Hacker News

This is how AI will solve the paperclip problem.

I read this in Linda Hamiltons voice

Interesting coincidence that both this article and the Hugging Face "security incident" post were on July 16. These will be mandatory reading in future Skynet History classes, assuming we survive the war.

No worries, these will all be controlled by AIs sitting in orbiting data centers part of the Starlink constellation.

And in any case, no one is building humanoid robots.

I find it quite sombre to see this comment from a peson with that handle.

Ha - this was the first thing that came to mind. I read this in Arnold's voice.

ref: https://youtu.be/4DQsG3TKQ0I?t=29

These are going to be experienced test pilots who are pretty good at actively managing autopilots and other automated systems. Plus they'll be backed up by telemetry to ground controllers. But ultimately the military is willing to accept some losses in order to advance key R&D programs.

> In the first Iraq war, they went in and won. In the second one, it took years to sort of occupy the country.

The scope of the first Iraq war was to force them out of Kuwait. That is very different then the invasion of 2003. If you just think about it this should be obvious.

Oh what a great CIA campaign that would be to give the impression that U.S. military isn’t up to par - to motivate and trick our enemies into making stupid moves while we keep our true cards hidden.

Such a crazy thought.

> In the first Iraq war, they went in and won. In the second one, it took years to sort of occupy the country.

The objectives for the two wars were incredibly different, making this an apples to oranges comparison. In the first war, the objective was to push Iraqi troops out of Kuwait (achieved) and then to contain the Iraqi military within Iraq (achieved). In the second war, the objectives were much broader and much harder to achieve. They achieved some objectives (toppling the existing government, defeating the existing military), and then utterly failed in nation building and got into a protracted conflict with state and non-state sponsored military groups.

https://en.wikipedia.org/wiki/The_Gulf_War_Did_Not_Take_Plac...

Bush Sr knew to keep the foreign-compromised media out of the war and so accomplished the objectives of the war swiftly. Bush Jr had to deal with the media second guessing everything.

> The article smells of AI psychosis...

China has been doing something similar since 2020 [0] and something we've experimented [1] on for decades [2][3].

[0] - https://www.defensenews.com/global/asia-pacific/2021/10/20/c...

[1] - https://www.af.mil/News/Article-Display/Article/467196/first...

[2] - https://www.af.mil/About-Us/Fact-Sheets/Display/Article/1045...

[3] - https://commons.wikimedia.org/wiki/File:QF-4E_GulfofMexico_1...

Drones really changed the equation more than Covid. Autonomous F-16's already can win against all human pilots. This would have been big 20 years ago. Now the cheap drones make it not the biggest deal anymore.

What are you talking about, Iraq fell in a month not years.

Don't tread on ... my gpu

Yes-ish.

There's a lot of different and overlapping programs in play here. I do not believe that autonomous F-16s are intended to be operational assets. That said...

VENOM - program that shows you can bolt on autonomous flight control into a standard F-16. Previously, the X-62A (a one-off prototype based F-16) demonstrated autonomous dogfighting back in 2023.

VENOM provides USAF/DARPA with a lot of high performance autonomous platforms (bolt-on F-16s) that they can then use to develop and validate software, tactics and training for actually fighting with autonomous systems. This would presumably be critical for getting the most of the CCA (aka loyal wingman) programs currently in development.

The AIR program referenced is basically "how do we get autonomous systems to control and fight effectively as a team (presumably arbitrary mixture of manned and unmanned systems) in a beyond visual range fight".

The ACE program referenced was basically "how do we get autonomous platforms to dogfight" with related questions like "how can we augment the human pilot in a dogfight - ie, how can we hand off the actual flying and execution of manoeuvres, and using all of the sensors to autonomous systems, and let the pilot choose actions, instead of hang flying the actions".

I'm sure there will be a next generation of jet drones that are cheaper, but the autonomy kit here is probably still cheaper than new jet drones.

I think most fully autonomous jet drone programs are also designed to work with a main crewed fighter jet that coordinates them, so there is also a benefit in getting the fighter jet to be able to fly better than a pilot.

More like it'll fly itself while the pilot engages in other tasks.

Every pound of weapons officer/navigator is a pound you can't put on target all else being equal.

Drones have human operators. This is a murderbot.

One Sharon Apple concert before the end!

The US military has thousands of old airframes, many of which can be made flyable with very short turnaround. They already use the F-16 as test targets.

Taking an obsolete airframe that is already paid for and adding AI to it is much cheaper than building a completely new AI-native airframe from scratch. This is finding a new use for old aircraft. You can send a bunch of old AI-driven F-16s on high-risk missions without worrying about a human pilot or cost. The airframes may be relatively obsolete but can still do plenty of damage on their day.

The new 6th generation aircraft are all being designed as "manned optional". That is probably the last generation with affordances for a pilot.

> when it would be better to build craft spec'd for AI flight without human life support systems.

Yeah this. https://en.wikipedia.org/wiki/Anduril_FQ-44

The "see also" section lists several competitors.

I guess if you already have a load of F-16s it might make sense, but I think that's a stretch.

This whole AI hype is such a convenient cover for the decisions of the 0.001%

That's not precisely how I remember that quote... But it's on the Internet, so my memory must be faulty.

With apologies, I don't particularly enjoy when people do this, and I don't think it adds a lot, but I want to leave this breadcrumb, I wrote a post on this a couple days ago here, on an anti-AI screed submission, saying this anti-politic, this accountability denial, is very much the point.

Taking this quote and seeing exactly that potential for abuse seems very very much like the point. Strong agree, strongly against leaving non-seeing non-thinking alien intelligences with no accountability and proprietary programming to judge us. https://news.ycombinator.com/item?id=48965083

> Is there any line where the public will become deeply concerned?

Is someone, somewhere using the wrong bathroom? Or is the wrong colored person crossing an invisible line in the desert? Those seem to be the only issues that reliably get the (American) public to the polls.

Dogfights aren't really a thing for a modern military. What the AI enables is much higher endurance than is feasible with a human in the loop. They planes can stay flying as long as they can be refueled, which is also going AI.

Are dogfights still relevant? Isn't it mostly shooting missiles outside of visual range these days?

These systems still require a human to command the jet, they jsut don't have to be inside them.

"This critical testing will pave the way for human pilots to seamlessly command and orchestrate teams of autonomous, uncrewed aircraft."

The reason they are doing this is because the Military is already building a bunch of autonomous drones and fighter jets that already run on this system so they are retrofitting existing equipment with the same system so they can all fly together.

"Ultimately, the capabilities advanced under AIR will enable a wide variety of future joint force operations, including Collaborative Combat Aircraft programs."

Which are these:

https://en.wikipedia.org/wiki/Manned-unmanned_teaming

And everyone is developing these.

> I'm sure they will soon find that AI pilots defeat human ones in dogfights 100% of the time,

Unfortunately you're wrong. It's not soon. It's 6 years ago.

https://en.wikipedia.org/wiki/DARPA_AlphaDogfight

>Just in the last month we have * AI that goes rogue and hacks billion-dollar companies

Except this never happened, it's a piece of marketing. It never "went rogue" but followed an intentional prompt (exploit-related benchmark), with hardly surprising results as the capabilities of this model were well known.

>AI controlled autonomous drones in Ukraine and F-16s in America

An entirely different kind of AI, not huge generalist models. Neural networks are being experimented with in sensors and control systems since the 80s and are widely used in everything since AlexNet, not sure what's new here, especially "this month" new.

does this mean we'll have preachy anti-war lectures while fighting them, too?

It's a step on from the F-16 flight simulator AI competition in 2020: https://en.wikipedia.org/wiki/DARPA_AlphaDogfight

As another poster already said, building new F-16s would be a waste of resources, but using the old existing F-16s by automating them is likely to be a better use of the existing resources, instead of scraping them. This also allows to gain experience that might be useful for designing the future unmanned airframes.

Fighter jets don't need 100%. In a war, we accept a much higher rate of casualties than we do in day to day driving. I would also imagine that while planes are harder for a person to pilot, they are easier for a computer. Most of what makes them hard is mechanical, not dealing with unexpected situations.

There are many ways that flying is an easier solve than driving, believe it or not.

I reject the notion that planes are harder to automate.

Planes might be easier, large empty space in the sky