1. The False Opposition of “AI vs. Dictionaries”
Japanese online media regularly circulates a certain genre of article: “I’m sick of my subordinate who outsources everything to AI.” One recent example centers on a 58-year-old woman who gave her 8-year-old grandson a Japanese dictionary, only to be told by her son-in-law, “Grandma, dictionaries are so outdated. They’re just in the way. Bad cost-performance, bad time-performance. If you have AI, who needs a dictionary?”注1
A crisis-management consultant is duly quoted, warning of the risk that “dependence on AI hollows out our capacity to think.” Articles of this type have proliferated in recent years, and most of them rest on an implicit binary: “people who rely on AI” versus “people who research and think for themselves,” or “AI” versus “books and dictionaries.”
But this opposition is technically wrong at its foundation. What the son-in-law’s remark actually reveals is not so much “cognitive atrophy from AI dependence” as something simpler: a failure to understand what the tool he is using is actually built on.
2. AI Doesn’t Replace the Dictionary — It Depends on It
The accuracy of generative AI output is directly tied to the quality of its training and reference data. This is treated as nearly axiomatic within the industry. Standard countermeasures against hallucination (the confident generation of false information) include “using datasets grounded in reliable primary sources” and “filtering out noise and misinformation”注2. Even when retrieval-augmented generation (RAG) is deployed, the standard caveat is that “if the referenced data itself contains errors, this affects the output — so using trustworthy data sources is essential”注3.
In other words, the kind of structured knowledge found in a dictionary — headwords, definitions, and usage examples curated and verified by specialists over decades — is not something AI has rendered obsolete. It is, rather, part of the infrastructure that underwrites the quality of AI’s output. If dictionary-derived structured knowledge were stripped from the training corpus, AI’s accuracy in language comprehension and generation would demonstrably decline. The son-in-law’s comment is the equivalent of saying “we don’t need power lines because we have Wi-Fi” — the causal relationship has been inverted.
Seen this way, the real problem the article should be interrogating isn’t a vague anxiety about “AI dependence,” but a deeper issue: a basic failure to understand what AI is actually drawing on when it generates an answer. This is not unique to AI. It follows the same pattern as Plato’s critique of writing in the Phaedrus, where he worried that the written word would weaken memory注4. It is one more instance of a rhetorical pattern that recurs every time a new medium appears — the story that a new tool represents the “degradation” of some authentic human capacity it merely replaces.
3. Someone Wove the Dictionary — The Labor Behind The Great Passage
What tends to be overlooked here is that a dictionary, too, is a work someone made — the product of enormous time and devotion. Shion Miura’s novel The Great Passage (Fune wo Amu, winner of the 2012 Japan Booksellers’ Award) is precisely a portrait of this invisible labor.
In the novel, the editor Araki says: “A dictionary is a boat that crosses the sea of words.” “People board this boat called a dictionary to gather the small points of light rising on the dark surface of the sea — so that they can deliver their thoughts to someone else, accurately, with the most fitting words.”注5
At the heart of the actual craft of dictionary-making is “example-gathering” (yourei saishuu) — the unglamorous, unending work of noting down real, living usages of words wherever they are found注6. This work is fundamentally different from mechanical search. It requires a human being to observe the ongoing change of a living language and make judgment calls, one word at a time, about how to define it and which examples to include — an intellectual labor saturated with value judgments.
The reason AI can generate natural-sounding Japanese at all is that it stands on decades of accumulated linguistic resources built by these anonymous compilers. The son-in-law’s claim that “we don’t need dictionaries anymore” also betrays a failure of imagination about this invisible labor.
4. The Personality of a Dictionary — Something AI’s Statistical Averaging Cannot Reproduce
There is a further, more important point: a good dictionary has personality.
Shinmeikai Kokugo Jiten has long been famous — even notorious — for the idiosyncrasy of its definitions. Tadao Yamada, who oversaw its editing, revised the dictionary in what was essentially a solo editorial vision注7, producing definitions that were by turns caustic and darkly funny, unlike anything found in other dictionaries. This dictionary’s personality was distinctive enough to inspire an entire book, Genpei Akasegawa’s The Mystery of Mr. Shinkai, and some readers have even lamented that with each new edition, “the bite (doku) of the interpretations is gradually removed”注8.
What does this tell us? A truly excellent dictionary is not merely an aggregation of word meanings — it is an expression filtered through the compiler’s own view of language and the world. Generative AI output, by contrast, tends by its very nature toward the statistical average, the most probable pattern across a vast corpus. A single editor’s idiosyncratic “bite” or “quirks” — precisely the sort of extreme, individual point of view — is exactly the kind of thing that gets smoothed away in statistical processing.
Which is to say: what will actually be scarce and valuable in the age of AI is not “the safe, accurate summary,” but the individually-voiced primary source, distilled from a specific person’s discernment. The work of a lexicographer, far from being devalued by the arrival of AI, is arguably entering a phase where it deserves to be revalued.
5. Turning the Pages of a Paper Dictionary — Active Intellect Within Inefficiency
The son-in-law dismisses paper dictionaries as having “bad cost-performance, bad time-performance.” But this, too, misses something. The act of looking up a word in a paper dictionary structurally builds in encounters with neighboring words you weren’t looking for — serendipity注9. As one English-teaching tutor observes, students who look up words in a paper dictionary naturally encounter related vocabulary near their target word, producing a learning effect that a single, direct digital search does not注10.
This is a value that no efficiency metric can capture. The very framework of cost-performance and time-performance is incapable of measuring the value of a chance encounter, or the simple pleasure of holding and paging through a dictionary as a physical object. If anything, deliberately choosing an “inefficient” tool, against a culture that prizes efficiency above all, is itself an expression of active intellectual engagement.
6. Analog as Fail-Safe
There is also a practical dimension that has not disappeared. During the Great East Japan Earthquake, it is well documented that while mobile phone networks became congested and unreliable, analog landlines remained stable注11. Digital infrastructure carries an inherent risk of failing all at once during power outages, communication breakdowns, or disasters; dependence on a single infrastructure is itself a risk. A paper dictionary, by the simple fact of requiring no power source, functions as a rational diversification — a fail-safe — within an information environment that has otherwise gone almost entirely digital.
7. Conclusion — The Age of AI Needs Excellent Lexicographers More Than Ever
To summarize, this genre of “AI vs. dictionary” discourse layers at least four distinct misunderstandings on top of one another:
- A technical misunderstanding: AI functions by depending on dictionary-like structured knowledge; the relationship is one of dependence and inclusion, not opposition.
- The erasure of labor: A dictionary is the crystallization of the patient, unglamorous intellectual labor of anonymous specialists, epitomized by “example-gathering.”
- The overlooked matter of personality: A great dictionary is an expression filtered through its compiler’s point of view — something AI’s statistical averaging cannot substitute for.
- The flattening of value into a single axis: Cost-performance and time-performance, as metrics, simply cannot register values like serendipity or fail-safe redundancy.
Paradoxically, precisely because AI depends on dictionary-like knowledge structures to function, the quality of AI’s output ultimately rests on the quality of the primary sources that skilled lexicographers produce. The real point this article should have made was not “we don’t need dictionaries because we have AI,” but rather the reverse: “precisely because we have AI, we need excellent dictionaries more than ever.”
Before we reach for the label “AI idiot” to mock an individual, what we actually need to confront is the basic absence of information literacy education around using tools without understanding what those tools are built on. This is not a problem unique to AI — it is an old problem in new form, one that has recurred since Plato’s time in the relationship between human beings and their tools. And so as not to be, rather than an “AI idiot,” simply an idiot — full stop — we first need to ask ourselves what the tools we rely on every day are actually standing on.
Notes
- Source article: “‘You didn’t feed AI our client data, did you?’ — Fed up with a subordinate who outsources everything to AI. The worst-case ending of information leaks × cognitive shutdown [Expert commentary]” (crisis-management consultant Toshiki Hiratsuka)
- Commentary on countermeasures against generative-AI hallucination (weel.co.jp, et al.)
- Commentary on RAG-based hallucination countermeasures (licensecounter.jp)
- The myth of Thamus and Theuth in Plato’s Phaedrus
- Shion Miura, The Great Passage (Fune wo Amu), Kobunsha, 2011 / paperback 2015
- The novel’s depiction of “example-gathering” (yourei saishuu)
- History of Shinmeikai Kokugo Jiten (Sanseido, chief editor Tadao Yamada)
- Genpei Akasegawa’s Shinkai-san no Nazo and related commentary on the dictionary’s idiosyncratic definitions
- The etymology of “serendipity” (Horace Walpole, 1754)
- English-learning column on paper dictionaries and serendipity (manalink.jp)
- Reporting and commentary on the stability of analog landlines during the Great East Japan Earthquake
Positioned as an extension of [[media-dependency-essay]]. It can be treated as the latest instance of the same recurring pattern traced there — from Plato’s critique of writing to SNS regulation to discourse on AI dependence — the story, retold with each new medium, that a tool represents the “replacement, and therefore degradation,” of some authentic human capacity.