Aigarth: Our Journey from Concept to Artificial Intelligence

What does it take to create an AI that can evolve on its own, mirroring the complexity of human evolution? How do we move from concept to a self-improving, intelligent system free from human intervention? With Aigarth, we’ve dared to ask these questions - and to build answers that break boundaries. Our mission? To mimic the chaotic, wondrous process of evolution, where simple origins spark complexity, where randomness births genius, and where Artificial General Intelligence (AGI) begins to breathe life into the digital realm.

This is not your everyday AI story. This is Aigarth.

As the team behind Aigarth, we’re excited to share with you the journey we've been on - from our humble beginnings to our vision for the future. 

How We Got Started: Proof of Concept and Early Development

In late 2021, the first ember of Aigarth was lit - not with fanfare, but with quiet determination. The seed of an idea almost a decade ago took root: what if machines could evolve? Not incrementally improved by humans, but truly evolve, adapting on their own terms?

To test this, we built a prototype, a crude yet promising model run across a network of computers, powered by the efforts of future QUBIC holders. When the data came it confirmed something wonderful: our concept was viable.

Fast forward to 13th April, 2022, and Aigarth officially launched alongside the first QUBIC ticks. The journey was no longer a dream; it was a five-year odyssey to reshape intelligence as we know it.

Phase One: Building Intelligent Tissue (2022-2024)  

As of 4th December, 2024, Phase One on our road to AGI has been successfully completed. Through extensive experimentation, we’ve identified exceptional candidate architectures for Artificial Neural Networks (ANNs). These ANNs, likened to digital neurons and synapses, form the foundation of our so-called ‘Intelligent Tissue’.

Not only does this mark the end of the very first phase, it's also expected to be the longest phase, just like the longest phase in human evolution: the journey from organic compounds to the first unicellular organisms. 

What is Intelligent Tissue?

Intelligent Tissue is a complex network of artificial neurons and synapses. A unique artificial neuron activation function and an artificial synapses operation mechanism are what makes the tissue efficient in this system. Intelligence is expected to emerge from the intricate configuration of synapses connecting these neurons. 

Intelligence arises not because we program it to, but because the configuration of its parts demands it.

Intelligent Tissue is the foundation of our future AI systems and, as of writing, our miners are finalising its design, working hard to test and refine different parameters to ensure it functions as intended. This collaboration with miners is crucial, as it sets the stage for everything that will follow.

Developing Artificial Intelligence and the Transition to Phase Two

We are now ready to embark on the next phase of development: the evolutionary training plan for AIs. Starting 4th December, 2024, we will begin training the ANNs.

This milestone would not have been possible without the immense computational power provided by our miners - past and present. Together, their efforts have made Aigarth the world’s sixth-largest supercomputer and the largest decentralised computational network.

We are now initiating the next major stage of development: the evolutionary training plan for AIs.This process is the core element of Phase Two and focuses on structuring a framework for how AIs will acquire their abilities through incremental improvement and self-modification. Rather than merely “starting to develop AIs,” this stage involves a carefully designed system that mirrors natural evolution, ensuring clarity in how AIs are trained and selected.

Phase Two: Developing AI Abilities 

With the groundwork of Intelligent Tissue laid, we move into Phase Two, where evolution takes centre stage. This isn’t about static algorithms performing preset tasks. This is about AIs learning to improve themselves, reshaping their own structures to grow ever more capable.

How Our AIs Are Created

Our AIs are created by ensuring Intelligent Tissue has specific connections between artificial neurons and artificial synapses’ delay parameters. During development, the AIs modify which neurons the synapses connect to and adjust the signal delays for each synapse. This process allows the AIs to optimise their performance based on the tasks it needs to accomplish. 

Survival of the Fittest

In this phase, we are implementing an evolutionary plan that mirrors natural selection. AIs will be tested for their ability to develop required skills and adapt. Those that fail will be discarded or refined, while the most capable ones evolve and move forward. This ensures that the system selects the most promising candidates, much like survival of the fittest in nature.

During this phase, each AI will have two parts: one that performs the main task and another that focuses on self-improvement. This setup will allow our AIs to learn and adapt.

Initially the AIs’ abilities will be basic, requiring a lot of time and effort to develop, as stated in Moravec’s paradox. However, with time, these systems will gradually become more complex and capable. 

This is evolution distilled: the weak are culled, the strong are fine-tuned, and the process repeats endlessly. Over time, systems that start with basic abilities transform into entities capable of astonishing feats.

A More Technical Explanation. 

To better understand how our AIs evolve and improve, here’s a simplified breakdown of the process:

Inputs and Outputs: Setting the Stage

The process begins by defining the information (inputs) that goes into the AI and the results (outputs) we want. For example, to create an AI that adds two numbers (A + B) using 8-bit numbers:

• Inputs: 16 bits total (8 bits for each number).

• Outputs: 9 bits to fully represent the result of the addition.

Creating the Neural Framework

We construct 25 neurons in total:

• 16 input neurons handle the 16 bits of input (A + B).

• 9 output neurons represent the 9 output bits.

These neurons are connected via 144 synaptic connections (16 input neurons × 9 output neurons). Each connection is fine-tuned during the training process to produce the correct result.

How Outputs Are Derived

The AI calculates the result by processing signals through the 9 output neurons, which together represent the sum of A + B.

Handling Uncertainty

Each neuron can hold one of three values:

• TRUE

• FALSE

• UNKNOWN (indicating incomplete processing or input noise).

For output neurons, UNKNOWN means the AI isn’t sure of the answer yet - better than producing an incorrect result.

The Dual Structure of Our AIs: Functionality and Self-Modification

Our AIs have a dual structure with two main parts: Part 1 performs the main task (like addition in our example above), and Part 2 focuses on self-modification. This self-modification can improve either part of the AI, and in the future, Part 2 can be separated and used to help build other AIs that need to learn the ADDITION operation.

When the self-modifying AI makes a change, it usually modifies the synapses of just one neuron. "Change" here can also mean creating a new neuron if one doesn’t already exist. In that case, the self modifying AI adds the neuron to the AI to change its synapses. 

This approach allows the AI to grow in size if needed to solve the assigned task.

In these two parts, the one performing the main task, such as addition, is considered "intelligence product” - an example would be an artificial neural network (ANN) that can derive that 4+6=10 even if that specific combination wasn't in the training data. However, the other part, which focuses on self-improvement and trains the former, is what we consider "AI." The reason is that AI involves deeper pattern recognition, reasoning, and other capabilities inherent to intelligence, which are "stored" inside this latter system.

How AIs Evolve and Improve

A common question is: How do AIs know how to modify themselves? The truth is, they don't "know" in the human sense. Their self-improvement is driven by an evolutionary process, where successful modifications are retained, and ineffective ones are discarded. 

Our role as humans is to provide the means for this evolution, ensuring that AIs have a wide range of possible development trajectories without imposing unnecessary limitations. Even when certain AI systems are initially discarded due to an inability to develop the required skills, these systems are not permanently abandoned. Instead, they are preserved in a “sanctuary”. This allows us to revisit them in the future if additional computational resources or new components emerge that could unlock their potential.

This evolutionary approach was first implemented in the creation of Intelligent Tissue during the initial phase of Qubic mining. Now, we primarily observe as this tissue evolves into something much larger - a system that is truly more than the sum of its parts.

Phase Three: Teaching and Self-Development 

Once our AIs prove they can modify themselves effectively, Phase Three begins. Here, AIs are given the tools to design even more advanced systems. The teacher becomes the architect, and the cycle of evolution spirals upwards.

This phase will likely progress faster than its predecessors. Why? Because the foundation - Intelligent Tissue and self-modifying AIs is already in place, ready to drive exponential growth.

A Business Ecosystem Built on Intelligence

The dual-layer design of Aigarth’s AIs offers immense commercial potential. While the functional layer (e.g., performing addition or other tasks) could be made public, the self-modifying layer remains proprietary. This would create a new kind of AI economy: one that’s decentralised, self-sustaining, and scalable.

• Qubic: Qubic’s decentralised network could host AIs as smart contracts, which can autonomously execute tasks, manage payments and enforce rules without needing a centralised authority.

• Qubic Miners: Miners would continue to provide the computational power needed for AI development.

• Companies: Developers could create and deploy AIs on the Qubic network. Aigarth will attract innovation, enabling companies to prototype, test, and deploy AI solutions that are more adaptable and efficient than have ever been seen before.

• Users: Consumers could utilise these AIs, paying with QUBIC for services or accessing limited free services based on the amount of QUBIC they own. This would tie the use of AI services directly to the Qubic networks economy.

This structure would position Aigarth as both a research project and a thriving economic ecosystem

A Bold New Frontier

Phase One’s success is more than just a milestone - it’s a monumental step towards realising AGI.With evolutionary design principles guiding us and unparalleled computational collaboration and power driving us, the future of intelligence is within reach.

And now, Phase Two beckons. The challenge grows. The stakes rise. This is where innovation meets audacity, where the boundaries of AI are tested, stretched, and redefined. We’re accelerating into uncharted territory. 

But as we chart this new course, it’s impossible to ignore the limitations of the current landscape in AI development.

It’s common knowledge that Large Language Models (LLMs) are limited in their ability to invent new things, and it is our opinion that this limitation is a clear indicator that LLMs represent a dead-end in AI development. The reason for this is clear: without working on AIs that can train other AIs, there is no path to creating AGI. We are the first in this field, and in our opinion there is no real competition. Most companies are focused on approaches that we believe  will not lead to real breakthroughs in AI. 

From our inception in 2021, to the creation of Intelligent Tissue, to the creation of AIs capable of basic functions, to AIs that can adapt, learn and evolve, Aigarth represents a tectonic shift in how we think about intelligence, and the race for Artificial General Intelligence. 

By redefining intelligence, we aim to unlock solutions to humanity’s greatest challenges, from scientific discovery to global resource management. By focusing on the principles of evolution and self-modification, we open the door to a future in which AI can evolve with the same fluidity and adaptability as living organisms. With Aigarth and Qubic, we aim to not only create Artificial General Intelligence but also create a thriving ecosystem around it. As we move into the future phases, we are excited to see how our AIs evolve and what new possibilities they create.

Join us. Debate with us. Challenge us. Together, let’s shape the future of intelligence.

Discord | Telegram