Towards Strong AI: Transformational Beliefs and Scientific Creativity

TL;DR

This paper tackles how to model scientific creativity—a key feature of strong AI—by introducing the Transformational Belief (TB) framework, a dynamic, data-driven process that iterates through Creation, Exploration, and Evaluation to generate transformed beliefs. It grounds TB in historical case studies (notably the Neptune/Uranus discoveries) and philosophical accounts of discovery, and demonstrates TB with a simple many-normal-means illustration, linking it to foundational ideas in Bayesian inference, fiducial reasoning, Dempster–Shafer theory, and inferential models. The work integrates these traditions to propose a unified logic of science within a TB architecture, and even conducts exploratory TB-evaluation using a language model, suggesting a pathway for computational creativity in AI. Collectively, the TB framework offers a principled route to quantify and foster scientific creativity, with potential to bridge epistemic philosophies and to guide future AI systems toward genuine exploratory problem-solving. The approach has practical implications for designing AI that can autonomously generate hypotheses, test them against data, and adapt beliefs as new information arrives, moving toward robust, creative, and interpretable strong AI.

Abstract

Strong artificial intelligence (AI) is envisioned to possess general cognitive abilities and scientific creativity comparable to human intelligence, encompassing both knowledge acquisition and problem-solving. While remarkable progress has been made in weak AI, the realization of strong AI remains a topic of intense debate and critical examination. In this paper, we explore pivotal innovations in the history of astronomy and physics, focusing on the discovery of Neptune and the concept of scientific revolutions as perceived by philosophers of science. Building on these insights, we introduce a simple theoretical and statistical framework of weak beliefs, termed the Transformational Belief (TB) framework, designed as a foundation for modeling scientific creativity. Through selected illustrative examples in statistical science, we demonstrate the TB framework's potential as a promising foundation for understanding, analyzing, and even fostering creativity -- paving the way toward the development of strong AI. We conclude with reflections on future research directions and potential advancements.

Figures (4)

• Figure 1: Initial estimates of the number of components by sample size using BIC (left) or 80-20 cross validation of the negative log likelihood (right) when the true model has $K=3$, $\theta = (-2\quad 2 \quad 5)^\top$ and $\pi = (0.3\quad 0.5\quad 0.2) ^\top$.
• Figure 2: An illustration of the effect of one new observation on model specification for an initial sample with $K_n = 1$, $X_i \sim N(0,1)$, and $i \in 1,\dots,n=10$.
• Figure 3: The fiducial set-valued mapping $\{\theta: F_\theta(x-1) < 1-u \leq F_\theta(x)\}$ for $u$ given $x$. The gray area is for the case with $n=10$ and $x=4$.
• Figure 4: The plausibility curve of the binomial example in Section \ref{['ss:ims']} for the case with $n=10$ and $x=4$. The 90% plausibility interval is given by the points with $\hbox{Pl}_X(\theta)\geq 0.1$.