AI Wins Against Humans In Simulated Air To Air Combat In What Air Force Colonel Calls a “Big Blue vs Gary Kasparov” Moment
Hypothetical question: What if future historians consider the big event of 2020 to be the time that AI was proven to be better than humans at Air to Air combat? It has just done that, and this demonstration via simulators will no doubt accelerate AI research in ways that has an impact on a broad range of both military and civilian applications.
First some background: The idea of artificial intelligence engaging in air to air combat is not new. But for as long as I can remember most serious professionals considered it a pipe dream, especially those that wanted a career. This is especially true in the Air Force. Since the Air Force culture is that you must be a pilot to become a senior officer or lead any significant organization, how can you have a service if there are no pilots?
The very few forward thinking military professionals I knew who supported pursuit of concepts of autonomy were not taken seriously. Many were ridiculed. It has been that way for decades. Even after years of advancements in drones and AI and incredible research into the potential of pilotless AI, that remains the dominate view in the Air Force. Academia and some in industry have built simulated AI in the past that humans could not beat, but research like this was discounted by the Air Force. Don’t believe me? Watch the reaction of the audience at the US Air Force at the Air Force Association Air Warfare Symposium on 28 Feb 2020 when Elon Musk tells them that “the fighter jet era has passed, it is drones.” The entire senior leadership of the US Air Force was there. No one cheered. They hemmed, hawed, murmured, laughed.
Now back to the news. DARPA has been funding a multi-year program called Air Combat Evolution (ACE) that includes many thrusts, including a competitive trial called AlphaDogfight. This time the AI was built to simulate flying real USAF aircraft and the simulations were built to exactly mimic the characteristics of real flight. This is simulating reality in ways the Air Force really should NOT discount. Then after a year of competitive trials a final three day competition was held between AI solutions where AI fought AI in dogfights. Then after this, the very best AI pilot was selected. The very best AI, the one who could beat every other AI in a simulated dogfight, was then put up against a human fighter pilot. The pilot, one of the Air Force’s very best, was in it to win it. But even with all his skill and human judgement, he could not keep up. He lost, in what one of the AI engineers and then Air Force Colonel Dan “Animal” Javorsek, DARPA’s PM of the project, called a “Big Blue vs Gary Kasparov moment.” It is not just that the AI won. It won again and again and again. It could not be beat.
What is the potential impact here? This is a demonstration that should reframe arguments and drive decisions. It will still be hard for the Air Force to reframe and look at this in a new way, so the nation may need to force some decisions on the Air Force that it does not want to make itself. This means it may require forceful direction from the Secretary of Defense, White House and the legislative branch. As for the other services, more automation must be in their future too. To not automate is to put our national security at risk, especially if our adversaries field working autonomous combat systems faster than we do.
It is also interesting to think about the potential impact this will have on other domains. There is already a great amount of research and operational experience being gained in AI driven autonomy in space and transportation. This may well accelerate. We should now look for new domains where AI driven autonomy can benefit the nation. Sectors like Agriculture, Mining and Manufacturing seem ripe for increased automation, and all have naysayers that might be more convinced of the potential of automation after seeing the results of this Darpa AlphaDogfight.
This video of this event was streamed live Thursday 20 August 2020. The link below starts at that 4 hour 20 minute mark, right before the AI vs Human segment:
Side note: was great hearing these pilots reference the creator of the OODA Loop concept John Boyd as they narrated these AI engagements. I only wish Boyd was around to see this. I have no doubt he would be encouraging all strategic thinkers to apply appropriate thought to the impact of this news on the military.
- AI Security: Four Things to Focus on Right Now – This is the only security framework we have seen that helps prevent AI issues before they develop
- A Decision-Maker’s Guide to Artificial Intelligence – This plain english overview will give you the insights you need to drive corporate decisions
- When Artificial Intelligence Goes Wrong – By studying issues we can help mitigate them
- Artificial Intelligence for Business Advantage – The reason we use AI in business is to accomplish goals. Here are best practices for doing just that
- The Future of AI Policy is Largely Unwritten – Congressman Will Hurd provides insight on the emerging technologies of AI and Machine Learning.
- Artificial Intelligence Sensemaking: Bringing together our special reports, daily AI news and references to AI from the most reliable sources we know.