The AI War and How to Win It
alexw.substack.com
alexw.substack.com
I'd be particularly bearish on AI for grand strategy. If LLMs need billions of examples of text to outperform humans, an analogous war strategy AI would need to be trained on more examples of warfare than have been conducted in human history.
Innovation is then left up to everyone else.
The defense budget is hundreds of billions of dollars
Reinforcement learning models are already crushing humans at strategy games from StarCraft to Diplomacy
Diplomacy especially has at most three actions that can be taken per turn (move, support, hold) maybe five if you include convoy and retreat. Generally each country does have pretty limited paths to win.
The impressive thing is the natural language processing of the AI and from what I read from one top diplomacy player the fact that the AI doesn’t try and take revenge when betrayed, doesn’t develop trust with any players or let any sort of subconscious feelings get in the way of its goal.
I don’t see a world where, as the article suggests, all war is automated.
I see a world where AI helps us be even more prepared for the last war, and equally unprepared for new wars.
The twist is that the AI itself will ultimately become the Authoritarian. Unless technological progress comes to a screeching halt (nuclear obliteration, or what have you) this outcome is probably inevitable and probably even the best possible outcome.