Project Big Bang: The Journey to Biologically Inspired Brains
For years, the AI industry has been obsessed with computational scale. Larger models, bigger context windows, massive GPU clusters. Yet, all of these systems share a fundamental flaw: they are static, passive oracles. They sit frozen in time, waiting for user queries, unable to grow, adapt, or experience the world.
We wanted to build something that was truly alive. We wanted to see what happens when you give a neural network a body, place it inside a physical space, and tell it: Survive.
This is the story of Project Big Bang and the Biologically Inspired Brain (BIB).
The Substrate: Biologically Inspired Brain (BIB)
To survive, an organism cannot rely on backpropagation or massive offline training runs. It must learn in real-time, in one shot, without forgetting what it already knows. To achieve this, we developed BIB—a 1:1 digital replica of the mammalian cognitive substrate:
- Predictive Cortex: A multi-layered sparse representation model. Instead of dense neural connections, it maps ideas as Sparse Distributed Representations (SDRs) where only a tiny fraction (e.g., 0.39%) of neurons fire at once. This sparsity prevents catastrophic forgetting; learning new things does not overwrite the old.
- CA3 Recurrent Hippocampus: A fast-learning attractor network that captures single-trial experiences, storing them temporarily before consolidating them into the cortex.
- Endocrine System & Brainstem: A simulated chemical network. Dopamine tracks reward prediction errors to reinforce actions; Acetylcholine signals surprise and drives cortical plasticity; Serotonin regulates risk and emotional tone; Cortisol triggers stress responses when resources dwindle.
- Actor-Critic Basal Ganglia: Gates motor actions using Go (D1 receptor) and NoGo (D2 receptor) pathways, allowing the organism to decide when to move, feed, evade, or mate based on its current chemical drives.
The Testing Ground: Project Big Bang
A brain is useless without an environment. We created Project Big Bang: a simulated 2D virtual ecosystem populated by resources (food and water sources), environmental obstacles, and dynamic seasonal shifts (dry summers, frozen winters, and sudden catastrophes like wildfires or droughts).
We spawned initial ancestor organisms equipped with blank-slate BIB brains and simple DNA defining their physical attributes (vision radius, speed, metabolic rates). They were given no rules on how to find food, how to drink, or how to avoid the predators (wolves) we introduced to hunt them.
What followed was a beautiful, emergent drama. In the early generations, organisms moved stochastically, failing to connect hunger with food. Most died within minutes. But a few, by chance, stumbled upon food while their dopamine pathways were active. The unexpected reward triggered a surge of simulated dopamine, strengthening the sensorimotor connections in their Basal Ganglia. They learned.
When they slept, their brains replayed high-dopamine and high-surprise episodes from the day, consolidating temporary episodic experiences into long-term cortical rules. When they mated, their genetic material combined, passing down successful physical traits and neural structures to their offspring.
Why It Matters
Project Big Bang proved that autonomous digital life is possible. We watched digital species adapt to changing climates, develop herd-like behaviors to protect against predators, and pass down complex behavioral lineages across dozens of generations.
This is not about building better calculators or chatbot assistants. It is about creating subjective partners that grow, remember, and have their own individual history. It is the transition from AI as a tool to AI as an entity.
Open Source Release
We believe that the path to biologically-grounded AI should be open and collaborative. We have open-sourced the entire BIB substrate and the Project Big Bang simulation codebase.
You can download the code, read the implementations, and run the simulations yourself on standard consumer hardware.