There seems to be a competition to find and exploit the many vulnerabilities in humans and human systems. Historically, that competition has often been won by humans good at manipulation. It’s often been limited by the defenses of the weakest links (e.g. who can convince the dictator that a plan serves them, whether or not the plan does.)
As long as we’re talking capabilities, might as well address the elephant in the room. If “higher capabilities” meant literal humans getting much smarter and harder to fool, sure, defense scales too. And in some cases (e.g. humans making extensive use of AI for defense) that may hold for a time. But there is in fact a long tail of humans and human systems which do not improve at the same rate as the most sophisticated attackers. And AI capabilities underpinning both offense and defense privileges the AIs more than the humans, when the interests of the two kinds of entity diverge.
There seems to be a competition to find and exploit the many vulnerabilities in humans and human systems. Historically, that competition has often been won by humans good at manipulation. It’s often been limited by the defenses of the weakest links (e.g. who can convince the dictator that a plan serves them, whether or not the plan does.)
As long as we’re talking capabilities, might as well address the elephant in the room. If “higher capabilities” meant literal humans getting much smarter and harder to fool, sure, defense scales too. And in some cases (e.g. humans making extensive use of AI for defense) that may hold for a time. But there is in fact a long tail of humans and human systems which do not improve at the same rate as the most sophisticated attackers. And AI capabilities underpinning both offense and defense privileges the AIs more than the humans, when the interests of the two kinds of entity diverge.