Publication

3 Ways to Ethically Innovate in Artificial Intelligence

What happens when machines make decisions?

Every day, artificial intelligence (AI) gets closer to making fantasy reality. Innovations in machine learning, high-performance computing, and more advanced computer reasoning are powering breakthroughs in health and science, helping airlines run more efficiently, and assembling new defensive capabilities to deter security threats at home and abroad.

But we are also rapidly approaching a future where artificial intelligence will be used to suggest whether you are stopped by police, whether you’re hired for a job, the mortgage loan rate you’re offered, or whether you’re admitted to the school of your choice—decisions with real consequences if the tool contains errors or unintended bias. This future reality raises serious ethical and policy concerns that must be addressed head-on.

At Booz Allen Hamilton, we’re laying out a future where advancements in artificial intelligence are shaped by a set of guiding principles borrowed from human subject research—beneficence, justice, and respect. We want to ensure that AI is beneficial, not harmful, to human welfare.

This was the focus of our response to a request for information issued by the White House in June 2016 for the pros, cons and other implications of artificial intelligence. We are calling for an approach where ethics are not simply tacked on at the end, but rather drive the U.S. approach to this experimental new technology.

We see that with the work being done to drive technical development and research for our evolving artificial intelligence capability, work that includes artificial learning and deep learning, quantum computing and more. We believe the potential positive applications for artificial intelligence far outweigh the threats, and by beginning with ethics and safety engineering in mind.

“There’s a huge influx of technology,” says Senior Vice President Young Bang. “We’re just touching the tip of iceberg. How things can be automated and ultimately make life easier for regular people as well as the government, it’s an exciting time for that. But again, there is an ethical dimension to the decisions we need to make. So, instead of pushing, ‘Here are the cool things we can do,’ which we can, we’re pushing ethics associated with AI. Thinking about things differently that way—that’s what I’m really excited about.”

Ethical Innovation Applied

So what does AI look like when applied? Here are some of our projects; this year promises to see even more activity, so stay tuned.

Art by Machines

We taught a computer to paint Picasso style—and not just images, actual live video. Walking up to the Art Mirror is like falling into a painting, the style of which depends on the filter applied. The most popular is probably Van Gogh’s Starry Night; when you look at yourself in the Mirror, you see your own image inside the world summoned by the artist’s brush – blurry swirls and colors. As you move, you move as some kind of fantastic version of yourself living inside this great work of art. You can also capture an image to email to yourself later.

Our developers took a high-performance machine and told it to look at a series of images and paintings—some by famous artists—others general styles, and filters “learn” how to paint that way. It was computationally intensive to do that and took a couple of days of “machine learning” for the system to process all the data. The inspiration for Art Mirror was an open source project by Gene Kogan, an artist and programmer. Our developers kicked their customized code back to the GitHub community so that others can benefit and improve upon it.

Beat Awake

The idea was conceived at Booz Allen’s Summer Games internship, by Michael Jacob, now a consultant, and André Nguyen, now a technologist. The Beat Awake technology pairs up to smartwatches and leverages biometric data to determine anomalies in heart rate, alerting the wearer of the watch when one occurs. The goal is to keep drivers from falling asleep at the wheel. We presented a mobile prototype for an inventive new product at the 2016 Amazon Web Services’ re:Invent conference.

Topcoder Open

We worked with Topcoder, the largest crowdsourcing community in the world, to host the Topcoder Open at our own DC Innovation Center, (now The Helix, Booz Allen's Center for Innovation). This annual event brings together the best coders from across the globe to compete against one another in six different types of competition, including best algorithm, prototype, and user interface design. This year’s Topcoder Open featured 62 competitors representing 26 countries. Six coders walked away winners. Find out who won and more at here.

3 Ways to Ethically Innovate

Adam Porter-Price is a leader in our artificial intelligence work. He and his team have some concrete steps to share on just how to ethically innovate in the field.

Test Models Extensively

You should review with a diverse group: Technologists should seek opinions from a broad community of technical and business or internal professional users to avoid making an embarrassing or dangerous misapplication of artificial intelligence tools. And consider piloting an “inert mode,” running AI tools in pilots parallel to the production environment to compare the results against a human-operated process.

Institute Boundaries

Define easy-to-understand categories of data that are always unacceptable to use. It’s nearly always unacceptable, for example, to include personal health information in a predictive model. This will let both business and technical leaders debate variables using a common language.

Establish a Governance Process

Create a mechanism for overseeing the application of AI tools at the executive level, incorporating review by senior business and technology leaders to oversee privacy, security, ethics, assumptions, input data, and operations issues. This group should constantly view review results from AI initiatives and post-mortem pilot programs, and should help promote a common understanding of AI tools and concepts across the organization.

The ethics issues we’ve seen to date in AI have largely been caused by accidents and poor application of techniques, rather than deliberate actions. Doing these three things will substantially reduce the chance of mistakes, and will provide a mechanism for spreading this increasingly useful technology across your enterprise. Machines may be increasing in ability daily, but they will still need to work closely with ingenious humans to answer tough questions over the next century.

3 Ways to Ethically Innovate in Artificial Intelligence