“Worrisome Trends for Digital Mind Evaluations” by Derek Shiller

This is a quick take, not a research report, reflecting the current thoughts and impressions of (some) members of Rethink Priorities' Worldview Investigation Team. We plan to occasionally post views we have (like this one) that we think are important but which we haven't had time to focus on in depth.

Evaluating AI systems for morally relevant properties like agency and consciousness is presently quite challenging. AI systems are complex and our understanding of their functioning remains rudimentary. Human and animal brains are, if anything, even less well understood. Coming up with verdicts on the presence of traits like valenced experience or morally-weighty desires involves distilling expert opinion, largely formulated in terms of humans and other animals, and applying the lessons derived to what we know about AI systems. Doing this well requires thought and conceptual nuance, and even in the ideal will leave us with significant uncertainty.

[...]

---

Outline:

(02:06) 1) Increasing secrecy

(05:51) 2) Wider exploration

(08:10) 3) AI-driven innovation

(11:37) Implications

---

First published:
February 20th, 2025

Source:
https://forum.effectivealtruism.org/posts/axHwbeiKA4ScDHik3/worrisome-trends-for-digital-mind-evaluations

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.