This paper presents an architecture of autopoietic intelligent systems (AIS) as systems of automated “software production”-like processes based on meta-engineering (ME) theory. A self-producing AIS potentially displays the characteristics of artificial general intelligence (AGI). The architecture describes a meta-engineering system (MES) comprising many subsystems which serve to produce increasingly refined “software-production”-like processes rather than producing a solution for a specific domain. ME-theory involves a whole order of MES and the ME-paradox, expressing the fact that MES can potentially achieve a general problem-solving capability by means of maximal specialization. We argue that high-order MES are readily observable in software production systems (sophisticated software organizations) and that engineering practices conducted in such domains can provide a great deal of insight on how AIS can actually work.

Key words: autopoiesis, meta-engineering, meta-learning, software production system, artificial general intelligence.

The development of artificial intelligence systems has to date been largely one of manual labor. This Con-structionist approach to A. I. has resulted in a diverse set of isolated solutions to relatively small problems. Small success stories of putting these pieces together in robotics, for example, has made people optimistic that continuing on this path would lead to artificial general intelligence. This is unlikely. “ The A. I. problem “ has been divided up without much guidance from science or theory, resulting in a fragmentation of the research community and a set of grossly incompatible approaches. Standard software development methods come with serious limitations in scaling; in A. I. the Constructionist approach results in systems with limited domain application and severe performance brittleness. Genuine integration, as required for general intelligence, is therefore practically and theoretically precluded. Yet going beyond current A. I. systems requires significantly more complex integration than attempted to date, especially regarding transversal functions such as attention and learning. The only way to address the challenge is replacing top-down architectural design as a major development methodology with methods focus-ing on self-generated code and self-organizing architec-tures. I call this Constructivist A. I., in reference to the self-constructive principles on which it must be based. Methodologies employed for Constructivist A. I. will be very different from today’s software development methods. In this paper I describe the argument in detail and examine some of the implications of this impending paradigm shift.

The development of artificial intelligence (AI) systems has to date been largely one of manual labor. This constructionist approach to AI has resulted in systems with limited-domain application and severe performance brittleness. No AI architecture to date incorporates, in a single system, the many features that make natural intelligence general-purpose, including system-wide attention, analogy-making, system-wide learning, and various other complex transversal functions. Going beyond current AI systems will require significantly more complex system architecture than attempted to date. The heavy reliance on direct human specification and intervention in constructionist AI brings severe theoretical and practical limitations to any system built that way. One way to address the challenge of artificial general intelligence (AGI) is replacing a top-down architectural design approach with methods that allow the system to manage its own growth. This calls for a fundamental shift from hand-crafting to self-organizing archi-tectures and self-generated code – what we call a constructivist AI approach, in reference to the self-constructive principles on which it must be based. Methodologies employed for constructivist AI will be very different from today’s software development methods; instead of relying on direct design of mental functions and their implementation in a cog-nitive architecture, they must address the principles – the “ seeds “ – from which a cognitive architecture can automatically grow. In this paper I describe the argument in detail and examine some of the implications of this impending paradigm shift.