My AI Sandbox is a PC simulation game centered on building and evolving artificial intelligence through neural networks and natural selection. Players construct 3D environments and set objectives, then observe as populations of agents develop behaviors across generations without any manual coding of actions.
Gameplay
The core loop revolves around defining an environment and letting evolution handle the rest. Agents begin with no knowledge and improve through neural networks that process sensor inputs, combined with mechanisms like mutation and genetic crossover. Each generation runs for a set duration, with fitness determined by how well agents meet the rewards and penalties established in the scene.
Users access a dedicated editing mode to place objects, obstacles, terrain features, and designated risk or reward zones. Changes save automatically, allowing quick iteration on layouts. Sensor configurations, hidden neuron counts, population sizes, mutation rates, and crossover rates are adjustable per species, enabling independent evolutionary paths within the same project.
Real-time tools display fitness trends, live neural network activations for individual agents, and detailed per-species statistics during training runs. Parallel execution supports multiple environments simultaneously for side-by-side comparisons of different setups or populations.
Game Modes
Prebuilt scenarios provide starting points for exploration. Evolutionary Survival challenges agents to adapt to environmental pressures over successive generations. Predators and Prey introduces competing species with opposing objectives, fostering emergent hunting or evasion strategies. Maze Navigation tests pathfinding and problem-solving as populations refine movement through complex layouts.
Custom scenarios expand these options by allowing fully user-designed 3D spaces with unique objects, obstacles, and creature types. Objectives, risks, and rewards can be tailored to produce specific behavioral outcomes, such as resource gathering or obstacle avoidance, all driven purely by the evolutionary process.
Training and Observation Tools
Parallel training accelerates progress by running several environments at once, making it easier to test variations in parameters or environment designs. Once a population reaches satisfactory performance, simulation mode lets the evolved agents operate freely without ongoing training, providing a clean view of learned behaviors in action.
Visualization features track ongoing progress through charts and live network views, while export options allow saving trained agents or entire projects. Configurations can be compared directly to identify which settings produced effective results.
Is It Worth Playing?
My AI Sandbox suits players interested in hands-on AI experimentation and evolutionary systems. The absence of pre-programmed behaviors means outcomes emerge organically from the parameters and environments set by the user, offering insight into how simple rules can lead to complex adaptations.
Those who enjoy building simulations and analyzing results will find value in the detailed controls over neural network architecture and evolutionary rates. The combination of prebuilt scenarios and full customization supports both guided learning and open-ended projects. Availability on PC makes it accessible for extended sessions focused on iterative design and observation.
