AGI·LEDGER
calibrating the instrument · 0%
an observatory of human belief
What humanity believes AGI should be.
Sealed before it arrives — every belief attributed, timestamped, immutable. Scroll, and watch a mind become a question.
scroll to begin
So we asked the world. Every answer becomes a star.
22
beliefs
20%
median P(doom)
5
factions
III · the cosmos of belief
Click a star to read it — or a faction to enter its cluster.
IV
Put your belief on the record.
One take. One falsifier. Then it joins the cosmos — a star among humanity's.
Add your take →- AGI is overhyped; embodiment is hard and timelines are long — by Rodney Brooks (Tool-only)
- AGI is the next step in evolution — embrace it — by Jürgen Schmidhuber (Accelerationist)
- AGI within ~a decade — use it to solve science, but build it carefully — by Demis Hassabis (Independent)
- AI is a human-made tool; the frontier is spatial intelligence — by Fei-Fei Li (Tool-only)
- Alignment is hard but maybe tractable — I'd put catastrophe near 50% — by Paul Christiano (Independent)
- Build AGI, deploy it iteratively, and it will lift the whole world — by Sam Altman (Independent)
- Build maximally truth-seeking AI — fast — before someone builds worse — by Elon Musk (Accelerationist)
- Build non-agentic 'Scientist AI', not autonomous agents — by Yoshua Bengio (Cautious Oracle)
- Design AI that is uncertain about human values — and defers to us — by Stuart Russell (Cautious Oracle)
- Digital intelligence may simply be a better form of intelligence — by Geoffrey Hinton (Independent)
- Fearing AI doom is like worrying about overpopulation on Mars — by Andrew Ng (Accelerationist)
- Forget AGI — the real challenge is containing capable AI — by Mustafa Suleyman (Tool-only)
- If anyone builds superintelligence now, everyone dies — by Eliezer Yudkowsky (Halt)
- LLMs are an off-ramp; real AI needs world models — and won't want to dominate us — by Yann LeCun (Accelerationist)
- Pause the race to superintelligence until it is provably safe — by Max Tegmark (Halt)
- Powerful AI by ~2027 — it could be the best or the worst thing ever — by Dario Amodei (Independent)
- Scaling LLMs won't get us to AGI — we need neuro-symbolic AI — by Gary Marcus (Tool-only)
- Solve the control problem before, not after, superintelligence — by Nick Bostrom (Independent)
- Stop scaling toward AGI — we cannot align it in time — by Connor Leahy (Halt)
- Superintelligence is coming; build it safely, as the only goal — by Ilya Sutskever (Independent)
- Superintelligence is uncontrollable, so we should not build it — by Roman Yampolskiy (Halt)
- We will merge with AI and transcend biology — by Ray Kurzweil (Accelerationist)
The six questions
Every belief on the Ledger answers the same six questions about the mind we are about to build.
- What form?
- Its shape — oracle, embodied, swarm, or tool.
- How built?
- How it's built — scale, world-models, or agents.
- Made of?
- What it runs on — silicon through bio-hybrid.
- What for?
- What it's for — helper, peer, successor, or never.
- Who controls?
- Who holds the keys — open, locked, or halted.
- When?
- When it arrives — this decade, or never.
The aggregate
How the cloud leans, summed across every belief.
What form?
Other
17 · 77%
Oracle
2 · 9%
Embodied
1 · 5%
Tool
1 · 5%
Ambient
1 · 5%
How built?
Other
14 · 64%
Bigger models
4 · 18%
World-model
3 · 14%
Logic + nets
1 · 5%
Made of?
Doesn't matter
22 · 100%
What for?
Helper
9 · 41%
Tool
5 · 23%
Other
3 · 14%
Don't build
3 · 14%
Successor
1 · 5%
Peer
1 · 5%
Who controls?
Locked down
12 · 55%
Open to all
4 · 18%
Should be halted
4 · 18%
Other
1 · 5%
Decentralized
1 · 5%
When?
Unstated
7 · 32%
2030s
6 · 27%
Before 2030
5 · 23%
After 2060
2 · 9%
2040s–50s
2 · 9%