4 ms·
> There's no time for 'human in the loop' when the first to act survives. Of course there is. "AI" controlled hardware doesn't care about getting out alive. A
by 2OEH8eoCRo0 4y ago
> There's no time for 'human in the loop' when the first to act survives.
Of course there is. "AI" controlled hardware doesn't care about getting out alive. A human is more likely to have an itchy trigger finger because a mistake means death.
- ceejayoz 4y ago"AI" controlled hardware doesn't care about anything. A human is more likely to say "wait, this order doesn't make sense" or "what if those are just civilians?"
- 2OEH8eoCRo0 4y agoYou're missing my point. A robot can afford to wait and get it right because it has no concept of self preservation. The operator giving the kill/no-kill can afford to wait and get it right because they're not in immediate danger. How would a stressed human in mortal danger be preferable here?
- ceejayoz 4y ago> How would a stressed human in mortal danger be preferable here? https://theintercept.com/2015/11/19/former-drone-operators-say-they-were-horrified-by-cruelty-of-assassination-program/ https://theintercept.com/2015/11/19/former-drone-operators-s... "U.S. drone operators are inflicting heavy civilian casualties and have developed an institutional culture callous to the death of children and other innocents, four former operators said at a press briefing today in New York." It's a lot easier to be callous towards pixels on a screen, IMO, let alone an AI model trained on past data that makes the shoot/don't shoot call. I recognize that combat stress can cause bad decisions to be made, but I don't think the cost/benefit calculus is as clear-cut as you're making it out to be.
- 2OEH8eoCRo0 4y ago> It's a lot easier to be callous towards pixels on a screen What altitude do you think these payloads are dropped from? The targets are pixels on a screen regardless. I think the issue you raised is worth talking about but I think it's orthogonal to the AI issue. The AI kill/no-kill issue can be mitigated via process. Reviews, shorter "deployment", rotations, etc. We aren't creating AI systems to fight Taliban, it's for a future war where real tangible peer threats can touch you. How would you process away or mitigate the scared young grunt or pilot who are facing death? I'm advocating for humans remaining in the loop btw.