US Air Force wants $6B to build 2k AI-powered drones | Hacker News Reader

US Air Force wants $6B to build 2k AI-powered drones

theregister.com

29 pointsBender24 comments

That's the problem with the recent overloading of the term "AI", I have no clue what this means. Are we talking about building and training an ML model? Did some general use chatGPT and told their assistant "I need this to pilot a drone"?

This is an interesting - albeit terrorizing - concept, but I have trouble understanding what it means in practice

Well for the money I would hope they at least come close to the drones from the movie Oblivion [1]. Even better if they can communicate out of band using the same sounds for psychological effect. Those drones could work together as a team those scenes missing from the linked clip. I hope that is the AI they are going for.

[1] - https://www.youtube.com/watch?v=rEby9OkePpg [video][3 min 33 sec]

Reading the article, it makes me think that they are making a real life aimbot for the Air Force.

It is getting silly though. Makes conversations with less than tech savvy people a little difficult lately since it's such a hot topic

Yes its a ML model.

> One of the U.S. Air Force's stealthy XQ-58A Valkyrie drones recently completed a successful test flight demonstrating the ability to carry out aerial combat tasks autonomously using new artificial intelligence-driven software. The service says the test is part of a tiered approach to maturing autonomy "agents," which involves training algorithms millions of times first in simulations and other testing. This includes the Collaborative Combat Aircraft program, or CCA, a key part of the larger Next Generation Air Dominance modernization initiative.

https://www.thedrive.com/the-war-zone/xq-58-valkyrie-solves-...

Somewhat related, Google has been using LLMs to train robots: https://www.theverge.com/2023/7/28/23811109/google-smart-rob...

I think it's a combination of ML to steer the craft and coordinate the assets. Eventually there will be something a bit like DeepMind's StarCraft AI. Probably with a text interface. They might even get that technology from DeepMind.

They are talking about building 2,000 drones, 30ft long. That they will be piloted by AI is interesting, but not as remarkable as the fact that they only cost 15% of what a predator drone costs, and will be the cheapest drone of this type the Air Force fields by at least a factor of 3.

There are no plans to arm them (yet), instead to make them do recon autonomously and report the results to humans. And the government is currently saying that when they do carry weapons, a human will have to authorize every use.

The US national defense budget is around $1 trillion per year ($800B in 2023). So this is about 3 days worth of spending.

If you've ever wondered the logistics of spending $2B a day is like, here's the daily contracts summary page [1]. It'll blow your mind. If you only had a vague notion of what the "military industrial complex" really is, this will bring it home.

1. https://www.defense.gov/News/Contracts/

And here we go with the autonomous killer drone swarms, what could possibly go wrong?

Good, let's perfect them before our enemies and prevent war over Taiwan and Ukr.

I get the arms-race angle and I know these developments are inevitable, but that doesn't make it any less of a dystopian nightmare. In the short-term it's very beneficial to be the first to develop this new paradigm-shifting arms tech so of course we will, if we don't someone else will. But in the long-term I think autonomous militaries are a one-way ticket to an autocratic surveillance state.

A soldier who is ordered to commit an atrocity or oppress their own citizens might disobey or rebel, an army of machines has no such problems. I'm less afraid of the 'self-aware AI' skynet takeover scenario than I am of putting so much military and technological might into the hands of so few real living people. When it becomes technically impossible for the military of an entire nation to disobey any order its given, no matter how terrible, I believe we will see man-made horrors beyond our comprehension. And it won't be because the machines became self-aware, they'll just be following orders.

Autocratic surveillance states are here in the form of countries like China, Russia, NK, etc. The US is clearly the leader against these types of dystopias (despite it's many past mistakes and current faults), so I think to lump them all together would be naive and dangerous.

I don't take the parent to be lumping them in with other countries. He is mearly musing that such tecnhology with incress the likelyhood of such form of goverment and will cement the positions of autocratic regimes.

Well I think a truly democratic country like the US (even though imperfect) getting these technologies first and isolating/defeating the autocratic countries would actually lead to some form of self-governance over the use of these technologies.

This would avoid the dystopian nightmare he worries about.

The US is only against them due to a perceived threat to global US hegemony. For instance, you forgot to list Saudi Arabia there, and they are one of our allies.

The US govt has been the largest exporter of misery and destruction the last 80 years. So yeah, you can lump them altogether. And yes, the US would happily use AI to exterminate 500k communists, 1 million Iraqis, or 1 million Vietnamese again.

"The US is only against them due to a perceived threat to global US hegemony. For instance, you forgot to list Saudi Arabia there, and they are one of our allies."

-- The US eventually wants democracy globally. Support of countries like SA etc are just stepping stones and temporary accepting.

"The US govt has been the largest exporter of misery and destruction the last 80 years. So yeah, you can lump them altogether. And yes, the US would happily use AI to exterminate 500k communists, 1 million Iraqis, or 1 million Vietnamese again."

-- This trite talking point is overused and mostly wrong. Since the US took partial hegemony in 1950s and complete hegemony in 1990s... things have improved greatly for the world as a whole. Despite many mistakes by the US, it is still pretty good. Do you think the alternatives of Nzi Germany (self-explanatory), USSR (millions killed by Stalin), or the CCP (millions dead, famines etc) would be better?

When has the US gov specifically targeted non-combatants for extermination? I do not believe that ever happened. Collateral damage due to military operations is terrible, but it's not willingly done.

So basically your answer is complete hyperbole, as well as mostly wrong.

So, you endup with best case 3.000.000 USD/unit drones. This for 2.000 drones, not considering budget overuns. Both sides in Ukraine use more than this in a single month. Again the obession with flashy technology in the US military ignores the reality of what is a large scale peer conflict. Those drone maybe great, but their cost and the possible manufacture complexity make it not very practical for large scale peer conflicts.

The unit cost goes down with scale. A Honda Civic would be extremely expensive if they only made 2000 of them. That doesn't mean it's too flashy for the job, it just means they haven't divided the development cost over enough units.

It makes sense to develop new military tech continuously, but only scale out production when needed.

The drones mentioned in the article is not the consumer drone (like DJI drones) used in Ukraine.

These drones are for accompanying new fighter planes. They aren't super expensive ground attack drones, which we already have in cruise missiles, but really cheap fighters. Peer conflicts is the only reason to have them.

Anyone know what the flight time is for one of these little Kratos hunter killers are?

Let them use the damn money they already have, over bloated budgets already. They don't need 6 billion for a drone. Come on that's ridiculous.

https://en.wikipedia.org/wiki/Skynet_(Terminator)

Wrong one. https://wikipedia.org/wiki/SKYNET_(surveillance_program)

Reply on news.ycombinator.com