Contemporary psychology—driven in part by modern neuroscience and following the path set out by the other natural sciences over the last centuries—is rapidly embarking upon a collective program of formalization. It is a natural scientific instinct to borrow formal intellectual frameworks, and tools, from disciplines in which they have already become established, foundational, dogma. Physics, in particular, presents the ambitious scientist with an extremely mature, and demonstrably effective, box of tools for formalizing natural processes. There is, however, a potential hidden danger in the eager application of models to domains for which they were not originally derived. Physics, it seems, is relatively “simple”: The objects of study are well defined; dimensionality is usually obvious (and low!); and it usually possible to design rigorous experiments to test, piecewise, the components of a formal mathematical model. Psychology, as gazed upon by an outsider, appears to posses none of these features. The “atomistic” objects of study remain a matter of open, and fierce, scientific debate; dimensionality is often unclear (other than that it is apparently high), and the massively heterogenous nature of neural systems makes the proposal of reductionist experiments enormously more challenging. Psychology is not simple, it is hard. Are we not then, to paraphrase von Neumann, “living in a state of sin” in attempting to apply simple models to hard problems? One answer to this query requires a brief detour into the theory of computability wherein one studies the types of functions that a given formal system can compute. The tablet, phone, or laptop upon which you are probably reading this commentary is a (very sophisticated) example of a system capable of universal computation; any function that can be computed (by any means in this universe) can be computed on the system you hold in your hands. This is probably unsurprising to you, given the dramatic role that electronic computers play in modern science. It is not at all obvious, however, that a symbol rewriting system with just two rewriting rules can compute any computable function, and yet this is precisely the combinatory logic that Sch€ofinkel demonstrated to be computationally universal in the 1920s! Contemporary “one instruction set computers” present a zoo of examples in which a single instruction is sufficient to implement universal computation (N€ urnberg, Wiil, and Hicks, 2003). Closer to the topic at hand, the famous “Rule 110” (Cook, 2004) shows the computational universality of very simple cellular automata—a class of systems that generalize dynamics of the sort observed in the Ising model. And so we arrive at the crux of our query: Dalege, Borsboom, van Harreveld, and van der Maas (this issue) describe a mapping between a generalized Ising model and a particular model of attitude dynamics that they have defined. The existence of this mapping is then, it seems, taken as an implicitly parsimonious finding supporting the proposed model of attitude dynamics. This is seductive reasoning: The Ising model is an apparently simple mathematical construct (and, of course, parsimony loves simplicity) that has been used very productively in formalizing the dynamics of many physical systems. The challenge with this leap of inference is that the Ising model, like Conway’s Game of Life, turns out to be extremely computationally powerful (Lucas, 2014). Indeed, for all their apparently beautiful simplicity, generalized Ising models have been shown to be capable of universal computation (Barahona, 1982; Lloyd, 1993). That such a model can capture an arbitrary model of attitudinal dynamics is, thus, unsurprising; the existence of this mapping tells us only that the model under study is computable. An argument for the correctness of the attitudinal entropy framework based on the ability to implement it as an Ising model is not compelling. This serves as a useful exemplar of the importance of studying and understanding, deeply, the tools one is using. It is insufficient to understand a model at the superficial level of simply being able to calculate with it; one must internalize both the details of the model and the broader intellectual context within which it exists. This requires serious commitment, caution, and restraint on the part of the modeler. What may be compelling, however, is the model itself— independent of the (Ising-based) implementation details. The authors articulate an entropic framework for reasoning about attitude that links to a rich supporting literature. To make a compelling link to an Ising-like implementation,

中文翻译：

---

Circe的受害者：我们是否太容易被数学物理学的海妖之歌所诱惑？

当代心理学——部分由现代神经科学驱动并遵循过去几个世纪其他自然科学所设定的道路——正在迅速开始一个形式化的集体计划。从已经确立的、基础的、教条的学科中借用正式的知识框架和工具是一种自然的科学本能。尤其是物理学，为雄心勃勃的科学家提供了一套极其成熟且明显有效的工具，用于将自然过程形式化。然而，将模型急切地应用于它们最初并非派生的领域中存在潜在的隐患。物理学似乎相对“简单”：研究对象明确；维度通常很明显（而且很低！）；并且通常可以设计严格的实验来分段测试正式数学模型的组成部分。在外人看来，心理学似乎不具备这些特征。“原子”的研究对象仍然是公开的、激烈的、科学辩论的问题。维数通常是不清楚的（除了它显然很高），并且神经系统的大量异质性使得还原论实验的提议更具挑战性。心理学不简单，很难。那么，用冯·诺依曼的话来说，难道我们不是在试图将简单的模型应用于困难问题时“生活在罪恶的状态中”吗？对此查询的一个答案需要简要介绍可计算性理论，其中研究给定形式系统可以计算的函数类型。平板电脑、手机、您可能正在阅读此评论的笔记本电脑或笔记本电脑是能够进行通用计算的系统的（非常复杂的）示例；任何可以计算的函数（在这个宇宙中通过任何方式）都可以在你手中的系统上计算。考虑到电子计算机在现代科学中扮演的重要角色，这对您来说可能不足为奇。然而，只有两个重写规则的符号重写系统可以计算任何可计算的函数，这一点并不明显，但这正是 Schöfinkel 在 1920 年代证明的计算通用的组合逻辑！当代的“单指令集计算机”展示了大量示例，其中一条指令足以实现通用计算（N€ urnberg、Wiil 和 Hicks，2003 年）。更接近手头的话题，著名的“规则 110”（Cook，2004 年）展示了非常简单的元胞自动机的计算通用性——一类可以概括 Ising 模型中观察到的那种动力学的系统。因此，我们到达了问题的关键：Dalege、Borsboom、van Harreveld 和 van der Maas（本期）描述了广义 Ising 模型与他们定义的特定姿态动力学模型之间的映射。然后，这种映射的存在似乎被视为支持所提出的姿态动力学模型的隐含简约的发现。这是一个引人入胜的推理：伊辛模型是一个明显简单的数学结构（当然，简约喜欢简单），它在许多物理系统的动力学形式化过程中被非常有效地使用。这种推理飞跃的挑战在于 Ising 模型，就像康威的生命游戏一样，它的计算能力非常强大（卢卡斯，2014 年）。事实上，尽管广义 Ising 模型表面上看起来很简单，但已被证明能够进行通用计算（Barahona，1982；Lloyd，1993）。因此，这样的模型可以捕获任意的态度动力学模型并不令人惊讶。这种映射的存在仅告诉我们所研究的模型是可计算的。基于将态度熵框架作为 Ising 模型实施的能力来论证态度熵框架的正确性并没有说服力。这是学习和深入理解一个人正在使用的工具的重要性的一个有用的例子。仅仅能够用它进行计算，仅仅在表面上理解模型是不够的；人们必须内化模型的细节和它存在的更广泛的知识背景。这需要建模者的认真承诺、谨慎和克制。然而，可能引人注目的是模型本身——独立于（基于 Ising 的）实现细节。作者阐明了一个熵框架，用于推理与丰富的支持文献相关的态度。为了与类似 Ising 的实现建立引人注目的链接，