5D Library › The Algorithmic Mirror · 中文
Chapter 7 Mirrors in the Classics
The mirror that does not hold
Source text
至人之用心若镜,不将不迎,应而不藏,故能胜物而不伤。(庄周, n.d.)
The perfected person uses the mind like a mirror: it does not escort things away or go out to welcome them; it responds but does not store. So it can meet all things without being harmed.
(Zhuangzi, “Responding to Emperors and Kings”)
Zhuangzi’s mirror is an ideal for the mind of the person, not a description of a machine. Its point is the absence of grasping: the mirror reflects whatever stands before it, fully and without distortion, and keeps nothing once it has gone. A language model resembles this mirror in one respect, that within a conversation it responds to what it is given, and differs from it in others: it is shaped by everything it was trained on, so it does “store”, and it can distort. If the image applies anywhere, it applies best to the user. Chapter 4 showed what a mind that “escorts and welcomes”, that goes out to meet the evidence with its conclusion already in hand, does to ambiguous text.
The bell, again
The Record of Learning’s bell (Chapter 2) belongs with the mirror. Both describe a responder whose answer is shaped by what comes to it, and both put the burden on the one who strikes or stands before it. The Record adds a point the mirror lacks: the bell needs time, “given time, it sounds to its full extent”. A good question is often a sequence of questions.
Indra’s net
The Huayan school of Chinese Buddhism, systematised in the seventh and eighth centuries by masters such as Fazang, used the image of the net of the god Indra: a net with a jewel at every knot, each jewel reflecting all the others, and the reflections reflecting reflections without end, 重重无尽, “layer upon layer, without exhaustion”. The image expresses the doctrine of the mutual containment and interpenetration of all phenomena (Cook 1977).
A high-dimensional embedding space invites the comparison. In a trained model, the representation of each word or concept is defined by its relations to all the others, and attention lets each position draw on every other. The comparison is suggestive but loose: the relations in a model are learned statistical associations of finite precision, not the metaphysical interpenetration Huayan describes.
Consciousness only
The Yogācāra school, transmitted to China above all through Xuanzang’s Cheng weishi lun (659), taught that what we take to be an external world is the manifestation of consciousness. Its analysis distinguishes eight consciousnesses; the eighth, the storehouse consciousness (ālayavijñāna), holds “seeds” (種子) laid down by past experience and action, which ripen into present experience (現行), while present activity in turn “perfumes” (熏習) new seeds (玄奘, n.d.; Lusthaus 2002). Later Chinese Buddhism summarised the school’s teaching in the phrase 万法唯识, “all phenomena are consciousness only”. The Zongjing lu (Record of the Source Mirror), compiled by Yongming Yanshou in 961, took the one mind as its source and its title from the image of a mirror.
The parallel with Chapter 4 is direct. What a reader finds in an ambiguous text depends on the seeds they bring: their expectations, attitudes and earlier experience. The parallel with a language model is weaker: a model’s “seeds” are its learned parameters, and its “manifestation” is conditioned by the context; but nothing in a model resembles the moral and soteriological purpose of Yogācāra analysis.
Note
These texts belong to their own religious and philosophical traditions, with their own purposes. Reading them beside information theory is an analogy that can illuminate both, not a claim that their authors anticipated machine learning or that machine learning confirms their teachings.
Hypothesis · not established
Not established. People trained in a practice of non-grasping attention of the kind Zhuangzi’s mirror describes (for example, several years of regular meditation practice) show less biased assimilation when rating ambiguous AI-generated text than matched people without such training. The hypothesis predicts a smaller effect of prior attitude on ratings in the trained group in the experiment of Chapter 4. It would be refuted if, with groups matched on age, education and attitude, the effect were as large in the trained group.
References
Fan, Y. (2026). The Algorithmic Mirror: Self-Attention Geometry, Ashby's Requisite Variety and Cognitive Projection in Information Theory and East Asian Philosophy of Mind [算法之镜] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23723557 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).