Dynamic Intelligence Ceilings: Measuring Long-Horizon Limits of Planning and Creativity in Artificial Systems

TL;DR

The paper tackles the problem that static benchmarks poorly capture long-horizon development and may miss premature convergence in AI systems. It introduces Dynamic Intelligence Ceilings (DIC), a moving frontier of effective intelligence defined as $\mathcal{C}(t) = \max_{\mathcal{T}} \; \mathcal{I}\bigl(t \mid \mathbf{R}(t), \mathbf{J}(t), \mathbf{S}(t)\bigr)$, and operationalizes it with two estimators: the Progressive Difficulty Ceiling $\widehat{\mathcal{C}}_{\mathrm{PDC}}(t)$ and the Ceiling Drift Rate $\mathrm{CDR}$. The authors design Workshop World, a procedurally generated benchmark with a difficulty vector $\delta=(H,K,C,A)$ to jointly assess long-horizon planning and structural creativity, enabling analysis of frontier formation, saturation, and novelty preservation. Empirical results show that some systems deepen exploitation within a fixed solution manifold, while others sustain frontier expansion, with positive drift coupled to persistent structural novelty. The framework is domain-agnostic and offers diagnostic tools to identify premature fixation of capabilities, supporting the development of AI that can continue to grow along meaningful trajectories rather than plateau early.

Abstract

Recent advances in artificial intelligence have produced systems capable of remarkable performance across a wide range of tasks. These gains, however, are increasingly accompanied by concerns regarding long-horizon developmental behavior, as many systems converge toward repetitive solution patterns rather than sustained growth. We argue that a central limitation of contemporary AI systems lies not in capability per se, but in the premature fixation of their performance frontier. To address this issue, we introduce the concept of a \emph{Dynamic Intelligence Ceiling} (DIC), defined as the highest level of effective intelligence attainable by a system at a given time under its current resources, internal intent, and structural configuration. To make this notion empirically tractable, we propose a trajectory-centric evaluation framework that measures intelligence as a moving frontier rather than a static snapshot. We operationalize DIC using two estimators: the \emph{Progressive Difficulty Ceiling} (PDC), which captures the maximal reliably solvable difficulty under constrained resources, and the \emph{Ceiling Drift Rate} (CDR), which quantifies the temporal evolution of this frontier. These estimators are instantiated through a procedurally generated benchmark that jointly evaluates long-horizon planning and structural creativity within a single controlled environment. Our results reveal a qualitative distinction between systems that deepen exploitation within a fixed solution manifold and those that sustain frontier expansion over time. Importantly, our framework does not posit unbounded intelligence, but reframes limits as dynamic and trajectory-dependent rather than static and prematurely fixed. \vspace{0.5em} \noindent\textbf{Keywords:} AI evaluation, planning and creativity, developmental intelligence, dynamic intelligence ceilings, complex adaptive systems

Figures (2)

• Figure 1: Planning success rates across increasing difficulty levels $\delta$ at multiple developmental phases. Each curve corresponds to a distinct evaluation phase under identical resource constraints. The horizontal dashed line indicates the success threshold $\tau=0.7$ used to estimate the Progressive Difficulty Ceiling (PDC). Static ceilings manifest as early plateaus, whereas dynamic ceilings appear as rightward shifts of the performance frontier over time.
• Figure 2: Temporal evolution of the Progressive Difficulty Ceiling (left axis) and mean structural novelty among solved instances (right axis). A positive Ceiling Drift Rate accompanied by sustained novelty indicates genuine frontier expansion rather than convergence toward repetitive solution patterns.