Neuro-Psychoanalytical Inspired Automation Systems

The research project addresses the demand of context- and situation-aware environments in computational areas. This means, that it is not sufficient to collect lots of sensor data as fast as possible and react on them by several predefined rules. A system has to extract the higher semantic meaning of a current situation. This advantage of situation awareness in a system is used to develop corresponding action plans within a decision unit that works on a very high semantic level. Having a look at the requirements of such a system, it has been decided to use a bionic approach to meet these demands. There are many different, partly conflicting models available, describing the human perception and decision making. They are describing the human behavior from different points of view. The neuro- psychoanalytical model of the human brain which was essentially improved in the last 20 years, is the most complete functional model of humans mind. A first approach of a neuro-psychoanalytical based concept has been developed at the Institute of Computer Technology with assistance of various technical centers of excellence and close collaboration with neurologists and psychoanalysts. The research shall discover and define the elements that are needed between the two borders of perception and action. The main focus is to improve the existing concept with further cooperations of the neuro-psychoanalytical association and to describe a technical model that enables an implementation within a technical system. This is done by transferring it into a technical model describing the functionality of each relevant function block. These blocks can then be interpreted and implemented in further research. It will be part of the research to implement several modules for testing purposes, using simulation environments. Autonomous embodied agents are used as a target platform for the developed model. These agents are equipped with sensors, indicating the status of their surrounding as well as their internal status, comparable to the homeostasis of the humans body. They can manipulate their environment with actuators. The agent has to perceive and recognize the current situation and take decisions that guarantees to hold the internal values (representing the homeostasis) within an optimum. Although, the target platform of the perception and decision unit are embodied autonomous agents, a further research in transforming this unit into other applications is considered. This can be e.g. applications in robotics, building automation, web agents or other areas making decisions for actions on perceived data.

The project`s aim was to research the possibility of applying neuro-psychoanalytical concepts to artificial intelligence. The yielded results are a rich model based for a decision unit and a sample implementation in a simulator. To generate the model, an interdisciplinary research team worked in a top-down manner to extract technically feasible terms from psychoanalytical theories. The starting points were concepts like the psychoanalytical entities Id, Ego, and Superego. Step by step, these concepts were divided into sub-concepts until the descriptions reached a detail grade which allowed it to implement them. This project is the first one which went the whole distance from the plain idea to implementing the result in a simulator. Thus, not all descriptions were available which are needed to define sub-concepts of the model. They have to be generated by the interdisciplinary research team such that they do not contradict the already used psychoanalytical concepts. The result of this top- down design process is a rich functional model. Bodily needs are in the form of drive representations as well as information originating from body sensor like stomach tension and world sensors like vision provide the three input sources for the model. The Id converts the drive representations into a psychic process-able format. Ego functions enrich the incoming sensor information with knowledge extracted from already experienced situations. The defense mechanisms of the Ego operate as filter. Non-fitting contents are blocked and send back to the Id. There they are changed and/or split apart until they reach a shape where they can pass the filter. Based on the filtered information stream, the Superego provides rules of fitting bans, bids, and rewards. The decision making - which is part of the Ego - uses the incoming information and rules together with stored knowledge to produce a decision of which action is to be performed best at the moment. Finally, the actuators of the body execute the actions and the next iteration of the decision process starts. This model is implemented for evaluation purposes into virtual embodied agents roaming in an artificial life simulation. There they are exposed to predefined settings and their resulting behavior is examined and - if necessary - adaptations to the model/implementation are made. The advantage of the produced neuro-psychoanalytically inspired decision unit to other artificial intelligence based control architectures is that it is based on a holistic theory - every aspect from sensing to decision making to acting is covered by neuro- psychoanalysis - and does not focus on a special tasks. Thus, once the model is designed and implemented it can be applied to all various kinds of tasks. Possible future applications for this model include but are not limited to robots, factory automation systems, computer games, airplanes, and cars.