Kingdom, learning to code, normalize data and solve celestial riddles with computers. In time, he moved to the U.S., and began to apply those skills to other kinds of data to solve new kinds of riddles.
Updating Decision Making
As a senior engagement manager and consultant with WWT, Milne helps government customers grapple with the data they collect and figure out how to use it to better inform decisions. Many of these decisions are highly important and time sensitive, and though the data may seem mundane, the payoffs can be tremendous.
Like when a large agency came to WWT looking for help with analyzing contracts.
“The analytics behind that were fascinating,” he says, his face lighting up at the irony of a former scientist getting pumped up about legal minutia. “It’s like you’re looking at decoding human language, and understanding the sentiment of different phrases, to say whether that specific phrase has a positive or negative sentiment and then say, ‘What’s the risk profile of that? How can I think about that in a business context?’”
Contract language is complicated because its purpose is to set down clear expectations of both sides, but humans being humans and language being subject to interpretation, contract wording can sometimes have unintended meanings leading to unintentional consequences. The aim in this case was to find a way to use computer intelligence to minimize such risks.
“You could have two legal experts read all these documents and come up with two vastly diverging opinions,” Milne explains. “So, to be able to use AI to reduce bias is a really good challenge. You are decoding human language and understanding differences.” But not just for the sake of solving a challenge, he says. If it were that simple, data scientists might just spend their days doing puzzles.
The real excitement, the pleasure, the satisfaction in this work is that the data analytics are really just a tool, a conduit, a means to accomplish remarkable things. “Our projects are not there to advance analytics so we can do math. It’s to save lives, reduce costs, reduce risk.”
“The real excitement, the pleasure, the satisfaction in this work is that the data analytics are really just a tool, a conduit, a means to accomplish remarkable things.”
Digging In
The age-old knock on consultants is that they use your watch to tell you the time and then charge you for it. But, in the AI era, that’s not necessarily a crazy proposition. Government agencies today are awash in data, but that data is often not in a form that makes it particularly useful. Moreover, agency leads don’t always know how to turn that data into knowledge and value. To put a twist on the watch metaphor, Milne and his team not only teach their clients how to tell time, they show them how to use those readings to save time, simplify processes, reduce risk, save money and deliver better customer service.
“We take a very consultative approach to things,” Milne says. We start by breaking everything down to first principles: What are we trying to do here? What’s the big strategic objective? Maybe it all comes down to spending less time doing this one task. Once we identify an objective: All changes we implement from there on out have to tie back to it. And if they don’t tie back, you need to have the confidence to be able to throw out ideas and come back and start over.”
Not all agencies are created equal when it comes to tapping AI to refine their data. Some agencies show up with specific ideas and objectives, problems they want to solve. Others have a vague idea, maybe a couple of different data sets that might be correlated to produce insights, in which WWT undertakes strategy projects in order to define objectives and narrow scope.
One thing that’s true across all projects, Milne notes, is that in order to develop AI solutions that produce worthwhile results, agencies need expertise in both the content and context of the data, expertise in the mathematics, programming and analytical tools, as well as methodologies for extracting knowledge from that data. The two must go hand in hand to get useful results, he says.
By way of example, he points to one agency that had a large collection of images and a desire to use analytics to identify subtle changes that might otherwise escape notice. Without the analytics experts to pair with its subject matter experts, the agency wasn’t able to get started.
Emerging technologies and government policies will shape the way federal agencies use their data. How might Artificial Intelligence optimize data analytics? Download the AI-Driven Data Analytics Flash Poll for more insights about federal employees and their use of data for actionable intelligence, perceptions of AI-driven data analytics and challenges implementing AI.
Innovating Effectively, Failing Fast
Rain pelts the windows and a mammoth display on the arena outside the window lights up the late afternoon gloom as Milne continues explaining the intricate dance necessary to solve many of these data problems.
AI projects are the kind that involve people from every corner of an enterprise, he explains. There are data collectors, of course, but also data experts who make use of the data analytics, and experts who can crunch the numbers to figure out answers. Moreover, there are also people who make those answers accessible and useful to the end users.
The key to wading through all the intricacies in any AI project is to keep in mind that among these disparate stakeholders, there is a work objective.
“This isn’t a project, right?” Milne says. “It’s not an academic exercise. It’s: How do we stop trucks from breaking down? It’s maintenance. It’s about: How do we improve our margins? How do we secure our networks?” In other words, it’s about getting the job done and creating better tools with that in mind.
“It’s about chunking the problem down into its constituent parts [about answering]: ‘What is the thing we’re trying to prove or disprove?’” he notes. “There’s a lot of meta statistical analysis we’ll do to find out what happens on a daily basis, to see what patterns we can identify. All those contextual pieces of information just help you get smarter and then you come back to the drawing board and identify what other useful questions you can ask.”
As with any science, however, the initial hypothesis as to how to best tackle a project is not always correct. In those cases, the trick is failing fast.
“You know, most problems, there are many levers,” he says. “You pull one and nothing happens. So, you move on to the other levers. If you’re working on a cost-reduction project, you choose a lever that maybe has the most spend, but it turns out that process is actually highly optimized. OK. But maybe the process with the second-greatest spend hasn’t been optimized, and there’s opportunity there.”
“Smart people, working in an organization 20 years — they’re so busy, they don’t have time to look up and say, hey, maybe there’s a better way to do this.”
Setting the Stage
Not all organizations are ready to launch full-scale AI projects, however. Oftentimes agencies find that while they have an abundance of data, it’s unstructured and untagged.
“This happens in business, it happens in government, everywhere,” Milne says. “Smart people, working in an organization 20 years — they’re so busy, they don’t have time to look up and say, ‘Hey, maybe there’s a better way to do this.’”
In those situations, the goal is to start small and aim at using that data to achieve something simple. “Normalize and standardize your data,” Milne says. “So now we can have a common language across what everyone does.”
That may not introduce AI to your organization, Milne acknowledges, but “it will save you an enormous amount in hours and effort — and it will improve employee satisfaction, and that will improve innovation. It’s worth doing.”
Tasks like this, while small, help set the stage for future innovation. Like tilling and fertilizing soil in a garden, it sets the conditions for future growth.
“If people don’t trust it, whether it’s because it’s too much of a black box, or they feel threatened, or they think they’re not getting enough information, then you won’t get acceptance.”
Cultivating Trust
But all innovative projects, especially among long-standing organizations with set workflows and infrastructures, involve cultivating trust. Trusting the data is the first imperative, followed quickly by trust in the data analysts and in the models they use to understand the data. Finally, if the data is to become truly valuable as a predictive tool, Milne notes that there must be trust in the algorithms developed to predict the future.
“If people don’t trust it, whether it’s because it’s too much of a black box, or they feel threatened, or they think they’re not getting enough information, then you won’t get acceptance.”
In order to cultivate the knowledge and skills necessary to generate trust, WWT relies on a vast and diverse talent pool — Milne’s AI team has over 90 people, many of whom grew up in different parts of the world. “There’s a lot of diversity of thought,” he says. Moreover, WWT also calls on its extensive technology partnerships. The company is an integrator, but its roots go back to its launch as a value-added reseller, which means those relationships are richer and less transactional than an integrator that is entirely vendor agnostic. WWT’s $600 million computer technology lab makes it possible to experiment with how different tools, systems, and technologies interact, and ultimately to find the best combination of tools to fulfill a need. And its relationships mean there are experts on hand and partners standing by ready to help.
This is what makes the business effective, Milne says. That, and maybe a little magic.
“The culture here is incredibly important to everything we do. Everyone who’s a leader here — anyone who has a direct report — goes to a conference every year for a week to talk about culture. How can we be better? How can we provide people better opportunities? A whole week talking about trust,” he says.
While WWT’s cultural approach differs from what Milne has previously experienced, he has come around to see just how valuable it is to shaping an innovative workplace.
“Now I see that this drives everything,” he says. “The more driven you are, the happier you are to get out of bed and come to work in the morning. And you know what? That’s going to produce a better product.
Connect with our experts, access our research and explore our labs, join our platform at WWT.com.