二次学習に基づく心的表象の進化的理論とその検討

在这项研究中，我们研究了一个关于认知起源的基本问题：表征认知是如何演变的？该项目包括理论工作(中心假设的表述和完善)和计算验证(w.r.t.空间认知和社会认知)。本研究项目的核心是一个假设，即表象认知是在二阶学习能力(即学会学习的能力)的选择下发展起来的。应用于社会认知这一主题，这意味着符合二阶学习条件的进化社会能力将产生各种形式的社会认知，这些形式通过形成对其他人心灵的表征来运作(“心理理论”)。我们通过让一种简单的社会认知在人工智能系统(神经网络)种群中进化来研究这一思想。计算结果：1)在模拟中，二阶学习能力选择下的进化导致了社会认知的一种表征形式，而一阶学习能力选择下的进化不会。这一结果支持了表象认知是二阶学习能力进化选择的产物的假说。2)在此过程中，我们展示了表征认知如何在神经网络中进化。这项技术可以被证明适用于例如社会机器人。理论工作的结果：完成了一篇详细介绍该理论的理论论文(与我们的计算工作一起说明)，并发表在Minds and Machines杂志上。

In this research we investigated a fundamental question about the origins of cognition: How did representational cognition evolve? The project consisted of theoretical work (formulation and refinement of the central hypothesis) and computational verification (w.r.t. spatial cognition and social cognition). At the heart of this research project is a hypothesis that representational cognition evolves under selection for 2nd order learning ability (i.e. the ability to "learn to learn"). Applied to the topic of social cognition, this implies that evolution social abilities that qualify as 2nd order learning will produce forms of social cognition that operate by forming representations of other individuals’ minds ("Theory of Mind"). We investigated this idea by letting a simple form of social cognition evolve in a population of AI systems (neural networks).Results of computational work: 1) We showed that in simulation, evolution under selection for 2nd order learning ability leads to a representational form of social cognition, whereas evolution under selection for 1st order learning ability does not. This result supports the hypothesis that representational cognition is a product of evolutionary selection for second order learning ability. 2) In doing so, we showed how representational cognition can be evolved in neural networks. This technique could prove applicable in e.g. social robotics.Results of theoretical work: A theoretical paper detailing the theory (illustrated with our computational work) was completed and published in Minds and Machines journal.