Milo Medin, vice president of Alphabet's access services and AWS18's keynote, spoke to attendees Feb. 22, 2018. Air Force Magazine photo by Mike Tsukamoto.
?Alphabet’s Milo Medin bluntly challenged the common refrain categorizing USAF as risk averse: The Air Force takes far more risk than any commercial entity ever would, he argued.
“This is anything but a risk averse culture,” Medin said Thursday, speaking as AWS18’s keynote. “There’s a price to be paid for lagging behind on innovation, it’s just that the price is not usually paid by the same people who settle for the status quo. That price is going to be paid, nonetheless.”
Medin has been with Google since 2010, most recently working for its parent company Alphabet, Inc., as vice president of its access services. He has focused on increasing the speed at which networks are fielded, and their infrastructural integrity and speeds. Medin is especially peeved with DOD’s requirements-driven strategy, saying it’s “more than inefficient, it’s become dangerous.”
He told attendees there are two digital realms the Air Force must recognize and deal with if it wants to own its future: software and machine learning, a type? of artificial intelligence.
The Air Force fights in a world rife with software, and all software will have bugs, and all bugs must be dealt with. USAF, therefore, must possess a culture allowing airmen to identify bugs and squash them, developing and delivering solutions quickly. But to do so requires a level of communication and transparency between engineer and end user (a pilot, for example), that’s nonexistent within the service.
“Software is never done,” Medin said, who added as example that paying by lines of code is passe in industry, despite it being foundational to DOD’s view on the value of code. Industry has been paying based on needs being met, not work being done, for years, he added. As an example of what rapid capabilities can do for USAF, Medin exemplified AOC Pathfinder, which has been around less than a year and pushed boundaries of software acquisitions and fielding for the service.
When it comes to AI, the real problem is today’s apathy to collecting the data AI will eventually need. The type of AI Medin—who also serves on the Defense Innovation Board—thinks is relevant to USAF isn’t the run-of-the-mill ultra futuristic type, where robots keep humans company and tell them jokes. Rather, it’s humans teaching machines to make decisions without explicit logic coded into them. For example, you can teach a machine to recognize a cat without fully describing one but describing a common denominator of cats by feeding massive amounts of cat photos and massive amounts of non-cat photos into it (think Reddit’s obsession with felines fed into a drive). Though limited since this AI may be able to only recognize a cat but not why it thinks that way, the capability is fundamental for defense, he noted.
“Don’t expect AI to work wonders,” Medin said, adding it doesn’t matter at the moment because DOD doesn’t have nearly as much data as it could, throwing away data from flown missions, for example. “If you’re not a leader in data, you will not be a leader in AI.”
Medin repeated what USAF and DOD leaders have been arguing, evidenced by various rapid acquisitions initiatives, like AOC Pathfinder: “The waterfall process must die.”
When software is never done, defining software needs is paradoxical, he argued. Agile development is a must, especially for DOD.
The pace of technological innovation is “increasing and more disruption is on the way,” Medin said of USAF and its adversaries. “Not tolerating the status quo anymore would be a good start.”