Anthropic is unwilling to cooperate with the U.S. Department of Defense, firmly maintains that AI must never be used in lethal autonomous weapon systems—this is regarded as an important “principle” within the AI ethics community. Since Anthropic refuses to develop AI designed to attack enemies, what real purpose does the Department of War have in approaching them for collaboration?
American AI companies often impose strict self-restrictions, declaring they cannot do this or that, while Chinese AI companies face no comparable moral constraints. Gunpowder was invented not only to blast open mountain paths but inevitably also to kill people. Nuclear energy was developed not just for electricity generation but, of course, also to create nuclear weapons.When American AI companies engage in endless internal debates about how to ethically regulate their own development—insisting on “do no harm to humanity” and the like—it ultimately amounts to little more than empty rhetoric.
Videos of Chinese robots performing with knives and guns are abundant online. If the United States chooses not to develop offensive AI, China will do it anyway.
When humanity entered the Iron Age, trying to prevent iron from being used in warfare was absurd. The sensible approach was always to build better iron weapons and defenses for oneself.The way to prevent nuclear war has never been to abstain from developing nuclear weapons, but to establish powerful defensive nuclear capabilities.Likewise, the way to stop AI from advancing to the point of threatening or exterminating humanity has never been to forbid AI from attacking humans altogether, but to build robust, defensive AI military capabilities. With such systems in place, AI police could immediately neutralize, through physical means, a deranged individual carrying out an indiscriminate massacre in a school or public space.