A Brain-Inspired Artificial General Intelligence Anyone Can Develop

May 20, 2024

brain with neural connections

Over recent years, AI development has occurred at lightning speed. Now, by typing in a few words, users can produce lifelike images, customized instructions, or even code in seconds. Doctors and researchers can use AI to help with medical imaging interpretation, climate modeling, and more. From fraud detection to autonomous service robots, AI is transforming our world. However, the actual development of AI remains in the hands of those with special skills, such as programmers and system engineers. Only a small minority have the power to develop genuinely new AI behaviors, and yet the average citizen stands to lose the most as AI transforms our societies.

FEAGI (Framework for Evolutionary Artificial General Intelligence) is a free no-code solution that enables anyone to invent new AI models and develop behaviors for robots. It exists in both downloadable package (FEAGI Playground) and web-hosted (Neurorobotics Studio) options, and is accessible to anyone with basic literacy, computer usage skills, and computer access. With this platform, users shift from being consumers and passengers in AI development to sitting in the driver's seat.

Artificial deep neural networks, which undoubtedly have taken over the world in numerous applications, are the result of decades of research and development that are finally commercialized. Each AI model is developed and fine-tuned to solve a specific problem ("narrow AI") such as image generation, text to speech, content generation, knowledge discovery, etc. Narrow AI has proven to be extremely successful in a given domain and highly scalable when given access to extensive datasets and compute resources.

Considering the success of deep learning models, many efforts are being directed toward combining such models to enable multi-modal capabilities, expanding them beyond their narrow focus and tilt them toward Artificial General Intelligence (AGI). In contrast, FEAGI has been developed from the ground up to solve the AGI problem by offering an easy to use and resource efficient solution that can give rise to safe, effective, and trustable AI models that could play the role of brains for robots.

How does it work?


FEAGI is inspired by the development process and structure of the human brain.

To understand, let's take a quick tour of the brain. Its development begins with DNA. DNA is composed of sequences called genes, which contain the instructions for making proteins. These proteins are essential for the formation and functioning of neurons. Neurons are special cells that receive input, transmit signals, and more. Your brain contains some 86 billion neurons!
Neurons in the brain communicate with one another across a vast interconnected neural network, which can be represented in a map called the connectome. The communication between two neurons is achieved via electrical or chemical signals. This connection is called a synapse. Neuron activation and communication across synapses is essential to our ability to process external input, think, and learn.
For example, if you were to pet a bunny, sensory receptors in your skin would pass an electrical signal to your brain. When the signal reached the brain, it would activate specific neurons. Those neurons would release chemicals called neurotransmitters, which would travel along the synapse and activate the next neuron. As your brain processes these signals, you would perceive the softness of the bunny's fur.

This elegant system and its concepts form the foundational inspiration, architecture, and terminology of FEAGI. FEAGI is not one persistent brain, but a system that allows the creation and development of artificial brains.

FEAGI brain creation begins with a digital genome that provides the genetic blueprint for building a brain. As it reads genetic instructions, the system generates a connectome that is also referred to as the artificial brain. FEAGI has the ability to dynamically regenerate the brain regions when they are modified where a user can modify properties of a specific region and observe changes in behavior in real time.

When the brain receives stimuli, digital neurons become activated and influence other neurons connected to them. Artificial neurons in FEAGI have the ability to strengthen or weaken the connectivity among themselves based on the synchrony of activations. This process is referred to as synaptic plasticity, and is responsible for the formation of memories and learning.


A brain developed by FEAGI begins with an artificial genome. The genome is used to construct a brain with a connectome that contains data on cortical areas, neuron locations, etc. Then, the brain receives inputs and its neuronal connections are selectively weakened or strengthened in response, forming abilities or behaviors. Using the GUI (graphical user interface), you can edit the connectome to influence learning. Finally, the developed genome and/or brain can be saved and evolved.

Let's dive in a little deeper! We'll touch on the user interface, brain initialization, input, learning, development, evolution, and memory.
Note: It's important to remember that while FEAGI "brains" share many elements of human brains in digital form, they are merely complex digital systems that follow rules and randomization using code and data structures to flexibly grow--they are not whatsoever sentient. Programmers are encouraged to check out the code in our open source repo!

User Interface

The UI (user interface) is run on the Godot game engine right in your browser. It's separated into the brain visualizer and circuit builder.
The brain visualizer lets you see currently activated cortical areas. For example, in the below image, the brain is receiving webcam input and processing it as shown in the vision IPU (input processing unit).

This is just one of many possible cortical areas that can be viewed simultaneously. When a voxel (individual block) in a cortical area is manually selected, it turns green, and when activated (manually or by FEAGI), it turns red. In this example, the brain's active visual voxels are constantly changing in response to a person moving on the webcam.

The circuit builder enables you to change the structure and wiring of various brain regions. It includes any cortical areas from your starting genome or that you've added, such as motor OPU (output processing unit). Users can learn how to use the circuit builder via our tutorials on Neurorobotics Studio.

Brain Initialization

Humans have a genetic blueprint that determines how they will grow and develop from the earliest stages of life. FEAGI similarly has genomes, data objects that contain structural and behavioral information created via user development.
When a genome is uploaded, Python code uses the genome object to create the basic structure of a connectome. This connectome will have digital cortical areas and their connections, specific numbers of neurons in each block, etc. This determines the initial brain structure and its starting capabilities.
Programmer Insight: Genomes are stored as JSON files! Check out a basic one here.


FEAGI can accept a wide variety of input types, including virtual commands and movements, screen capture, camera, real-life robotic movements, and much more. Under the hood, these input types are converted to a form FEAGI can understand by a controller. You can watch FEAGI receive inputs in the brain visualizer, like the webcam example above.


After the brain has been initialized, you can make edits in the circuit builder and provide inputs such as robot movements. In response to inputs, new neuronal connections are made, and existing connections are strengthened, deteriorated, or eliminated.

You can watch your embodiment, such as a virtual car or physical drone, respond to FEAGI's commands. And you can see FEAGI react to inputs in the brain visualizer. Observing these will enable you to make informed changes.

Hebbian theory states that in biological brains, “Neurons that fire together, wire together.” FEAGI works the same way. When the brain receives input, the data and synapses (connections) of associated neurons are updated, or "fire." When they fire simultaneously, their synapses are strengthened. Once this happens enough times, the synapse becomes fixed and the association is learned.

A famous example of associative learning is Pavlov's dogs: by ringing a bell when the dogs were fed, he trained the dogs to salivate at the sound of the bell. FEAGI works in much the same way.

For example, FEAGI was trained to recognize letters in various styles in images. By feeding the letter "a" to the brain's UTF-8 (standard font) processing unit at the same time as an image version was fed to its visual processing unit, the association between the two was strengthened. Eventually, it learned to associate images of "a" with the UTF-8 letter. It could then move to learning "b," and so on. This is much like how children learn their alphabet!

FEAGI is also capable of reinforcement learning. When FEAGI's "pain receptors" are activated, connections between neurons that were activated at that time are weakened. (Note: "Pain" is just a concept represented by reducing stored connections in data.)


If FEAGI is not learning as intended, or you want to change what/how it learns, you can make edits using the circuit builder.

For example, the version of Pong we provide with FEAGI produces consistent noise when the ball is caught by the paddle, and random noise when it is missed. This leads to a strengthening of connections over time when the ball is caught, and a weakening when it is missed. If you wanted the brain to learn more effectively, a smart move would be to create a cortical area that remembers the path of the ball across many instances, so that FEAGI can learn to predict movements based on the angle of the path--the same way that we learn to intuitively predict the path of a ball that hits a wall at a certain angle.

Here is FEAGI receiving a mass of random noise when it missed the ball:

In essence, developing FEAGI is all about finding the best ways to strengthen neural connections for desirable behaviors, and/or weaken them for undesirable behaviors. Users can learn how to do this via our tutorials on Neurorobotics Studio, and of course by actually tinkering with the brain visualizer and circuit builder.


Developing a FEAGI behavior isn't always a matter of creating a single genome and sticking with it. Just like in nature, an effective behavior can optionally be honed through many generations of genomes and survival of the fittest. Note: We are still developing this feature.

When performing the evolutionary process, if the brain fails at a desired behavior, such as turning a robot left when you want it to go right, the brain dies and its performance data is stored in the genome. A new brain is formed: either from the same genome, or one that is similar, mutated (randomly altered in various parameters by a given percentage), or crossbred (merged) with another genome. The performance data of that new brain is stored in its genome, and so on. Over time, the best-performing genomes rise to the top.

Since the earliest known civilizations, humans have bred animals for ideal characteristics. Today, users of FEAGI essentially will be able to take on the role of breeders of digital brains, developing new capabilities via an artificial evolutionary process within hours.

Memory & Retention

By saving a genome, a user can capture the blueprint for an effective brain's connectome structure that has good learning potential for their desired behavior. The behavior itself, however, must be learned anew by each new brain created from a genome, unless the entire brain is saved. This additionally includes data on individual neurons and synapses, and saves both the learning potential and the behavior itself.

Programmer Insight: As of now, and without any compression or optimization, each neuron requires about 1 kilobytes of storage. The size of the connectome scales linearly with the number of neurons. If an entire human brain with about 100 billion neurons were simulated, it would require 100 terabytes of storage, which is well within the supported limits of our current database, MongoDB.

What can FEAGI do?

Thanks to its robust brain-inspired architecture, FEAGI is ideal for testing neuroscience hypotheses and for learning about neuroscience. Anyone from a professional researcher to curious enthusiast can benefit from our tutorials and interactive digital brains.

However, the real power of FEAGI lies in its codeless AGI development and relative ease of use for true AI and robotics development, democratizing the field for everyone. Although it may look complex at first, most people can pick up the fundamentals after just a few hours of tutorials and experimentation. Even professional roboticists and programmers will find that developing new capabilities and behaviors takes a fraction of the time with FEAGI. Rather than writing bespoke code, you can encourage or discourage behaviors by associating inputs, outcomes, and so on in cortical areas using the circuit builder, then let the brain learn on its own over rapidly occurring iterations.
At present, FEAGI has many useful abilities. It can learn to differentiate objects from background noise, in various lighting conditions, from different angles, and from varying distances. It can perform behaviors, such as robot movements, in response to webcam movements. It can effectively control sprites in virtual games to navigate environments. It can perform line tracking. It can recognize letters in images. These are just a handful of FEAGI's vast capabilities, and with further development, FEAGI will be capable of understanding object permanence and many more complex behaviors.
FEAGI is a continually evolving system, and even we haven't thought of everything it might be able to do today. Visit Neurorobotics Studio to explore, develop behaviors, share and evolve genomes with other users, and be instrumental in expanding the boundaries of what is possible in artificial intelligence and robotics!

© 2024 Neuraville Inc.

ServicesCareersContactNewsBlogPrivacy & Credits