← Workbench

Clarkson AI Institute · Workbench

Defender · harden

Defender · watch

Attacker · exploit

+1, −1

+3, −3

Attacker · probe

+2, −2

0, 0

You are the attacker. Pick a row — the AI defender will try to contain you.

round 0/12your score +0value line +1.5/rd

Games & Strategy

Wargames built as scientific instruments — each one a contest with a measurable outcome.

$$v^{*}=\max_{x}\ \min_{y}\ x^{\top} A\, y \;=\; \min_{y}\ \max_{x}\ x^{\top} A\, y$$

a contest with a stable balance of advantage — play it above — you hold the structural edge

The Asymmetry

The attacker needs one way in. The defender has to hold every door.

In cyberspace the defender protects every vector while the attacker chooses one. False narratives spread faster than corrections. Exploitation costs less than resilience. Across every domain we model, the balance tips the same way — and the games are designed to surface it, measure it, and expose it.

one breach winsall must hold

$\text{attacker wins}\iff \exists\,v:\ \mathrm{breach}(v)\qquad \text{defender wins}\iff \forall\,v:\ \mathrm{hold}(v)$

The Suite

green = defender · gold = attacker

MINDSPACE

Cognitive Warfare

Four asymmetric factions contest a population’s attention, trust, and resilience — built on the DISARM influence-operations taxonomy.

$s_{t+1}=f(s_t,\ \mathrm{TTP}_t)$ — population state updated by each technique played

Attacker-dominant — three attack surfaces, one defender covering all.

▶ Play

INFOSPHERE

Influence · Belief Spread

One player seeds disinformation, the other contains it — over validated models of how beliefs spread through social contact.

$x(t{+}1)=W\,x(t)$ — DeGroot & Deffuant opinion dynamics

Spreader wins — reproducible across play conditions, matching the literature.

▶ Play

CyberWar

Network Intrusion

AI agents attack and defend a network — probing, moving laterally, patching — configured on MITRE ATT&CK concepts.

attacker $\exists\,v:\mathrm{breach}(v)$ vs defender $\forall\,v:\mathrm{hold}(v)$

Attacker-dominant — measured success rates, detection windows, failure modes.

in development →

Cog & Capital

Behavioral · AI Advisors

Victorian industrialists compete for market dominance while AI-powered advisors shape their decisions.

$a^{*}=\arg\max_a \mathbb{E}[\,u\mid \mathrm{advice}\,]$ — does the advice change the choice?

Open question — how AI-mediated advice changes human strategy.

in development →

The Great Awakening

AI-Adjudicated Geopolitical Crisis

Five factions submit orders in plain English; an AI Oracle reads the situation and decides what happens, in narrative form — no action menus, no probability tables.

$s_{t+1}\sim \mathrm{Oracle}(s_t,\ o^{1},\dots,o^{n})$ — the model is the referee

Research frontier — what AI adjudication can and cannot do reliably: consistency, creative play, escalation, balance.

▶ Play

+ next game

Wargames as scientific instruments

A simulation grounded in a verified model produces observations you can analyze, compare, and publish. Players take asymmetric roles; AI agents execute strategies; the games generate outcome data. That data feeds academic research.

The suite’s central claim — that attackers hold structural advantages that persist across domains, and that those advantages come from how defense obligations are distributed rather than attacker ingenuity — is tested domain by domain.

Research in preparation for Simulation & Gaming. Deployment data from student cohorts will constitute the empirical section.

In the Classroom

These games grew out of a Clarkson course where students learn history — from the Berlin Airlift to Cold War nuclear rivalry — by playing wargames and designing their own.

Dr. Ali Kocho-Williams built the course and its game library through Clarkson’s President’s Challenge for Faculty. Student teams have produced functional game prototypes that drew interest from commercial publishers, and the class launched its own podcast, “Playing with the Past.”

Read the Clarkson feature →

Deploy in your course or workshop

All games are browser-based, require no installation, and run on any device — suitable for security studies, information warfare, and cognitive science courses, practitioner workshops, and public engagement.

Faculty Lead — Alastair “Ali” Kocho-Williams
Arts, Culture & Technology · Humanities & Social Sciences

akochowi@clarkson.edu contact → Cold Alchemy Games → DISARM Foundation →

BUILD-WORLD_WITHAI · WORKBENCH · GAMES & STRATEGY
Clarkson AI Institute — prototype draft, not yet published

← Back to the Workbench