Neural Networks Help Unravel Complexity Of Self-awareness

The doctoral research, conducted by Milton Martínez Luaces, was directed by the School of Computing's professor Alfonso Rodríguez-Patón.

This research represents a dual advance in the modelling of awareness-associated cognitive functions. On the theoretical side, it applies the theory of informons and holons to awareness structures. An informon is an information entity. It can take the form of data, news or knowledge. The term holon refers to autonomous entities that act both as a part and as a whole.

On the practical side, awareness-associated processes were simulated and influenced the design and development of software models. To do this, modular neural networks were used to develop multi-entity simulators. The networks were used to simulate several scenarios of interaction between their own potential and that of the systems with which they interact.

Self-awareness and the time dimension

In the case of human beings, self-awareness does not imply just an abstract image of what one is, but also an image of one’s trajectory throughout time. This research also modelled the time dimension of self-awareness using time-delay neural networks. These networks have shown, in different interaction scenarios, that the image that each entity has of its qualities in the past or its expectations for the future has an impact on how it interacts with other entities. For example, a figure reveals how interaction with other artificial entities, in this case in a competitive scenario, enables these entities to develop self-awareness.

In the same way as individuals tend to form groups with common interests and, as a result, develop a sense of belonging at more than one level, artificial entities may interact similarly to achieve a particular purpose.

Possible applications

The proposed models and their neural network implementations have basically two possible fields of application.

First they are useful for research into plausible models for explaining biological models. This approach tackles the problem of awareness through the formulation and computer simulation of artificial models.

Second these models could, from a practical viewpoint, also be applicable in the field of artificial intelligence. This would involve the prospective deployment of some of these features in multi-purpose artificial systems (robots, softbots, multiagent systems, etc.). In actual fact, several researchers are working on applying software models to robots in competitive or collaborative environments.

The models that have so far been implemented in artificial intelligence systems to approximate awareness are the result of huge simplifications because it is hard to model such complex reality.

No doubt this also applies to the models presented in this research. However, many of the concepts suggested by the proposed models and the high-level modelling principles match the results of animal experiments and observations and outcomes of non-invasive experiments conducted on human beings.

These early advances are not in themselves a solution to the complex open questions, the researchers note. Even so, many others are tackling the subject of awareness from different angles, configuring an attack on several fronts, such as cognitive psychology, neurobiology and artificial intelligence.