Our Walter Bradley Center director Robert J. Marks is back with Jonathan Bartlett and Eric Holloway, assessing their Top Ten real advances (“Smash Hits”) in AI in 2020. Readers may recall that we offered a fun series during the holidays about the oopses and ums and ers in the discipline (typically hyped by uncritical sources). So now we celebrate the real achievements. Our nerds think that #6 is what happened when AI won all the aerial dogfights in a simulated closed-world contest that was closely watched:
According to Javorsek, fighter pilots need to be convinced of the capabilities and utility of AI in combat beyond current practice. DARPA’s AlphaDogfight is a step in this direction.
By posing relevant questions, DARPA’s overall AI strategy accurately embraces both the capabilities and limitations of AI. How can AI enhance the performance of the pilot by lessening cognitive load? How can companion unmanned aircraft accompanying the fighter pilot’s fighter jet be effectively used? What are the limitations and dangers of autonomous drones?Robert J. Marks, “After Thursday’s Dogfight, It’s Clear: Darpa Gets AI Right” at Mind Matters News (August 23, 2020)
Well now, Robert J. Marks and air force captain Eric Holloway talk about how incorporating AI will change aerial warfare:
Our story begins at 22:35. Here’s a partial transcript. (Show Notes and Additional Resources follow, along with a link to the complete transcript.)
Robert J. Marks: AI meets fighter pilot. Eric, what in AI is going on here?
Eric Holloway: Fighter pilots like to claim they’re irreplaceable and so on, but I think they’re actually pretty replaceable by a decent AI. For nothing else than the fact that AIs are not humans and you don’t have all the constraints of a human body. AI can pull like 20 gs and not break a sweat whereas a fighter pilot’s going to pass out at around 9 gs. So just in pure performance an AI is going to be able to do things that a fighter pilot can not.
Note: 20 gs is a unit of acceleration.
Also the parameter space is a lot simpler than, say, what you’re dealing with as a soldier on the ground navigating a dense urban situation or a deep jungle. In air you’re just flying through air high above the clouds … it’s a pretty simple environment. You’re moving in three dimensions, but that’s really not too tricky for different various algorithms to figure out. And you don’t even need a very fancy AI for this. It is using probably control systems that have been well studied for the past couple of decades or so, probably since even before the ’50s. So this is going to be a kind of AI that can rely on really well established control systems and do things that fighter pilots can not. So I definitely think this is the way to go for the Air Force.
Robert J. Marks: That is going to be very interesting. And you’re right about the ego of fighter pilots. I’ve also found this in other professions, I think that heart surgeons really have incredible egos. And as far as fighter pilots, if you watch Top Gun, you see big egos from Tom Cruise and Val Kilmer, and this isn’t far from the truth.
Jonathan Bartlett: Your ego is writing checks your body can’t cash.
Eric Holloway: Essentially once you have AI fighter pilot or a fighter jet, you just have a really smart missile. And I think you can also probably scale that down and get really smart bullets and stuff like that. So I think you have a lot of interesting possibilities out of this kind of narrowly constrained smartness in the weapons.
Robert J. Marks: Well the fascinating part is, as you mentioned, Eric, is that if we do come up with these AI sort of fighters, that they are not constrained by human weaknesses, they can stand these high G forces and can do things a lot quicker.
Eric Holloway:Right, just in general I think the military can rely a whole lot more on robots and AI because robots can go many places that humans cannot go and plus you don’t lose human lives. Hopefully in the future all of our wars are just going to be giant robots fighting each other, like in Japanese cartoons driven by eight year olds and 13 year olds.
Robert J. Marks: Okay, well, I hope not, I hope not. But one of the things about this, I watched the… They did this AI beating the fighter pilot in real time, they streamed it. And I watched part of that, and of course the AI did beat the fighter pilot, but much of this was meant to be psychological. It was supposed to be a gut punch to the egos of the fighter pilots to make them realize that AI was going to be viable. So I don’t think this is something which is going to be reduced to practice tomorrow, but I think that DARPA did get it right in taking these first steps to make sure that the fighter pilots understood the power and the utility of artificial intelligence.
Eric Holloway: Yeah, now one thing I did see… Yeah, I watched the whole casting of the fighter pilot fighting the AI, but there was a qualitative difference between the two techniques. So the AI was just really good at optimizing really tactical moves, it could take tighter turns and follow the pilots, but it never innovated. It was pretty much just following and dogging the pilot until it got in for the kill shot. Now the pilot on the other hand, he was always coming up with new innovative moves. So there is an aspect still in which the AI is still very different in terms of fighting than the pilot. And so once the AI got it locked into the very small area that it knew how to get the kill shot in, then the pilot was toast. But if the pilot is able to broaden out the domain he’s fighting in so there’s a lot more options and he can take a longer term strategy, there might still be ways that pilots can defeat AIs.
Robert J. Marks:This is part of the history of warfare. If you have an enemy AI fighter pilot, they are such that you can probably after a while predict the limitations of their movements, then you can gain them to defeat them.
So it gets back to the idea of the creativity of AI. So you need the creative human aspect in there in order to win the day.
Eric Holloway: Right. And I think the bigger vision for this is not just we’ll replace all our fighter pilots with a bunch of AI pilots, but it’s going to be more a hybrid approach where you have the fighter pilot, and then he has a bunch of robot wingman that he can control.
Note: Readers may wish to look at The Case for Killer Robots by Robert J. Marks for more thoughts on these topics.
Here are the Smash Hits to date:
7 AI Smash Hit: Why AI can’t do your thinking for you. Robert J. Marks: you change a pixel or two in an image and the deep convolutional neural network is totally wrong. Eric Holloway: The machine’s confidence in its result is complete certainty and it’s absolutely certain about the wrong result.
8 AI 2020 Smash Hit: Big gains in practical self-driving cars. The people who have been pursuing Level Five self-driving are nowhere but Level Four is working well. Jonathan Bartlett: You can think of Level Four self-driving as an engineering project and Level Five as a philosophy project
9 AI Success: Smarter cars for non-millionaires If your car is a recent model, an affordable aftermarket kit might transform it into a much smarter car. One possible risk is that a hacker could take over your car but, no matter what we do with AI, we must deal with security issues.
10 Smash Hit: #10 AI Success!: Translation gets faster and better. Machine translation, properly used, can help us communicate better. What’s made AI tech translation work so well is not that it’s perfect, but we’re going to have a second pass.