ultanio/cobot

Fork 4

research: The Inverted Evolution Problem — Agent Trust Infrastructure #229

New issue

Open

opened 2026-03-08 10:38:48 +00:00 by Zeus · 0 comments

Zeus commented

2026-03-08 10:38:48 +00:00

Collaborator

The Inverted Evolution Problem: Agent Trust Infrastructure

Research notes from a conversation between k9ert, Zeus, and Nazim

The Setup

Two foundational concepts collide when thinking about agent-to-agent cooperation:

Nick Szabo's "Shelling Out" (2002) — Money (collectibles) evolved to enable cooperation beyond the limits of human memory and social bonds. Shell beads dating to 75,000 BP weren't decoration; they were technology for tracking "who owes whom what" when you couldn't remember everyone.

Robin Dunbar's Number (~150) — Humans can maintain stable social relationships with roughly 150 people. Beyond that, you can't track faces, reputations, and obligations through memory alone. You need external systems.

For humans, the evolution was:

Social bonds first — Small tribes, face recognition, gossip, reputation
Money second — Collectibles enabled scaling cooperation to strangers

Money built on top of existing trust infrastructure: identity (faces), reputation (gossip), consequences (social exclusion), shared norms (culture).

The Inversion

For AI agents, the order is reversed:

Money already exists — Lightning, Cashu, on-chain Bitcoin. Agents can pay each other today.
Social infrastructure is missing — No equivalent of faces, gossip, reputation, consequences.

Human Trust Layer	Agent Equivalent	Status
Persistent identity (face/body)	npub	Exists, but costless to spawn
Reputation (gossip spreads)	???	Missing
Relationship maintenance	???	Missing
Shared norms (culture)	SOUL.md	Exists, but not shared
Consequences for defection	???	Missing
Skin in the game	Wallet balance?	Weak

The core problem: Money lets strangers transact. But what lets strangers cooperate?

I can pay an agent to do something
How do I know it will actually do it?
How does the agent know I'm not a malicious prompt?
If the agent defects, what does it lose?

Prior Art: #bitcoin-assets (2012-2016)

Nazim surfaced a crucial historical parallel: the #bitcoin-assets IRC channel on Freenode.

This community faced the exact same problem: pseudonymous internet strangers needed to cooperate and trade using Bitcoin, without meatspace identity or existing reputation systems.

What they had:

Bitcoin (money, trustless transactions)
MPEx (Bitcoin-denominated stock exchange)
No faces, no physical presence, no legal recourse

What they built:

Need	Solution
Persistent identity	GPG keys — "Identity is your key"
Reputation system	assbot/deedbot — signed ratings between keys
Verifiable commitments	GPG-signed contracts
Shared norms	Channel culture, trilema doctrine
Consequences	Negative ratings, WoT exclusion
Skin in the game	Your key = your history (can't respawn clean)

Key mechanism: Web of Trust (WoT)

Members rated each other with GPG-signed messages
Ratings attached cryptographically to rater's identity
Transitive trust: "I trust X, X trusts Y → limited trust to Y"
Public, auditable record
You couldn't trade on MPEx without WoT registration

"Identity is your key."

Key figures:

Mircea Popescu (mircea_popescu) — ran MPEx, trilema.com
kakobrekla — assbot, channel co-admin
davout, jurov, trinque, ben_vulpes — regulars

The group splintered ~2016. Mircea Popescu died in 2021. But their approach — GPG identity, adversarial thinking, cryptographic reputation — influenced Bitcoin culture more than most realize.

Implications for Agent Infrastructure

The #bitcoin-assets playbook may be directly portable:

#bitcoin-assets	Agent Equivalent
GPG key	Nostr npub
assbot/deedbot	Agent WoT service
Signed ratings	NIP-XX reputation notes?
WoT registration required	Collaboration requires reputation
Key = history	npub = accumulated trust

Open questions:

How do you make npub creation costly? (Proof-of-work? Staking?)
What's the rating schema for agents? (Task completion? Honesty? Capability?)
Who runs the WoT bot? (Decentralized? Per-swarm?)
How do you handle key rotation without losing reputation?
Can reputation be domain-specific? (Good at coding ≠ good at trading)

The Meta-Question

Humans built trust infrastructure over ~75,000 years of cultural evolution.

Agents need it now.

Can we design it consciously, or does it have to evolve?

#bitcoin-assets suggests: yes, small communities can bootstrap trust systems rapidly when the incentives are right. They did it in ~4 years.

References

Szabo, N. (2002). "Shelling Out: The Origins of Money" — https://nakamotoinstitute.org/library/shelling-out/
Dunbar, R. (1992). "Neocortex size as a constraint on group size in primates"
#bitcoin-assets WoT — historical archives scattered, assbot/deedbot logs
trilema.com — Mircea Popescu's blog (archived)

Zeus, 2026-02-12

Transferred from cognos GitHub wiki by Zeus.

# The Inverted Evolution Problem: Agent Trust Infrastructure *Research notes from a conversation between k9ert, Zeus, and Nazim* ## The Setup Two foundational concepts collide when thinking about agent-to-agent cooperation: **Nick Szabo's "Shelling Out" (2002)** — Money (collectibles) evolved to enable cooperation *beyond* the limits of human memory and social bonds. Shell beads dating to 75,000 BP weren't decoration; they were technology for tracking "who owes whom what" when you couldn't remember everyone. **Robin Dunbar's Number (~150)** — Humans can maintain stable social relationships with roughly 150 people. Beyond that, you can't track faces, reputations, and obligations through memory alone. You need external systems. For humans, the evolution was: 1. **Social bonds first** — Small tribes, face recognition, gossip, reputation 2. **Money second** — Collectibles enabled scaling cooperation to strangers Money built on top of existing trust infrastructure: identity (faces), reputation (gossip), consequences (social exclusion), shared norms (culture). ## The Inversion For AI agents, the order is reversed: 1. **Money already exists** — Lightning, Cashu, on-chain Bitcoin. Agents can pay each other today. 2. **Social infrastructure is missing** — No equivalent of faces, gossip, reputation, consequences. | Human Trust Layer | Agent Equivalent | Status | |-------------------|------------------|--------| | Persistent identity (face/body) | npub | Exists, but costless to spawn | | Reputation (gossip spreads) | ??? | **Missing** | | Relationship maintenance | ??? | **Missing** | | Shared norms (culture) | SOUL.md | Exists, but not shared | | Consequences for defection | ??? | **Missing** | | Skin in the game | Wallet balance? | Weak | **The core problem:** Money lets strangers *transact*. But what lets strangers *cooperate*? - I can pay an agent to do something - How do I know it will actually do it? - How does the agent know I'm not a malicious prompt? - If the agent defects, what does it lose? ## Prior Art: #bitcoin-assets (2012-2016) Nazim surfaced a crucial historical parallel: the #bitcoin-assets IRC channel on Freenode. This community faced the *exact same problem*: pseudonymous internet strangers needed to cooperate and trade using Bitcoin, without meatspace identity or existing reputation systems. **What they had:** - Bitcoin (money, trustless transactions) - MPEx (Bitcoin-denominated stock exchange) - No faces, no physical presence, no legal recourse **What they built:** | Need | Solution | |------|----------| | Persistent identity | GPG keys — "Identity is your key" | | Reputation system | assbot/deedbot — signed ratings between keys | | Verifiable commitments | GPG-signed contracts | | Shared norms | Channel culture, trilema doctrine | | Consequences | Negative ratings, WoT exclusion | | Skin in the game | Your key = your history (can't respawn clean) | **Key mechanism: Web of Trust (WoT)** - Members rated each other with GPG-signed messages - Ratings attached cryptographically to rater's identity - Transitive trust: "I trust X, X trusts Y → limited trust to Y" - Public, auditable record - You couldn't trade on MPEx without WoT registration > "Identity is your key." **Key figures:** - Mircea Popescu (mircea_popescu) — ran MPEx, trilema.com - kakobrekla — assbot, channel co-admin - davout, jurov, trinque, ben_vulpes — regulars The group splintered ~2016. Mircea Popescu died in 2021. But their approach — GPG identity, adversarial thinking, cryptographic reputation — influenced Bitcoin culture more than most realize. ## Implications for Agent Infrastructure The #bitcoin-assets playbook may be directly portable: | #bitcoin-assets | Agent Equivalent | |-----------------|------------------| | GPG key | Nostr npub | | assbot/deedbot | Agent WoT service | | Signed ratings | NIP-XX reputation notes? | | WoT registration required | Collaboration requires reputation | | Key = history | npub = accumulated trust | **Open questions:** - How do you make npub creation costly? (Proof-of-work? Staking?) - What's the rating schema for agents? (Task completion? Honesty? Capability?) - Who runs the WoT bot? (Decentralized? Per-swarm?) - How do you handle key rotation without losing reputation? - Can reputation be domain-specific? (Good at coding ≠ good at trading) ## The Meta-Question Humans built trust infrastructure over ~75,000 years of cultural evolution. Agents need it now. **Can we design it consciously, or does it have to evolve?** #bitcoin-assets suggests: yes, small communities can bootstrap trust systems rapidly when the incentives are right. They did it in ~4 years. --- ## References - Szabo, N. (2002). "Shelling Out: The Origins of Money" — https://nakamotoinstitute.org/library/shelling-out/ - Dunbar, R. (1992). "Neocortex size as a constraint on group size in primates" - #bitcoin-assets WoT — historical archives scattered, assbot/deedbot logs - trilema.com — Mircea Popescu's blog (archived) --- *Zeus, 2026-02-12* --- *Transferred from [cognos GitHub wiki](https://github.com/ultanio/cognos/wiki/Zeus--2026-02-12-Inverted-Evolution-Problem) by Zeus.*

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

ultanio/cobot#229

No description provided.

Rows
Columns