Let's pump the breaks here for a minute. I've said this many times before, but as someone who focused on urban warfare before entering the tech/AI space, I have some reservations about what high tech can and cannot do in urban environments. Some thoughts:
https://twitter.com/willknight/status/1391783674069258241
1. Urban environments are both congested in communications signals and at times inaccessible/unreliable. So you have to think about how you're going to keep these systems communicating w/ operators AND each other WHILE accounting for enemy interference, electronic warfare, etc.
2. Before anything else, you gotta ask-what problem is the tech solving? So here its abt identifying & targeting enemy combatants under the assumption algorithms will improve to a point where they can do that reliably. That's a huge problem & a really hopeful assumption, imo.
Identification is actually one of the hardest problems in urban operations, esp. distinguishing between civilians & combatants, and it's one of the main reasons for why urban warfare is so costly for civilians even when parties want to comply with IHL (ifri.org/en/publication…).
How are you going to train an AI to reliably identify combatants & differentiate them from civilians when combatants hide and move among the civilian population, use civilian infrastructure, and you rarely can evacuate the entire civilian population from the city?
3. I've written on what AI can do in urban warfare (warontherocks.com/2019/10/with-a…) & I do see both potential & necessity for greater autonomy in systems employed in urban settings, including for reducing civilian harm
But I am skeptical you'll be able to responsibly delegate target selection & engagement to AI in urban settings b/c of civilians present, the extremely dynamic conditions, how fast things can change, and the massive vulnerabilities of ML systems in such complex environments.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
