Search papers, labs, and topics across Lattice.
This study investigates the alignment of thinking effort between humans and large reasoning models (LRMs) in the context of abductive reasoning, which presents unique challenges that prevent models from exploiting shortcuts. By comparing human reaction times with LRM reasoning traces, the authors reveal that both humans and LRMs exhibit similar reasoning costs and error patterns. Notably, the use of advanced decoding methods that allow models to explore multiple reasoning paths enhances this alignment, suggesting a deeper understanding of cognitive processes in AI systems.
LRMs mirror human reasoning effort in abductive tasks, revealing shared cognitive challenges and error patterns that could redefine our approach to AI reasoning.
A major question in cognitive modeling concerns the behavioral alignment between large language models and humans across linguistic and non-linguistic tasks. Unlike standard LLMs, large reasoning models (LRMs) are optimized with reinforcement learning from verifiable rewards, encouraging correct solutions to reasoning tasks rather than preference-aligned responses. Recent work (de Varda et al., 2025) investigates the cost of thinking in humans and LRMs by comparing human reaction times with model reasoning traces across a range of reasoning tasks. We isolate this alignment by turning to abductive reasoning: unlike deductive tasks, its difficulty cannot be inferred from formal structure and offers no shortcuts a model could exploit to mimic effort without genuine search, providing firmer ground for empirical claims of shared effort. We find further evidence of alignment between LRM and human reasoning effort, as well as evidence that models and humans tend to make similar errors. Finally, we show that decoding methods that let models explore multiple reasoning paths increase alignment in reasoning cost between humans and LRMs across the three models tested.
