Towards Self-constructive Artificial Intelligence: Algorithmic basis (Part I)

TL;DR

This paper introduces Self-constructive Artificial Intelligence (SCAI) as a framework for artificial general intelligence inspired by brain-like self-organization. It centers on three organizing principles—self-growing, self-experimental, and self-repairing—and employs Schema-based Learning (SBL) to autonomously construct predictive schemas, dual schemas, and goal schemas that support open-ended, modular internal models. By integrating forward (predictive) and inverse (dual) models with goal-driven schemas and a distal learning mechanism, the approach enables robust planning, learning, and adaptation through cause–effect experimentation and coherence maximization. The work also provides a pseudo-algorithm and mechanisms for open-ended schema construction and causal-relations discovery, arguing that SCAI offers a general, embodied framework potentially applicable to navigation, motor control, and decision-support domains, with future extensions to social and ethical considerations.

Abstract

Artificial Intelligence frameworks should allow for ever more autonomous and general systems in contrast to very narrow and restricted (human pre-defined) domain systems, in analogy to how the brain works. Self-constructive Artificial Intelligence ($SCAI$) is one such possible framework. We herein propose that $SCAI$ is based on three principles of organization: self-growing, self-experimental and self-repairing. Self-growing: the ability to autonomously and incrementally construct structures and functionality as needed to solve encountered (sub)problems. Self-experimental: the ability to internally simulate, anticipate and take decisions based on these expectations. Self-repairing: the ability to autonomously re-construct a previously successful functionality or pattern of interaction lost from a possible sub-component failure (damage). To implement these principles of organization, a constructive architecture capable of evolving adaptive autonomous agents is required. We present Schema-based learning as one such architecture capable of incrementally constructing a myriad of internal models of three kinds: predictive schemas, dual (inverse models) schemas and goal schemas as they are necessary to autonomously develop increasing functionality. We claim that artificial systems, whether in the digital or in the physical world, can benefit very much form this constructive architecture and should be organized around these principles of organization. To illustrate the generality of the proposed framework, we include several test cases in structural adaptive navigation in artificial intelligence systems in Paper II of this series, and resilient robot motor control in Paper III of this series. Paper IV of this series will also include $SCAI$ for problem structural discovery in predictive Business Intelligence.

Figures (4)

• Figure 1: The agent immersed in the environment E. The agent receives input from the environment through the receptors in primitive perceptual schemas (e.g. $S_k$) and affects the environment through the effectors in primitive motor schemas (e.g. $S_m$). All sorts of internal interactions take place defining in some sense another "internal environment". The different internal drives affect the inner dynamics (e. g. by setting goals to be achieved). This diagram only illustrates one example for each of the different possible internal interactions
• Figure 2: Representation of the cause-effect relations in the simplified example consisting of five schemas, three of which are motor schemas A, B, C and two of which are perceptual schemas 1, 2. The black squares represent the existence of a cause-effect relation between the respective motor and perceptual schemas with that specific delay $\tau_i$. On the other hand, the white squares represent the lack of any cause-effect relation between the respective motor and perceptual schemas.
• Figure 3: Unexpected can be generated in two cases: (C.c) by a new predictive cause-effect relation discovered by the cause-effect dynamics (as described in section 4.1); and (C.a) when there is an incoherence, that is, the predictive response does not match the actual observed response $(o^x(t+1) - \hat{o}^{x,y}(t+1)) \neq 0$ and, at the same time, either the cause or the effect schemas are inactive, i.e., $o^y(t) = \sharp$ or $o^x(t+1) = \sharp$.
• Figure 4: Generalized process of schema(s) construction. Construction of the new predictive schema $P^{x,y}$ and its dual schema $S^{y,x}$. In turn, cause schema $S^y$ is adapted to instantiate a modulatory input port from the dual schema (represented by dashed lines). This in turn adapts its behavioral specification as now it has to take into account this instantiated modulatory signal.