Hacker News Reader

Comment by Waterluvian | Hacker News Reader

I have a feeling that humans will remain overwhelmingly responsible for most of the wartime mistakes.

I think this will further reduce conflict between advanced nations (robots killing robots just becomes a stalemate) and will intensely escalate any asymmetric conflicts where one advanced nation can crush whoever without having to pay the price of deploying humans (think CIA drone strikes).

View on HN

I'd say more countries will choose a nuclear route over being droned to death at near zero cost to attacker.

Atomic weapons are forties technology, and even nations as locked down, and downtrodden like North Korea, or Israel (as it was in sixties) can not only make one, but seemingly genuinely capable of keeping it hidden against modern spy, and nuclear detection tech.

Nuclear weapons still aren't easy to make by any means. It's not so much the nuclear physics that's the challenge as it is all of the engineering best-practices. That, and the difficulty of obtaining raw materials.

And nuclear weapons aren't a binary either. North Korea has demonstrated the actual bomb, but their delivery systems are still far from reliable. Their rockets might technically have the range to threaten the US, but what are the chances they could actually hit their target? And how many launch vehicles do they realistically have? Not likely enough to seriously threated any other nuclear powers. And this after making decades of enormous political and economic sacrifices to pursue nuclear weapons. I don't think most countries will feel tempted to follow their lead.

You don't need 100% reliance on ICBMs if your only aim just to hit enemy's mainland for the sake of it. Cheaper means are on the table.

Much more relevant theatre range weapons are things even NK can manufacture. I was very surprised to learn that North Korea not only designed a working cruise missile with their economy, but had a working terrain hugging flight on it.

This got me thinking about the common misperception of how empty the Earth is:

If you dropped a bomb of a certain type and strength on a random coordinate in <country> how many people do you kill on average?

On average, or by percentile? Because I think one is more likely to be revealing than the other.

A 1Mt device might only have a 50% chance of killing 1+ people if detonated at the proper altitude at a random location within the US, (47% of the US is uninhabited) but average 1000 deaths (~6mi lethal radius, US overall population density of 90p/sqmi).

Good point on that. Yes, you've described my thought in much better detail.

> I have a feeling that humans will remain overwhelmingly responsible for most of the wartime mistakes.

Of course we will. It's not as though they're programming themselves. Somebody has to tell the computer what to do.

Hah. Good point. But I meant like self driving: we suuuuuuuck. Let's not fool ourselves acting like it's critical that there's a human touch to things.

Nor should we fool ourselves that it's important for a machine to take over. We don't "suuuck" at driving -- or at killing other people, for that matter: as a species, on average, we're fairly competent at both tasks. The fact that a machine might have a lower failure rate isn't a singular reason to embrace the tech.

I agree with that. I was also referring to how much we suck at not killing the wrong people. I imagine systemic racism would still exist, but at least the robots are making decisions consistently.

> I imagine systemic racism would still exist, but at least the robots are making decisions consistently.

We program our robots with our prejudices[1]. If you don't control the machines, you can't change their programming. It's entirely possible that such prejudices are cemented through machines that cannot be influenced or changed except through actions of their owners.

What happens when the owners of these machines are indifferent towards, or outright supportive of, prejudice? History is rife with owners of companies, industries and governments exploiting and furthering prejudice for their own benefit.

[1] https://www.americanbar.org/groups/judicial/publications/jud...

Exactly. If your fire-control system is trained on a dataset that (through carelessness or malice) has some bias, you could end up with a racist machine gun turret.

the problem with this is, our prejudices can change, sometimes very quickly. A person who loves dogs, for example, might hate them after being mauled by one. Simply put, i don't think we can simply write down our emotional processes easily enough to make advanced ai relevant.

"wartime mistakes" is a nice euphemism. And I'm sure that people will continue be held to account. For instance when somebody whoopsies and dronestrikes a wedding party in Jemen.

Reply on news.ycombinator.com