Systems, Algorithms, and Cognitive Models to Predict Individual Human Reasoning (PREDIR)

该项目专注于认知人工智能研究的一个核心挑战：开发一个系统，该系统能够预测一个人的信息处理，特别是基于少数观察到的例子得出的未来结论。通过关注个体，这个项目与认知心理学的方法有所不同，例如，认知心理学由于统计和数据分析的原因，忽视了个体实验参与者的越轨行为，因此只关注聚合实验数据和识别一般认知现象的认知模型。在这个项目中，我们将重点放在个人的角度来重新分析数据，开发能够在稀疏背景信息下预测个人未来结论的计算系统，并分析和评估认知心理学、认知科学和数据驱动方法(如推荐系统)中现有方法的性能。特别是，我们将确定、然后开发并系统地改进方法，这些方法不仅能够实现准确的预测，而且还可以确定数据驱动的上限。此外，哪些机制是人类推理的基础，以及这些见解如何反映在自适应认知建模方法的自动开发中的问题将得到澄清。该项目开发了能够自适应和预测性地调整个人思维过程并预测结论的认知人工智能模型。

This project focuses on a central challenge of cognitive AI research: to develop a system that is able to predict an individual's information processing and in particular future conclusions of based on few observed examples. By focusing on the individual, this project differs from approaches in, for example, cognitive psychology, which, due to statistical, data-analytical reasons, disregards deviant behavior of individual experimental participants and thus focuses only on cognitive models for aggregated experimental data and identification of general cognitive phenomena. In this project, we reanalyze data with a focus on an individual perspective, develop computational systems capable of predicting an individual's future conclusions given sparse background information, and analyze and evaluate the performance of existing approaches in cognitive psychology, cognitive science, and data-driven methods such as recommender systems. In particular, we will identify and then develop and systematically improve methods that enable accurate predictions but also identify data-driven upper bounds. Furthermore, the question of which mechanisms underlie human reasoning and how these insights can be reflected in an automatic development of adaptive cognitive modeling approaches will be clarified. This project develops cognitive AI models that are able to adaptively and predictively adjust to individual thought processes and predict conclusions.