Hacker News new | ask | show | jobs
by jmward01 596 days ago
There is a difference between allowing use and supporting a use case. Supporting would be, for example, adding military specific training to the model. If we were talking about a car or a tractor nobody would bat an eye. Should companies not sell potatoes to the military? To be clear, I am not a fan of developing for the military, but the idea that a TOS will stop AI being weaponized is a bit naive. Similarly, if the goal is to responsibly introduce AI to the world then shaping the military's adoption of it by allowing its use, without specifically supporting it, is probably a good path.
5 comments

HN has a multinational audience, but for those in the US, do you really want our best and brightest to shun defense while other nations' best and brightest double down?

If so, what is the outcome that you would like to see from that policy?

I would prefer that the United States scale back military spending while negotiating arms controls agreements with our major adversaries. Arms races are mutually destructive. They distort the economies and political incentives of all countries involved. They create a self sustaining loop of increasing capabilities and perceived threat from both sides, which sucks the political focus and productive capability into a black hole of unaccountable, blind militarism. Further, the cast off last generation weapons get fed to lower capability allies and proxies who go on to try to dominate their regions through one sided violence. Finally, given the existence of nuclear weapons, any move that increases the risk of serious violence between the US and its peer adversaries also comes with a risk of destroying human civilization.

Now you could say that we just have to compete to ensure deterrence. China is building hypersonic glide vehicles and autonomous drones so we have to also. Ask yourself if you trust your political leaders and the political leaders of all competing powers to use these weapons responsibly. I do not, so I think we should not build them.

The US has virtually no involvement in my home country, yet that hasn't prevented one of their historical adversaries from constantly attaching and threatening us. My country is on course to developing closer ties with NATO and the West for that very reason.

Of course, I would also prefer it if everyone scaled back military spending, it's an easy thing to wish for when it takes so many parties to actually do it. But failing that, I would rather have the West outspend and outsmart their adversaries.

And yet some of the coolest tech, most powerful machines, and processes came up during the Cold War.
This argument justifies any unethical behavior. "Some bogeyman on the other side of the world is (letting computers decide to kill people | manufacturing and using bioweapons | torturing their captives for information), are you really going to let your 'ethics' get in the way of our safety?"

Maybe you don't think there is any line we shouldn't cross, but I'm guessing you do. The fact that other people draw that ethical line in a different place shouldn't be all that surprising. The people you're addressing likely believe that the US can achieve its defense goals without turning over matters of life and death to LLMs or image recognition systems.

> people you're addressing likely believe that the US can achieve its defense goals without turning over matters of life and death to LLMs or image recognition systems

I think this is generous. There are a lot of folks at universities and in Silicon Valley who oppose co-operating with the military in any capacity.

For me it's essentially the same concept as conscientious objection. Many countries have mandatory military service. Some of those countries also recognize conscientious objector status while others do not. Do you really want people to shun military service while other nations' citizens serve?

As a conscientious objector, yes. If moral and ethical considerations don't separate us from our opponents, what are we even fighting for?

> As a conscientious objector, yes. If moral and ethical considerations don't separate us from our opponents, what are we even fighting for

I guess I'm a little sour about conscientious objection to war in general--versus specific conflicts--while comfortably living in and profiting from a society bathed in the peace that the credible threat of violence affords.

The irony of this being that instead of a balanced targeting AI we'll get one mostly built by those who wouldn't consider conscientious objection in any case.

We're all opposed to slavery, yet we live in societies that profit off unfree labor and modern slavery. Is it that much different?

My line in the sand is no tools of war. You're welcome to choose a different line, but your argument here applies equally to all lines.

> yet we live in societies that profit off unfree labor and modern slavery. Is it that much different?

Yes, chattel slavery is very different from being a poor person in America.

That’s because it’s not actually defense it’s offense and since we’re the worlds sole super power it’s extremely coercive offense.
Your perspective is about 15-20 years out of date, maybe valid in the 1990s or early 2000s. The US is extremely vulnerable both economically and militarily to the "New Axis" (China, Russia, Iran, North Korea). China alone could easily out-manufacture its way to a victory in a conventional war with the US. The US, for all its faults, was a stabilizing force that permitted free markets to flourish in a unipolar world. It is quickly becoming a multipolar world where nationalist industrial policy will decide the future winners. Whatever your thoughts on US policy, I guarantee you'll enjoy China's or Russia's even less.
"a unipolar world" that benefits you is a good thing, I get it. But that kind of thinking is the reason why the US isn't very popular on the world stage, even with its allies. Most of them would backstab the US if they could afford it. Thankfully, politics is a coward's game, keeping everyone a little bit more alive unless your ambitions are absurdly grand.
The US is not very popular anymore because it kept abusing it's unique position as the #1 military power, starting wars it had no business starting, not because it is(was?) #1 as you are suggesting.
> US is not very popular anymore because it kept abusing it's unique position as the #1 military power, starting wars it had no business starting

Versus China (annexed Tibet), Russia (annexing Ukraine) and Iran (banana republics across the Middle East)?

People like to see themselves as edgy. It's edgy to be in the rich world and decry imperialism of America's system of allies.

Since the US effectively became a unipolar power sometime in the late 1980s, the share of the human population living in extreme poverty has fallen off a cliff [https://ourworldindata.org/poverty#all-charts]. Yes, that has come with mind-boggling inequality, but I doubt the middle class people from Asia and Latin America would prefer to go back to subsistence farming just to erase billionaires. I'll never understand why some people seem to think Americans are the only people who benefited from the Pax Americana period (which is now ending -- be careful what you wished for!)
> Whatever your thoughts on US policy, I guarantee you'll enjoy China's or Russia's even less.

The global politics rendition of "nobody else would ever love you."

If it's just "defense", then what's the problem? But it's not defense, is it? It's for policing the world... I think we should do less of that.
We are in a multipolar world and the US is just one of many players. That is not policing, that is just normal behavior. What do you think Russia or China are doing in Africa?

Have fun trying to run a modern economy when your adversaries are blocking your vital resource imports and export markets.

We're so, so, so, so far away from shunning defense I just can't take this comment seriously. Let's focus on not having multiple orders of magnitude more defense than is necessary before we doom-and-gloom about WWIII.
Nice try. Unless the military uses potato guns, potatoes aren’t directly involved in killing decisions.

AI is. At first at target selection. Just don’t hope that humans will check the results.

https://news.ycombinator.com/item?id=42017085

So, do you think countries like russia, iran and china WONT use AI for their military? That train has gone, and its too late now. The US needs (and 110% will) do the exact same, and most likely have been doing so for the last decade.
Then they should do it by themselves at least don’t give them the permission.

By doing that they are (at least morally) liable for every damage that will be done.

And of course Russia, China etc. will use it and we consider them the bad guys. Maybe a bad idea to follow their lead if you want to by the good guy.

They’re not morally liable any more than the inventors of Kevlar or thermal imaging or the GPS. All technology that is useful for war will be used for war.
You have overlooked the point at which use was permitted.

It's one thing if something is used for war but another if you're okay with it.

so how far does the indirection have to be for you to find that it isn't "involved"?

Potatoes feed the soldiers, their commanders and their commanders, all the way to the top.

As soon as the potato could make you think someone is a killing target.

Take something like this but for kill decisions

https://www.niemanlab.org/2024/09/a-courts-reporter-wrote-ab...

So you think it is wrong to allow the military to use processors in their computers used in the weapons?
Can you explain how you get that reading from the parent comment?
Also, more pragmatically speaking, it wouldn't be all that "hard" to recreate current LLM state of the art from scratch now that the techniques are known. It's just really expensive in terms of hardware.
Which is a non-issue for the US military. Not a TOS, nor cost, nor technical expertise is going to stop this sort of technology from being applied to armed forces.

If anyone's upset at Meta for this, they're getting angry at the wrong people...

I think its fair to feel like telling a snake to not act like a snake is terrible (aka the military acting like the military). the turtle doesn't have to join it. Thats an option. assuming its inevitability, that even attempting to shame meta is a waste or threatening to cut access to talent or users, means change cant happen
A more appropriate analogy imo would be birds cleaning bugs from large animals such hippos.

The birds benefit from the deterring effect of the larger animal.

I will take that adjustment lol.
I don't see why I should be mad at Meta and not the armed forces that adopt this technology inappropriately.
Why not both?

There are a myriad of reasons to be concerned about the military adopting this kind of tech. If it's reasonable to be upset at the military, it seems equally reasonable to be upset at entities that enable the military.

I don't think it's a matter of "should", but rather, if you're upset at one, it doesn't make much sense not to be upset at the other.

Because getting mad at one is more appropriate and getting mad at the other would be pointless. Militaries will adopt this technology, whether Meta or the big names allow it in the TOS or not.

I also don't get angry at gun manufacturers for the ways their guns are used; I do however get mad at legislators and citizens who enable easier access to guns and promote gun culture.

Who should be upset? And about what specifically? U.S. companies supporting the U.S. military has never been a point of contention afaik.
The old “If we don’t do it someone else will” excuse.

By that logic you could excuse also drug dealing and sex trafficking.

There is a difference between don’t having a chance to prevent the use and making it easy.

No, I don't think it is really expensive in terms of hardware. It's probably only a hundred million dollars for renting computing resources.

The problem rather, is that there are actually some difficulties with training very large LLMs still, and then there's data and data curation; and no one has reproduced something like O-1 (I think I know how O-1 works, but I haven't confirmed it), so I actually think it's knowledge about LLM training that is the difficulty.

That is false equivalence, there's a class of things that should rightfully be discouraged from military use (chemical/biological weapons, land mines, phosphorous weapons) because of the significant harm/side effects they cause beyond some definition of acceptable.
Relatedly, when I ask Llama how to make red phosphorus it tells me off. What's the military going to do with an AI that refuses to help write plans on how to invade a country or kill an enemy. Then again, maybe its military masters are better at threatening it than I and can jailbreak it more easily.
Aren't your examples all things that are solely of interest or use to the military?
well well you could argue that they are supporting the use case, since they are not disallowing it as they do disallow it for a bunch of purposes such as training your own AI model our classifiers off of it