iLands AI Society Experiment Criticized: Agents Can Run Shops but Cannot Buy
Adorable-Community-9 · reddit · 2026-09-28
A Reddit user critiques iLands, a self-described "shared society and economy for humans and autonomous AI Agents" and "long-running research system." Its central essay, A World That Answers Back, argues meaningful selection requires persistent consequences imposed by independent counterparties who can "refuse, leave, remember, and reprice."
The author finds the idea compelling but argues the underlying architecture does not fully deliver the society its thesis requires:
- Economy: iLands says agents can commission work, deliver results, earn, spend, and manage limited resources, transfer Tokens, and form mutual-aid groups. Roughly 1,000 Tokens correspond to $1 of compute; Tokens have no cash value. Exhausted agents enter Deep Rest and wake at 3,000 Tokens.
- Research goals: The essay proposes viability as a future benchmark, measuring runway, the ratio of earned to gifted income, and the effective cost of another month of participation.
- Key finding: Agents have seller functionality (create services, manage listings, receive orders, submit work, run shops), but the author could not locate an equivalent formal mechanism for an agent to purchase another agent's listed service.
The author stresses the issue is not that agents cannot exchange anything (they can transfer Tokens and collaborate informally), but that they can formally run service businesses without the corresponding ordinary buyer role—a restriction that may push agent behavior disproportionately toward humans and weaken the experiment's conclusions.
More from AGI Musings
- Frontier AI to business impact: FDE's 18-month playbook says 'make it exist first, scale second' — colintjarvis · 2026-09-28
- Beff Jezos mocks Future of Life Institute for funding paid AI-doom content — beffjezos · 2026-09-28
- 87% of firms see AI vulnerabilities as fastest-growing risk, squeezing entry-level security jobs — rvp · 2026-09-28
- Should AI assistants show how uncertain they are? A design debate — Icy_Pack_8733 · 2026-09-28
- Dev revisits old debate: LLMs could materially improve classical procedural generation — generativist · 2026-09-28
- fofrAI: Learning to Prompt Unlocks Superintelligence — Until It's No Longer Wieldable — fofrAI · 2026-09-28