An interactive,
Reinforcement Learning (RL) and Behavioral Economics dashboard designed to model Stop Work Authority
(SWA) decision-making in industrial environments.
Overview
This application models why workers frequently choose not to use their Stop Work Authority,
even when explicitly told they are allowed to. Rather than treating human workers as strictly
rational actors calculating objective risk, the simulation engine calculates the Perceived
Value of Working vs. The Perceived Value of Stopping.
The mathematical engine driving the dashboard utilizes models established by Kahneman & Tversky
(Prospect Theory), Hyperbolic Discounting, and Stochastic Softmax action selection.
Behavioral Mechanics
The underlying JavaScript engine features several academic concepts:
- Decoupled Probabilities: Factoring the probability
of a hazard separately from the probability of a fatal accident.
- Temporal Risk Discounting (γ): Modeling how the
human brain discounts abstract, future injury risk compared to immediate financial incentives
like piece-rate bonuses.
- Loss Aversion (λ): Utilizing Prospect Theory,
negative penalties (such as Supervisory wrath or False Alarms) are mathematically weighted
drastically heavier than equivalent positive gains.
- Social Capital Buffer: An algorithmic dampener
giving veteran workers structural immunity to the social penalty of a False Alarm, while making
the same penalty devastating for new hires.
- Softmax Action Selection: Workers exhibit a
stochastic probability distribution of behavior based on temperature-scaled differences in
Q-values, modeling the sheer unpredictability of human action.
Usage
This simulation requires no build tools or servers. The variables are entirely controlled by the user
via the left sidebar, which recalculates expected values and logic dynamically via Javascript.