Harnessing Human Intelligence and Artificial Intelligence for Collaborative Problem Solving: a Framework for Enhanced Decision-making

Abstract

While Artificial Intelligence (AI) can now rival or surpass man in performing broad classes of data-intensive tasks such as narrow decision-making and calculation, complex, real-world decision-making, which requires case-dependent understanding, ethical values and creativity—hallmarks of human reasoning—remains exclusive to human intelligence. This paper presents a formal human-AI Collaborative Problem-Solving (HACPS) model, which proposes and empirically tests five falsifiable and explicit statistical hypotheses for task partitioning of decision sub-tasks between human and machine agents based on their comparative strengths. It employs a mixed-methods design, including a 214-practitioner structured survey of practitioners in the healthcare, financial, and technology domains; a controlled between-subjects decision-simulation experiment (N = 96, 4 conditions) to see if there is a significant difference in decision accuracy, decision time, and calibrated trust by experimental condition (Hypotheses 1–3, formal testing via one-way ANOVA, Tukey's HSD); and the qualitative analysis of themes from interviews with practitioners (N = 18) to test whether the underlying complementary principle is consistent—AI-led task ownership is significantly higher in analytical/data-intensive stages of the pipeline than in judgment/ethical stages (Hypothesis 5, formal testing via a paired-sample t-test). Results of all five hypotheses revealed positive. The Collaborative condition achieved statistically higher mean Trust (6.3/7) and decision accuracy (91%) compared to the other conditions (Human-only, AI-only, sequential Human-AI); decision time decreased by 52% and decision quality increased by 26 percentage points across eight cycles of collaboration (linear trend, r²> .97, p<.001); and AI-led ownership was about 52 percentage points higher at analytical stages versus judgment stages (t(213)≈50, p<.001), statistically confirming the complementarity principle. The paper includes the results of the statistical analysis used to arrive at the conclusions, with formulas and worked calculations provided as such, and introduces the validated, reusable framework, an empirically contingent task-allocation matrix, and a research agenda for scaling the human-AI collaboration to high-stakes organizational decision situations.