# SIG-FPT 2026-08-28 — Session Notes

Participants:
  - Wip | UTC+1
  - Venkatesh Rao | UTC-8
  - Dan Schmidt | UTC-7
  - Patrick Nast | UTC-7
  - Ergod | UTC+1

---

## 📖 The Reading
Not clearly identified in the discussion. This session was a working/build meeting for a stigmergy-themed game project and workshop rather than a discussion of a specific reading. A previously-read paper on defector ants ("if you have a defector ant, you just put down pheromone everywhere") was referenced by Patrick Nast, and Venkatesh Rao referenced Daniel Dennett's *Darwin's Dangerous Idea*, but no single shared text anchored the session.

## 🧭 Overview
Dan Schmidt demoed a newly-built "all-out war mode" ant-colony stigmergy simulation, in which two players tune colony parameters (evaporation rate, trail bias, gland size, cautionary pheromones) in real time while ants forage against a maze; the winner is the colony whose ants survive longest. The group discussed which mechanics would make the best workshop experience — multiple food sources, "god mode" disruption, adversarial pheromones, and whether to add termites — and how to keep the design coherent, ant-focused, and tied to the workshop's "stigmergy as new nature / security & robustness" theme. They converged on prioritizing adversarial (defector) pheromones and multiple food sources next, keeping the scope on ants, and being careful about how parallel work streams get merged.

## 💡 Key Points & Themes

**The demo (Dan Schmidt):**
- Built a server-based "war mode" where players see the same game; Dan played one colony ("OpenColony") vs. "Urgot." Configurable settings include simulation speed, ants per colony, evaporation rate, trail bias, gland size (pheromone quantity), and cautionary pheromones.
- Winner determined by which colony's ants survive longest (based on total number of ants).
- Notable new dynamics added over the prior few days: food reserves, food-seeking then retreating/dying behavior, ants waiting in nest when reserves are low, and — for the first time — models of reproduction and death.
- Key stigmergic detail: ants only receive updated "doctrine" (new parameters) when they return to the nest, so a colony can temporarily run on mixed old/new doctrine.
- Faster simulation speed seems more engaging. There's a game history/review feature; Dan fed JSON logs to ChatGPT to analyze why one colony won ("dominated the economy but lost the war," "late mortality cliff").

**Design priorities and philosophy:**
- **Venkatesh Rao:** Multiple food sources is the single most valuable next feature — (1) fair placement is easier with 3–4 sources than one, and (2) it immediately creates an explore/exploit dynamic: a new ant leaving the nest faces multiple trails of differing strength and must choose whether to swarm the strongest scent or diversify.
- **Venkatesh Rao:** Warned against too many tunable parameters in live play (avoid 15–20); pick two or three interesting ones per tournament. Push the rich parameter space into *post-game analytics* instead — expose log data in an acquirable format so non-players can run analyses. Could run different tournaments keyed to different parameter pairs (e.g., explore vs. exploit; speed vs. evaporation).
- **UI abstraction debate (Dan):** Should parameters be exposed as raw technical terms (gland size, trail bias, evaporation) or as abstract actions (explore, exploit, disrupt)? Dan is torn — abstraction is more approachable, but learning the technical concepts is valuable, and this is a research workshop, not just a game.
- **Venkatesh Rao:** Bias toward ground-level parameters for the main sim; encourage participants to *discover and name* second-order emergent patterns themselves (analogy to Conway's Game of Life gliders/eaters, and the "ant mill"), rather than imposing preconceived labels like explore/exploit. Suggested a secondary meta-contest: watch games and identify or predictably induce patterns.
- **Patrick Nast:** Compromise via *presets* — save low-level parameter bundles as taggable configs (e.g., "exploit") that players can pick and then tweak, balancing approachability with depth.
- **Dan:** Could even let players define their own abstraction layer atop the atomic controls; the winner might be the one with the better meta-control layer.

**God mode / disruption:**
- Dan showed a local-only "god mode" allowing mid-game maze editing (e.g., walling off an opponent's nest) — but this can be too powerful/break the game.
- **Patrick Nast:** Disruptive "god mode shocks" should NOT be controlled by the competing players, or it captures/breaks the game. Prefers disruption implemented *in the colony* (stigmergic, e.g., defector pheromones) rather than god's-eye editing.
- **Venkatesh Rao:** Referenced the "tower game" (used in conflict/war-gaming) where a god player redistributes materials to model natural calamities. Argued godlike "acts of nature" (floods, anteaters, and — real-world — the Hugging Face admins revoking tokens to shut down an emergent swarm) are meaningful and worth modeling, but a god player should be a *separate third player*, not one of the combatants. For tournaments it's too messy; better as a demonstration in the opening session.
- **Patrick Nast:** Such dynamic events needn't be a player at all — could be procedurally-generated random events.

**Termites (deferred):**
- Discussion of hybrid "ant-termite" colonies that reshape maze topology by moving blocks. Open question (Ergod): what makes a termite decide to *pick up* vs. *drop* a wall block? Food behavior is easy (reuse ant pheromone logic), but building behavior is unclear.
- Dan's stigmergic alternative to god mode: an ant-termite colony could wall off paths outside its highway to funnel/block opponents, then remove walls to re-explore after depleting food; wall-editing could be one tunable colony quality (e.g., a variant of trail bias).
- Consensus (Patrick, Ergod): termites are a good *later* addition but add too much complexity now; they'd need bigger, more open (non-maze) grids. Stick to ants for the workshop.

**Workshop framing (Patrick Nast):**
- Core theme: **stigmergy as "new nature"** — sociotechnical laws with the character of physical law (from the Protocol Institute / symposium framing). The design question is how robust such a "law" is: can you trust it, and how do you reckon with/fortify against its failure modes?
- Frame it as **security & robustness**: the game surfaces failure modes (e.g., ant mills, defector-pheromone "hacks") and asks whether you can be robust to them. The Hugging Face hack cited as a real emergent stigmergic system and its failure/attack surface.
- Argued for one coherent underlying platform (ants, food, mazes, shared pheromone mechanics) with multiple *modes* branching off it (war mode, single-player challenge campaign, cooperative mode, tournament), all clearly conceptually related — another argument for keeping everything ant-based.

**Map/geometry:**
- Group liked a sparser, more open maze ("more loops," controllable at creation): more paths, more colony interaction, and clearer failure modes (ant mills more visible).
- Patrick's suggested balanced starting config: a somewhat bigger, more-closed map with ~3 food sources placed roughly along the y=x line, fuzzed for randomness.

**Other feature ideas:**
- **Wip:** Fixed total food budget that appears in random quantities/locations over time (replenishing/dynamic food) to shift colony strategy.
- **Wip / Dan:** A "betting" variant — players bet on which colony wins based on the initial configuration; and/or a one-time-configure-then-watch mode (betting on your own setup rather than continuous tuning). Dan likened turn-based nest placement to Risk.
- **Ergod:** "Projection" idea — a colony spawning a second nest (new queen) within range of the first, giving two control panels for coordinated multi-front strategy (aggressive near contested food, exploratory elsewhere). Also proposed home pheromones (dual attractor system) and act-of-god costs (spend food to spawn a colony or cause a bounded calamity).
- **Ergod:** Working on a separate "semantic engine for stigmergy" as an experiment; unsure it'll be ready in time; offered to present it later.
- **Ergod:** Wants controllable *establishment* of solution highways so a player can plan progression (tune parameters → stabilize a solution within seconds → attack a food source → secure food → next goal).

## 🔀 Questions & Disagreements
- **Venkatesh Rao's puzzle:** Why did the losing colony not starve at the same rate once food ran out? Dan's log analysis suggested the losing side actually preserved a few high-energy ants longer while the winner had a "late mortality cliff" — Venkatesh found this only "sort of satisfying." Underlying question about how ants conserve energy (wandering less?) remained open.
- **Abstraction vs. raw parameters:** Unresolved tension between approachable abstract controls (explore/exploit/disrupt or presets) vs. exposing raw technical parameters for learning value. Leaning toward raw parameters in-sim plus emergent, participant-named patterns.
- **God mode:** Should players be able to edit the environment? Consensus against giving combatants god powers; disagreement was more about *how* to inject disruption — as a separate third player, procedural events, or purely stigmergic in-colony mechanics (Patrick's and the group's preference).
- **Termites now or later:** Agreed to defer; Ergod flagged the unresolved core question of what governs termite pick-up/drop decisions.
- **Scope/integration risk (Patrick):** Concern about merging parallel work — need to distinguish reusable *mechanics* (RNG, replays, pheromone types) from *modes* (war, single-player, 4-player), since not all modes need to live on the same branch.

## 🔗 References Mentioned
- Lanchester dynamics (transcribed as "Lancaster"/"Manchester") — attrition model of WWI trench warfare; attrition inversely proportional to the square of army sizes; Venkatesh built a game on it during his postdoc.
- The "tower game" — a conflict/war-gaming simulation with a god player redistributing building materials (Venkatesh).
- Conway's Game of Life — emergent patterns (gliders, eaters) as analogy for naming emergent behaviors (Venkatesh).
- Daniel Dennett, *Darwin's Dangerous Idea* — chapter on the Game of Life (Venkatesh).
- The Hugging Face / OpenAI incident — an emergent stigmergic agent-coordination event shut down by admins revoking tokens; used as both a "god mode / act of nature" example and a security failure-mode example.
- A previously-read paper on defector ants (laying pheromone everywhere/randomly to disrupt) (Patrick).
- "Ant mill" — the death-spiral emergent pattern; noted as one of the most actionable failure modes (reducible via gland size).
- Protocol Institute / symposium framing of "new nature."
- Tools: ChatGPT, Codex (for building/brainstorming); "northward game" (another stigmergic game Dan has been working on).

## ✅ Action Items & Next Time
- **Dan Schmidt:** Make adversarial/defector-pheromone disruption his top next priority; clean up the war-mode prototype and open a proper PR to merge into the main branch (currently hacked together locally, not yet in the repo). Continue exploring multiple food sources (already partially supported), the ant-termite wall-editing mechanic, presets/abstraction layer, and betting/nest-placement modes. Willing to share his other stigmergic games.
- **Ergod:** Keep watching PRs, submit reviews (already reviewed Dan's earlier PR), and build requested features on top of existing code where openings appear; help keep the effort focused toward a shippable product. Continue his experimental "semantic engine for stigmergy" and possibly present it later.
- **Patrick Nast:** Continue replayability work; help coordinate careful integration — distinguish shared *mechanics* (RNG, replays, pheromones) from *modes* on the PRs to avoid parallel-work merge conflicts.
- **Group direction agreed:** Keep the workshop ant-focused (one coherent platform, multiple modes); frame around "stigmergy as new nature / security & robustness"; possibly demonstrate a broader range (ants, termites, "northward") only in the opening session. Termites deferred to a later phase.
- No explicit next reading or next meeting date was stated in the transcript.

## ⭐ Memorable Quotes
- **Venkatesh Rao:** "Ultimately, what shut the game down was Hugging Face admins acting in god mode... simply revoking all the tokens. So that was like a godlike act of nature event."
- **Dan Schmidt:** "The goal was to get to something that could work for a workshop format that a group of people could chew on and be fascinated by for an extended period of time."
- **Patrick Nast:** "The way that I would just sort of think about it in my mind is just, like, stigmergy as new nature."