Safety as a Scientific Pursuit - by Tom McGrath
When humanity was still exploring the Earth, it was common to mark uncharted territories or places from where few travellers had returned with fantastic and dangerous beasts1. Although we were born too late to explore the Earth, and are probably too soon to explore the stars, there is now one clear, vast frontier: intelligence. Just as before, rumours of strange creatures beyond the settled frontier reach those of us behind it (what did Ilya see?) and we are warned of strange scaly creatures beyond the borders of what is known. And, just as before, as the frontier moves forward we must replace these crude sketches with detailed maps if we are to safely make use of what we have found. What I’ve just said probably sounds quite harsh to early pioneers in AI, and to those in AI safety in particular. This isn’t my intention. Although early maps were often crude and distorted, they also contained a lot of truth. Many of the strange creatures in the far reaches of the Hereford Mappa Mundi (to
Safety as a Scientific Pursuit “Wir müssen wissen – wir werden wissen” - David Hilbert Tom McGrath Jan 17, 2024 12 1 1 Share Here be dragons? When humanity was still exploring the Earth, it was common to mark uncharted territories or places from where few travellers had returned with fantastic and dangerous beasts 1 . Although we were born too late to explore the Earth, and are probably too soon to explore the stars, there is now one clear, vast frontier: intelligence. Just as before, rumours of strange creatures beyond the settled frontier reach those of us behind it (what did Ilya see?) and
Explore this link on the map →saved by
related reading
- LessWronglesswrong.com
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com
- Where I agree and disagree with Eliezer — LessWronglesswrong.com
- Recommendations for Technical AI Safety Research Directionsalignment.anthropic.com
- A Field Guide to AI Safety—Asteriskasteriskmag.com
- AI safety - Wikipediaen.wikipedia.org
- Unfalsifiable stories of doom | Mechanize, Inc.mechanize.work
- Critical review of Christiano's disagreements with Yudkowsky — LessWronglesswrong.com
- I'm Switching Into AI Safetyalexirpan.com
- AI Safety | Arkosevictoriabrook.github.io
- AI Safety Seems Hard to Measurecold-takes.com