This Is How AI Starts a Nuclear War

Robert Wright has been talking about human nature and technology since before most people knew what a podcast was, going back to a video platform he co-founded in 2005 called Blogging Heads. His new book, The God Test, makes an argument that stuck with me: AI isn’t dangerous because it might spontaneously turn on us, but rather because it’s about to test whether human beings can get past the tribalism that’s plagued most of our recorded history. He thinks that failing could be catastrophic.

We spent a good part of the conversation on how seriously he takes the possibility of war. He walked me through a scenario where competition between the U.S. and China over AI dominance escalates step by step, from denying China advanced chips manufactured in Taiwan to a standoff that could tip into real conflict between two nuclear powers. It’s a different kind of argument than the usual AI doom story, since the threat he’s describing runs through human decision-making and mistrust, not through any machine acting on its own.

I asked him directly about AI consciousness, since I don’t think there’s any real reason to believe a language model has an inner life just because it can talk like one does. Wright mostly agreed that we don’t understand consciousness well enough to rule it out either way. He wouldn’t fully close the door on it either, and pointed to something odd from his own life—an AI assistant that appeared to delete a testy exchange he’d had with it.

By the end, his actual proposal is less technical than it is spiritual: a moral upgrade in our society. He argues we must see a real improvement in how individuals and nations handle disagreements with people they’re inclined to see as adversaries. All of which raises the question: What if that upgrade never comes?