Perceptions of vaping addiction and cessation among Australian young adults: A qualitative study.

Human cognition is marked by its remarkable versatility, effortlessly transitioning from instant intuitive judgments such as recognizing faces or interpreting emotional cues to deliberate analytical reasoning involved in solving complex puzzles or making detailed plans. According to the influential dual-process theory popularized by Daniel Kahneman in (), these cognitive abilities stem from two fundamentally different modes of thought: , which operates quickly, intuitively, and with minimal cognitive effort; and , which is slower, more deliberate, analytical, and capable of managing demanding cognitive tasks (; ). System 1 continuously generates intuitions, impressions, and rapid judgments, often sufficient for routine or straightforward decisions. However, complex tasks, those involving multi-step planning, retrieval of external knowledge, and careful evidence integration necessitate engaging the more effortful System 2 (; ; ; ). However, even these complex reasoning tasks are rarely executed entirely by System 2 alone; instead, most human reasoning processes fluidly integrate rapid intuitive responses with occasional deeper reflection, triggered when intuition proves insufficient or potentially erroneous (; , ). Motivated by this understanding of human cognitive efficiency, we introduce PRIME (Planning and Retrieval-Integrated Memory for Enhanced Reasoning), a multi-agent reasoning framework designed explicitly to mimic this dual-process model. PRIME initially employs a (System 1) to produce intuitive answers rapidly. Subsequently, a specialized performs explicit self-reflection to critically evaluate the intuitive response. If self-reflection identifies uncertainty, potential errors, or risk of hallucinations, PRIME activates a more deliberative reasoning process (System 2), orchestrating specialized agents responsible for systematic planning, hypothesis generation, targeted knowledge retrieval, information integration, and comprehensive reasoning. This adaptive mechanism ensures efficient use of computational resources, engaging slower, computationally intensive reasoning only when necessary, significantly reducing unnecessary deliberation.