We have designed a machine that becomes increasingly better at behaving in underspecified circumstances, in a goal-directed way, on the job, by modeling itself and its environment as experience accumulates. Based on principles of autocatalysis, endogeny, and reflectivity, the work provides an architectural blueprint for constructing systems with high levels of operational autonomy in underspecified circumstances, starting from a small seed. Through value-driven dynamic priority scheduling controlling the parallel execution of a vast number of reasoning threads, the system achieves recursive self-improvement after it leaves the lab, within the boundaries imposed by its designers. A prototype system has been implemented and demonstrated to learn a complex real-world task, real-time multimodal dialogue with humans, by on-line observation. Our work presents solutions to several challenges that must be solved for achieving artificial general intelligence.

Bounded Recursive Self-Improvement. E. Nivel, K. R. Thórisson, B. R. Steunebrink, H. Dindo, G. Pezzulo, M. Rodriguez, C. Hernandez, D. Ognibene, J. Schmidhuber, R. Sanz, H. P. Helgason, A. Chella, G. K. Jonsson

Article @ arXiv

Robot control software endows robots with advanced capabilities for autonomous operation, such as navigation, object recognition or manipulation, in unstructured and dynamic environments. However, there is a steady need for more robust oper- ation, where robots should perform complex tasks by reliably exploiting these novel capabilities. Mission-level resilience is re- quired in the presence of component faults through failure recovery. To address this challenge, a novel self-adaptation framework based on functional knowledge for augmented autonomy is presented. A metacontroller is integrated on top of the robot control system, and it uses an explicit run-time model of the robot’s controller and its mission to adapt to operational changes. The model is grounded on a functional ontology that relates the robot’s mission with the robot’s architecture, and it is generated during the robot’s development from its engineering models. Advantages are discussed from both theoretical and practical viewpoints. An application example in a real autonomous mobile robot is provided. In this example, the generic metacontroller uses the robot’s functional model to adapt the control architecture to recover from a sensor failure.

A self-adaptation framework based on functional knowledge for augmented autonomy in robots. Carlos Hernández, Julita Bermejo-Alonso and Ricardo Sanz. Integrated Computer-Aided Engineering 2018, IOS Press

Article @ ASlab

This is an editor's introduction to a special issue of the Journal of Artificial General Intelligence on the topic of Self-Programming and Constructivist Methodologies for AGI

Thórisson, K. R., Nivel, E., Sanz, R., and Wang, P. (2012). Approaches and assumptions of self-programming in achieving artificial general intelligence. Journal of Artificial General Intelligence, 3(3):1–10.

Article @ AGI Journal

Journal of Mind Theory An ASLab Journal We all realise that the number of publications in the field of cognitive science is continuously growing. This makes the task of getting a state-of-the-art picture of the field an impossible task for any normal human. We feel the need for simplification and focusing. We believe that the pursuit of the ultimate understanding of mind shall be easier if we are able to get rid of the decorative literature. While that kind of text usually embellishes the many insights on the nature of mind, a narrower focus on the very core issues is absolutely neccessary. Succinctness becomes a major target. Hence, in the old way of the hard sciences, we strive for terse formalisations that will minimise the need of ink and paper and will hopefully convey precise, non-interpretable expressions of theories or hypotheses on mind nature.

If we are succesful in this simplifying and focusing attempt, then there will be a single journal in the reading pile.