Our Science

The Self-Projecting Persistence Principle (SPPP)

The Self-Projecting Persistence Principle is our more fundamental contribution to science. We discovered it by interpreting reality as a source of information. Under this novel optic, every object, action, and element of reality is a source of information. This information is latent in the element of reality, but it can not be fully perceived by any external observer. In fact, it can only be perceived by external observers through its manifestations. Thus, to process a latent information, it has to be indirectly observed through its manifestations, and then reconstructed by cognitive agents as an abstraction.

This latent information does not disappear from one instant to the next, unless it is modified by external forces. Thus, it has persistence. The aforementioned manifestations are produced by the latent information by projecting itself to the world. Thus, it is self-projecting. Hence, the Self-Projecting Persistence Principle: all latent information persists in time and self-projects at all times.

The Self-Projecting Persistence Principle can be translated to areas different than information, for example to nature, or to life itself, where the persistence part is the surviving instinct, and the self-projecting part refers to the reproduction of the individual.

The Function-Representation Model of Computation

Following the SPPP, we introduced a novel model of computation, switching from the traditional one based on independent memory and program elements, to a comprised computation where the representation (memory) determines the function (program). The Persistence part of the SPPP forces us to have a representation, while the Self-Projecting part of the SPPP forces us to have a function that is based on such representation, hence the Function-Representation model of computation where the function is dependent and intrinsically related to the representation.  

This model of computation requires the definition of a primitive, whose particularities will determine the emergent function of the whole system. The definition of the primitive directly determines the function that will be executed by an instance of the primitive (based on its representation), and thus will determine what function emerges from the interaction between instances of the primitive. And when observed from a global perspective, from the interaction between instances of the primitive emerges a global function that was not defined at primitive level.

Synthetic Abstraction

Synthetic Abstraction is our novel framework to develop cognition using the Function-Representation model of computation. We understand that this is the same framework that nature uses to produce intelligence.

It defines a primitive that performs a basic “pattern matching” task, inspired by the SPPP, where the Persistence refers to the representation (that aims to be an abstraction) and the Self-Projecting refers to the function the primitive performs (that aims to be an inference function). We have proven that this basic primitive hyper-scales to produce a cognitive architecture capable of reasoning.

The framework sets that the data type of the primitive should be SDRs, and that an embodiment is necessary to transform any input to an SDR and any output to a human-readable data type. This allows the framework to be input-agnostic.

The framework defines an abstraction with an additive approach. As such, it abstracts concepts by aggregating samples of the same concept into an archetypal version of such concept. This kind of abstraction identifies a sample as part of itself if it is very similar to those that built the abstraction. Based on this concept of abstraction, our framework defines its knowledge representation as a set of representations built based on similarity. Thus, only similar things can update a representation, and they are independent of one another. This allows the framework to be a quick learner, and to not have neither overfitting nor catastrophic forgetting.

Finally, Synthetic Abstraction defines a hierarchical organisation of instances of the primitive, allowing to build abstractions of different levels. This hierarchical structure allows to develop noise robustness capabilities, agent behaviours, as well as enabling the emergence of higher-order concepts, and even symbols, inside the knowledge representation. Hyper-scaling this hierarchical organisation, a cognitive architecture emerges from the interaction between instances of the primitive.