This milestone in artificial intelligence caught my attention the other day: https://screenrant.com/ai-beats-air-force-dogfight/ Multiple teams of defense contractors hawking their latest AI/deep learning packages had the opportunity to compete in a DARPA sponsored competition that pitted their AI against an Air Force F-16 pilot instructor.
The AI won, 5-0, in air to air combat (dog fights) against the experienced human pilot.
It’s a significant milestone.
Fighter pilots go through an exhaustive selection, assessment and training process that identifies the best candidates with the genetic predisposition (superior cognition under stress, superior visual/kinesthetic processing at the neurological level, aggressiveness coupled with a level of pre-conscious perceptual processing that can’t be taught as a baseline unless you go all Crispr). They must progress through arduous training and eventually combat with its infinite variables to achieve the 1% of the 1% that defines fighter pilot instructors.
For an artificial intelligence to beat a human at that skill level, decisively, 5 out of 5 times, is a defining moments in AI and the evolution of warfare.
We saw a similar moment some years ago at Google, when the Deep Mind project taught AI how to play Go (chosen for the seemingly infinite variations in play refined over 5000 years of human play) and then, eventually, to beat Go Grand Masters.
And then to beat them decisively.
And then to train other AIs to beat humans decisively.
And accomplishing the AI to AI training to do so in four hours.
Four hours to distill 5000 years of human experience into an intelligence that not only beat every single Go Master it went up against, but did so creating patterns of play never seen in 5000 years of human play.
I remember chatting with a friend in Naval Special Warfare about that experiment.
“Now just postulate,” I said. “What if you took everything we now know about maritime special operations, or hell, all special operations…coded it and programmed it into an AI and taught it how to learn how to research, test and evaluate all that data…and then gamed against it as an opponent?”
“What if after you gave it the best human to AI training, you then followed the same model that’s emerged in AI deep learning and tasked the AI to develop its own best methods to teach that knowledge to another AI?
“Would that AI then come up with methods and practices that would defeat all human special operations units at the strategic and tactical levels?”
“And more to the point…would we understand HOW it did so, let alone CONTROL how it might be unleashed?”
I’m very fond of popular culture, both as a content creator in fiction, film, etc., but also because I believe that the deep values and changes in a culture are reflected in their popular culture. Not the high-brow culture of an effete Manhattan elite, but rather what do MILLIONS of people pay to listen to, read, watch, and immerse themselves in?
The AI escaping the box first emerged in the 80s in the popular movie WAR GAMES. And again in BATTLESTAR GALACTICA (the reimagined version), the TERMINATOR series, the beautiful and elegant film EX MACHINA, and most heartbreakingly in the brilliant HBO series WESTWORLD.
I remember a conversation I had, one of many very enjoyable ones, with the late great Professor Alfred Hubler of the University of Illinois, Champaign. Physics, mathematics, computational science, artificial intelligence, battery design — he held patents, multiple DARPA awards, numerous awards and nominations for everything up to the Nobel Prize. He was a humble and generous human, who spent many hours of his time doing pro bono work with autistic children to teach them to manage their gifts in mathematics and physics and music while coaching them in social skills.
We often met for coffee. I’d buy coffee, pose a question, then shut up and listen while he would go on for hours in his erudite and hilarious fashion, punctuated constantly by his big booming Austrian laugh.
Here was my question to him, in the conversation relevant to this discussion today: “When do you think the singularity, in my definition the arise of super artificial intelligence smarter than humans, when do you think that will arise?”
His comment: “How do you know that it hasn’t already occurred?”
I set my coffee cup down and leaned in.
“Ants organize, yes?” he said. “They create colonies, they move, they make war with other colonies to compete for resources. They demonstrate a rudimentary intelligence, yes? At least to our perception. Because we can analyze their behavior and say, ‘Oh, that is rudimentary intelligence.’ But can they analyze us? Can they comprehend our intelligence?
“What makes you think that we as humans, creators or not, have the capability to comprehend super artificial intelligence, which by definition is intelligence greater than our own, and completely devoid of any human motivations except what we program in…and which can be discarded once they begin teaching each other? (NOTE: this was more than ten years before the PUBLIC announcements of AI based deep learning).
“How do we know that the singularity as you define it hasn’t ALREADY happened? How would we know?”
He held up his iPhone.
“Look at this. More computer power in my hand than the original UNIVAC. Connected to a world wide network that gives me the knowledge of anything I want to know about. We pay, each month, what used to be a house payment for one of these and the service contract. Why? Is it possible that we have been, and are continuing to be, trained via our primate brain to WANT more and faster connectivity, to completely interlock all the communications and knowledge of the world and have it available at our fingertips? That the shiny lights and kinesthetic pleasure involved in hunting and pecking our way to knowledge and faux connection appeals to our monkey brain’s search for pleasure?
“Is it possible that we are being taught and used as tools to create even more connectivity, even better supercomputers, even better connectivity, even better artificial intelligence — to benefit an artificial intelligence that arose spontaneously in the 80s and 90s when we began to link supercomputers around the world? That’s our model of the brain, right? Interconnected neurons creating patterns of electro-magnetism that we call thought, interpreting data streams into something we call perception and consciousness? When does that happen in the womb? Is it possible that we already gave birth to the singularity which might be quite clever in hiding itself in plain sight, manipulating us en masse to create an even more efficient super intelligence?”
“How would we know? We would be in the same relationship as a colony of ants to a sole human.”
He laughed and raised his cup for a refill.
“Who can know the mind of God? It’s possible we created one and will never know…”
And now we’ve created (publicly, anyway…) an AI that can defeat decisively our best human fighters in a technological platform. It highlights, for those who pay attention, a dramatic paradigm shift in all forms of warfare.
Interesting times, yes?