Artificial intelligence, machine learning and deep learning are powerful technologies increasingly deployed across many industries. While AI can mimic human intelligence, often exceeding humans in speed, accuracy and capacity, it is entirely dependent on the data that is fed into it.
Bad data — whether simply inaccurate, deliberately misleading or inappropriately interpreted — has led to deeply flawed results. Just as human decisions can be distorted by bad information, automated systems like AI/ML can take erroneous data and come up with a perfectly wrong answer. The cost of poor quality data in the U.S. alone reached an astonishing $3.1 trillion in 2016, according to IBM, which concluded that one out of every three business leaders distrust the data they use to make decisions.
The Department of Defense is a case in point. DOD has been actively engaged in leveraging AI to investigate, learn and apply new advances from industry using troves of DOD data in support of digital transformation. However, when I meet with DOD clients to discuss AI projects, I often discover that what is referred to as AI means something different to the executives than it does to the engineers.
The executive levels of DOD generally support making responsible use of AI with ethical, law-abiding, governed and audited data – measured to fine-tune strategies and standards. A memorandum on plans for ethical AI was recently released by the Deputy Secretary of Defense, Kathleen Hicks. Her memorandum advocated transforming the Defense Department into a data-centric organization with the goal of “improving warfighting performance and creating decision advantage at all echelons from the battlespace to the board room.”
But first, it’s essential that DOD’s customers and suppliers speak a common language – share an understanding of terminology, technology, process, frameworks and methodology.
What, exactly, does a common understanding involve? To help explain the differences between AI, ML and deep learning, I use the following definitions from Microsoft:
Artificial intelligence is any technique that enables computers to mimic human intelligence. It includes machine learning.
Machine learning is a subset of AI that includes techniques that enable machines to improve at tasks with experience. It includes deep learning.
Deep learning is a subset of machine learning based on neural networks that permit a machine to train itself to perform a task.
Starting with this common language, I work with clients to learn their goals and then apply AI, ML or DL to that project to enhance efficiency and reduce cost of the project. How does a program achieve efficiency and reduced cost though use of AI? Applying the system lifecycle of these concepts focuses on the topic of data readiness, a critical component of AI and any data strategy.
“Data is essential to preserving military advantage, supporting our people and serving the public,” Hicks stated in her memo, which also empowers the DOD chief data officer to provide leadership and issue guidance regarding data sharing, data architecture and data lifecycle management.
The first guiding principle of the DOD Data Strategy is to treat data as a strategic asset. The strategy provides a wealth of knowledge directly supporting the vision that “DOD is a data-centric organization that uses data at speed and scale for operational advantage and increased efficiency.”
But why the focus on data readiness? Data readiness drives the outcomes, the analytics or the automated responses depending on the use of AI. Recall that data will drive the path of AI-mentored learning or a semi-mentored ML. As DOD’s Joint Artificial Intelligence Center blog post put it, “AI projects do not succeed without AI data. As the Department of Defense begins to transform itself through AI, creating ‘AI ready’ data will be a key determinant of success.”
About the Author
Frank Manuel is the Solutions Executive, Director, at ARRAY.