The attempt to make fully autonomous Artificial Intelligence can create a lot of confusion in people. For some – it is simply not possible. To others – if possible, then it is too scary.
But what if such advanced and self-learning AI would be fully transparent in the same time?
These are the attributes of our technology, where we can examine and understand everything what is stored in memory embedded into neural networks and associated decision models.
The machine can learn a lot by observing, make own assumptions and test them in the internal event simulator.
Most of just described attributes are inaccessible to the current AI technology. Instead relying on human supervision and collection of datasets with thousands of examples.
Relying on human supervision and collection of datasets is a dead end.
That is why we had to invent our own neuro-symbolic architecture, able to learn both low-level sensory information and high-level symbolic information – about associations between objects, the causality and common-sense knowledge.
Our innovative 3D Neural Networks allow the computational units to connect to any other ones, regardless of their virtual „location” – as we are not restricted by any physical limitations. The data can not only flow from left to right or right to left – but in all directions.
We use only the most useful parts of the architecture – that is how we can make it both extremely capable and energy-efficient at the same time. The flow of data can start and end in any fragment. One unit can initiate some process or be in the middle of a different one.
Incremental real-time learning mechanisms modify only relevant units, not affecting previously stored knowledge.
We know from neuroscience that human brain is not a randomly initiated network of networks – and we used a similar approach in our technology.
With high-level structure and low-level adaptation we can get the explainability, energy-efficiency and real-time learning.
Because both the system and the user can know how (and why) data is routed through specific computational units.
We extend our 3D neural networks with a whole range of additional modules – inspired by subcortical components of the brain. By adding several numeric parameters, mimicking neurochemicals in the brain – we allow machine to decide what to prioritize.
Is it better to explore the environment or act immediately. Reason, simulate events or interact with objects or people. Seek for good enough solution or the absolutely best one.
The system or the user can make changes in real-time, modifying exactly those elements that need it – leaving the rest untouched.
We are absolutely convinced that this is the future of Artificial Intelligence.
And thrilled to take you with us on this journey…