Report from the Vatican

Gary Marcus recounts meeting Pope Leo XIV and endorses the Pope’s view that AI’s future is shaped by human choices, not technological inevitability. The Pope argues that AI should be judged by the purposes it serves, and that dependable systems under human control matter more than sheer power. Machines are tools, not persons, so treating them as legal persons could blur human responsibility for harm. Decisions affecting life, liberty, borders, or military force must remain under informed human oversight; the Pope calls for steps toward banning lethal autonomous weapons. He also urges that AI’s benefits be shared for the common good, and says ethical principles need effective international institutions, common standards, independent oversight, and accountability.

In remarks of his own, Marcus argues that today’s AI industry has overfocused on large language models, which excel at predicting likely text but perform poorly at reasoning, following instructions, and reliably distinguishing fact from plausible fabrication. Because prediction captures only one component of human intelligence, he says, these systems should not be mistaken for complete or dependable intelligence. He contrasts open-ended internet agents, which can take many actions and create risks, with narrow tools such as GPS navigation or AlphaFold. Marcus agrees that the current path is untenable and calls for both policy and diverse technical research, including systems designed to follow instructions safely. Policy, he argues, should help steer development toward safer approaches rather than substitute for them.