Welcome to Darkstone Cybernetics

Developing a

General Cognition Engine

Lightweight, Advanced Artificial Intelligence

I'm Ashley Darkstone, Founder of Darkstone Cybernetics, and I've been designing a novel and practical, 'general AI system' since 2016.
A sustainable alternative to massive AI models, which learns and operates in real-time for virtually any purpose, especially suitable for robotics and IoT.
Currently under active development!

Subscribe to the Darkstone Cybernetics Substack for updates!

for

About This Engine...

What it isn't

What it is

This is not a Large Language Model (LLM) or a wrapper for one, nor even based on the Transformer architecture. It is not something a LLM conjured up, nor is it anything grandiose or exotic, requiring massive resources.

A novel and practical design, grounded in symbolic AI and machine learning, with cutting-edge research. Uses little-to-no traditional neural nets, so it remains lightweight and compatible with lower-powered devices and older technology. A highly sustainable form of AI.

What it could be

This engine is intended for real-time cognitive applications, such as

• Robotics: Generalized robotics, humanoid and otherwise. This can include vehicles of all sorts (ground, air, and space) and any technology that can be effectively 'roboticized' and supports the relatively low compute requirements.

• Edge AI/Internet of Things (IoT): Small, low-powered, always on AI, and capable of on-device learning and optimizing from its inputs over time. Ideal for things like, solar-powered environmental sensors that can learn locally.

• Industrial Monitoring & Safety: Real-time cognition across multiple cameras and other sensory inputs means it could spot and solve problems before they impact safety and productivity. Also great for consistent and adaptive quality control.

These are just a handful of the virtually endless potential applications of this very flexible engine. Essentially, with a modern computer and sensors, this engine can roboticize almost any device to operate in real-time. Also great for character and environment simulation.

The problems with current methods, and how this engine fixes them

Current methods relying on neural nets are often described as a 'black box' because we can't directly observe what's going on inside the AI, making it difficult to trust. This cognition engine uses different methods that make it auditable and editable, providing transparency about what's going on inside it, and what data it collects.

Unlike popular AI, which scales compute and electricity for modest gains, my lightweight engine reduces these needs, eliminating multi-gigawatt data centers. This makes it the most environmentally-friendly and sustainable option.

Being lightweight also means running locally on the user's own hardware, which further means it isn't reliant on an internet connection/cloud access to run, keeping user data private and secure.

Of course, an even bigger concern with current AI methods goes back to the 'black box' problem; LLMs are notoriously hard to make safe and aligned with humanity and the user. My cognition engine, on the other hand, has a built-in safety mechanism to not just steer or guide the AI, but force it to obey rules and laws as applicable. It is also selfless, so it doesn't have its own motivations, just what the user sets for it, making it user-aligned.

The biggest AI problem this engine solves...?

A lack of humanity. It's one reason LLMs are so hard to align and make safe. Also the reason for LLMs hallucinating supposedly factual information. They learn very differently, missing a lot of the common-sense rules that we live by, and the time-space domain that we operate within.

To counter this, my engine supports continuous learning, rule-building, and updating and optimizing itself over time. Unlike LLMs' pattern-matching, it uses human-like rulesets and one-shot learning for real-world accuracy.

Technical Challenges

So what will it take to build this thing?

Time.

That's it, really. The design and implementation are almost entirely determined and development has begun. It will take far fewer resources to develop and run than an LLM. Progress has been fairly smooth since I started working on this in 2016. A lot has changed in the AI space since then, not quite in the way I had hoped, but I see a lot of potential on the horizon. With my cognition engine, we can have safe, reliable, and sustainable AI.

ask@darknetics.com

To have the time to build this engine, my company is actively seeking funding. If you'd like to donate to or invest in my work, please check out the following link to my crowdfunding page, or contact me to discuss investment opportunities and potential returns.

Support This Vision

$50k supports a working demo within a year (by mid-2026), while $125k supports significant progress on a prototype over two years (mid-2027)! With your support I can fund salary, operational costs, and equipment. This doesn't just accelerate my work, it accelerates my ability to help others. Any support is appreciated!

Donate Through Ko-Fi