Henry Smith

General Comments

When Professor Tomohiro Inoue wrote our source text last summer, I doubt he was thinking about translators, but this text could have been tailor made as a test for us. We must seamlessly incorporate a flirting couple, employment statistics, Vincent van Gogh, and the state of Japanese education into an engaging and coherent message on the economic prospects offered by artificial intelligence. All five finalists have amply demonstrated why the world needs human translators— rather the machines—to craft such coherent messages.

All our finalists produced predominantly complete and accurate translations with some ingenious choices to replicate the literary quality of the original; however, in my initial comment, I want to focus on another thing that algorithms cannot do, but translators can (and should): creating an authentic voice for our author. A little background research tells us that Professor Inoue strongly advocates a universal basic income (UBI). Indeed, he has stated (in English) that the favourable results of a recent Japanese UBI experiment are ‘deeply moving’ because they strongly suggest a ‘connection between cash benefits and a happier and fulfilled life’. Clearly, our author is a macroeconomic expert who focuses on people’s welfare, and this fact should help translators to find the right voice for this message on an AI-driven economy of the future.


Four of the five finalists used ‘idea’ and ‘product’ in their titles, but retroconverting loan words may not achieve the desired effect. Professor Inoue is on record that 「アイデア即プロダクト」 is a deliberately キャッチコピー的 term intended to spark debate on the economy of 2030. アイデアandプロダクト may add an exotic touch in Japanese, but ‘idea’ and product’ do not resonate like a catchphrase back in English, in my opinion. Buzzwords that might describe our concept abound on corporate websites in the Anglosphere. My own brief search suggests ‘Instant Imagineering Economy’ as just one possible candidate.

 今話題のChatGPTという言語生成AIは、あらゆる質問に答えられるだけではなく、論文や小説、プログラムなどを生成できる。一方、Stable Diffusion のような「画像生成AI」は、例えば、「浜辺で戯れるカップル」などと言葉を入力するだけで、相応する絵や写真の画像を生成可能だ。

 こうした「生成AIによって …

Our contestants have painstakingly transferred all the information from the first two sentences into their English versions, but a little more attention to coherence could have paid off when presenting all this information. Our text structure has a few clues that may help. These sentences are a 序論 (preamble) to a 本論 (main argument) starting at こうした「生成AI」によって, and the whole text itself is a 尾括型 (reasons>conclusion) essay. Thus, this two-sentence preamble would not conventionally be the place for key, argument-supporting facts or a foreshadowed conclusion; its main job is just to prepare readers for a discussion on generative AI. The 鉤括弧 around 生成AI do a lot of that work by using their grammatically legitimate 強調する function to establish a coherent thread on the topic, and clearly that function has nothing to do with quotation. So, our finalists were correct to reject 鉤括弧 ‘s English cousin, the quotation mark but, but I believe they needed to do more to compensate for loss of function. With some subtle changes (maybe in word order and subject-verb choice), a translated preamble here could use the given information to highlight a topic (as I believe the author intended), rather than appearing to tell readers things (especially about ChatGPT) they probably already know.

For the term 生成AI itself, our finalists proposed a mix of ‘generating,’ ‘generation,’ and ‘generator,’ all of which may be acceptable (although it is better if the chosen term is used consistently). However, based on ChatGPT’s response when prompted ‘How do you define yourself?’, I suspect that ‘generative AI model’ may be the optimal term, and it would also resolve any singular vs. plural dilemmas.

As we reach the 本論, all our finalists deserve credit for putting people on the page. The finalists’ choices of ‘people,’ ‘users,’ ‘we,’ and ‘you’ correspond to the unstated but very much present subjects and objects in Japanese sentences, which machine translators often kill off. That would be particularly disastrous here, since this column is essentially about people, not machines. Interestingly, when our finalists identify unstated human actors, as here, they use terms like ‘people’ or ‘we’, but when actually translating 人間later in the text, they almost invariably propose ‘humans,’ which (I believe) is a potential source of inconsistency in the author’s voice. We should remember he is an economist not an anthropologist; in a quick search of his published work and pronouncements in English, I found a lot of ‘people’ but not a single instance of the noun ‘human’.

 AIが可能にするのは、アイデアをすぐ形にできる「アイデア即プロダクト」の経済だ。。そこでは …誰でもクリエイターになり得る。

 その反面、… アーティスト(の)職業 … 成り立ちにくくなる。

Our finalists rendered 可能にする as ‘enables,’ ‘will enable,’ or ‘opens the door,’ the last of which best shows the imaginative use of English expected in a journal article of this nature, but I want to focus on tense here. Our author chose to write AIが可能にする (not 可能にした), and the economy of 2030 has been a dominant theme in his recent work, so I argue that tense should be chosen on the basis that AI is now opening up a future economic prospect. Choice of tense is bound to be complex when dealing with an evolving phenomenon such as the one here, but our choices should at least allow the そこではand その反面 sentences to co-exist in the same space in time, even though the そこでは situation may arise first. Choosing the future tense for those two sentences may also act as a subtle indicator that our article is going to finish on a future-orientated note, where the text takes a somewhat unexpected turn to the education of the next generation in its final sentence.


Our finalists made some gallant efforts to make this statistic more digestible, but I wonder if just to putting ‘one in five jobseekers’ near the front of the sentence would have been easier. Either way, I think our readers should be told these are Japanese jobseekers, not some globally surveyed group of aspiring artists.


When we come to these three disabilities of AI, there are some clear choices for 「意志」 and 「体験」, and on the face of it, translating 価値判断 as ‘value judgement’ might seem to be an open-and-shut case. But can we treat ‘value judgement’ as an uncountable characteristic in this way? Japanese may have more flexibility, allowing 価値判断 to be AIに欠けている here, and then the object of 行う a couple of sentences further down the page, but I find it difficult to imagine the same word being the grammatical object of (not) ‘possessed’ and ‘performed’ in English. Our finalists chose ‘value judgement’ or just ‘judgement’, and I certainly would not insist those choices were wrong, but I would be tempted to look at other alternatives to indicate that the meaning here relates to AI’s (uncountable) inability to make subjective judgments.

Interestingly, in the passage that follows, all our finalists have stuck fairly closely to the word order in the original Japanese sentences, where each disability is stated again towards the end of its own one- or two-sentence mini-section. If exactly the same ideas were being written up by an Anglophone economist, I suspect each disability would be the opening term of its own mini-section, and this would certainly make a more coherent text for the English reader. One of the (many) challenges that make our work interesting is to consider how tweaked word orders can make our complete and accurate translations look more natural.


In this lively passage, スーパー偏差値エリートand 指示待ち人間 certainly catch our eye, and they represent great case studies on how to translate single units of meaning. Recreating a unit of meaning in English by translating individual words within it tends to produce stilted and often nonsensical prose. Rather, we have to find some matching English idiom (often impossible) or craft our own version. Our versions should succinctly and immediately get the essential point across (something like exam-passing prowess vs. zero dynamism here) and, given that the 言わば promised the original readers some playful language, they should not be dull. Atmospherics can be important too; for example, 指示待ち人間 is most often used in workplace situations to identify the poor guy who will not be getting his initiative bonus again this year, so a workplace-related expression would work well in our translation.


To appreciate the effect of ことができない, consider how the meaning would change if 生み出せないwere in the terminal position. By using ことができない, (in tandem with 人間のごとく), the author shows that AI fails, but people shine, at every stage of the creative process. We should consider what reinforcement a simple ‘cannot’ may need to achieve the same effect in English. The opening したがって actually points back to AI’s failings enumerated across three 段落, but an English reader used to looking to paragraph structure for contextual clues might mistakenly assume this point is just a link in a cause-effect statement about judgement, unless したがって is handled correctly. As the scholar Jay Rubin warns us, ‘Never trust Japanese paragraphing to work as it does in English.’

The translator’s job changes a little from this point. Through the 「意志」「体験」「価値判断」passage, we acted as guides grandly gesturing at the view; in the next passage, we have to guide our readers carefully along a twisting path. We must go step by step from AI’s total failure to be original now, to our exciting chance to be original in the future. To translate a sequence like this, we must avoid treating sentences in isolation, and keep an open mind of how the step-by-step instructions may differ between context-rich Japanese and content-rich English.


The first step on our path involves 難しい, which surely has a wider spectrum of meaning than ‘difficult’ does in English. Given the context, having just established AI’s failings with originality, I argue that our translation should be as close to ‘impossible’ as we can get on this spectrum. Just ‘challenging’ may not be enough.


OO風 may look like the most exciting term to translate here (although I think it is just plain old 何々風) but そもそもの provides the crucial step in this sequence about originality. On a minor point, we need to take care with 描いて. Our sequence builds upon allegories about van Gogh and oil paints, so ‘paint’ would better than ‘draw’. However, can an AI model really ‘paint’? Should we start putting quote marks inside quote marks here? The best solution could be to find a completely different term.


Our path now changes direction (from AI’s failings to our prospects), and the turn is nicely executed in the source text with a clever interplay of words. To see what I mean, let us imagine how this sentence may play out as a subliminal dialogue between the author and his readers. The author could be saying, ‘I agree no one can be as original as van Gogh anymore—in the ancient medium of painting’, and readers might reply, ‘Wait a minute! Do you mean there is a chance to be original in some other medium?’, and then they read on to find out. The ideal translated sentence here should have a similar effect. Many terms contribute to achieve this dramatic effect, but I argue that もはや is especially important to the logic of the sequence, given what comes next about exhausting and exploring artistic possibilities.


I notice that our finalists all stuck fairly closely to the word order of the Japanese sentence here. However, our readers are still pondering the question of media raised at the end of the previous sentence, so should translators consider a bold re-ordering with ‘media’ coming before ‘technology’?



Here, we reach the end of our path-guiding duties. This is where our readers turn around and take a sharp breath as a new horizon of artistic possibility is revealed to them. My interpretation is partly based on some background research: these words encapsulate the author’s initial economic theory that a UBI is affordable when we have a high rate of technological change, and that this rate of change is achievable when we all have the leisure to follow our chosen creative pursuits. Furthermore, it makes sense that this shift from a negative prospect to a positive view of our future (thanks to AI) is a crucial point in the essay. I argue we want this pair of sentences to have a combined, dramatic effect, and achieving that effect in English means giving the translation a bit of thought. For example, we could specify ‘people’ as the actors behind 模索and 取尽くす, so the word appears before and after the 逆に言うとtranslation, assuring a mirror-image effect. 生みappears in the both the source-text mirror images, but our finalists have proposed relatively dissimilar parings (produces/will be created, leads to arrivals/are generated, gives birth/will take form, leads to new forms/will be born, and create/continue to emerge). A paring like ‘spring from/spring up’ might better indicate a single phenomenon being viewed from different angles. My suggestions may or may not work, but this pair of sentences is a fascinating case study in translating sentences as logical links in a flow, rather than standalone messages in a procession.



Of course, I cannot see into the author’s mind, but I wonder if the AI時代 sentence here was the true 結論, and the sentence on education was an addition (maybe the foreshadowing of the next question to discuss). Whether I am right or wrong, I think the translator needs to consider the transition to the final sentence. I argue that a purely faithful translation of each sentence in isolation risks making the final transition even more abrupt than it should be. By adding a ‘that-is-why’ wording in front of the education sentence, we could make it more like a conclusion. Alternatively, we could try some metatext, such as ‘That brings me finally to education, which in Japan ...

At the final hurdle, some finalists let だろう distract them into a slightly muffed shot. This term just shows that the statement is the author’s (obviously strong) opinion, and is not really an indication of probability. So, it is best ignored, or perhaps turned rhetorically into something like ‘Wouldn’t you agree?’ as the parting shot.


E05 produced a very accurate translation, but at times the influence of the source text prevented the expression from being as natural as it could have been. Any translation needs to have both transparency and opacity; transparency, so that authors and checkers (and contest judges) can see the link back to the source text, and opacity, so that readers never guess the target text is a translation. E05 did well on the transparency side, but that came at the cost of opacity.

The economy of ‘idea to product’ made possible by AI

On the title, I think E05 made the best choice for 可能にする with ‘made possible.’ A title is supposed to encapsulate the whole text, and ‘made possible’ foreshadows the discussion on what is needed for us to seize the title’s economic opportunity, up to the final call for education reform. However, I worry that E05 has lost 即 in translation. Turning ideas into products has presumably been happening for millennia, and here the author is looking at how that transformation can now be achieved in a mouse click, thanks to AI.

The now much-discussed language-generation AI ChatGPT can not only answer any question, but can generate articles, novels, and programs, while an image-generation AI like Stable Diffusion is able to produce an appropriate picture or photographic image with only a text input such as, for example, ‘a couple playing on the beach’.

E05’s opening sentence did achieve the right balance between AI and ChatGPT and make a logical thread to the 本論 (partly thanks to a good choice of ‘while’ for 一方). However, I fear some readers may struggle with this paragraph-spanning blockbuster. Long sentences can work in English when each part clearly relates to the central idea, and this often requires using repeated grammatical patterns to underline some point of comparison or similarity or a progression (that is to say, parallel structures). None of that really is evident here, which is why readers might spot it as a translation.

On smaller points within the sentence itself, I think ‘any type of question’ would be better than the literal ‘any question’ for あらゆる質問, and we may need to draw the dots between ‘appropriate’ and ‘text input’ more explicitly. The relationship covered by 相応する here has been described by Stable Diffusion’s own developers as the production of ‘text-guided images,’ so I wonder if something like, ‘corresponding to a simple text input’ would make the second half of the paragraph clearer.

With this kind of generative AI we can create picture books or manga and also sell them easily as electronic documents. Soon, by connecting with devices such as 3D printers, individuals will be able to make things like accessories or furniture instantly. Soon, by connecting with devices such as 3D printers, individuals will be able to make things like accessories or furniture instantly.

I think E05’s choice of ‘we’ as the actor at the main argument’s opening was spot on. In my (subjective) view, this choice of subject best suits an authentic translated voice for our author. Very pedantically, I think a comma just before ‘we’ would have helped readers better appreciate the excellent choice of subject. The singular ‘generative AI’ at the start of the sentence may leave readers unsure whether the discussion relates to Stable Diffusion or ChatGPT here.

E05’s next sentence has a bit of problem with ‘connecting,’ which makes it seem as if individuals are hooking themselves up to devices Matrix-style; using ‘combining’ and another reference to AI models might have avoided that problem.

On the other hand, those of average skill will be unable to compete with AI, making it difficult to pursue art as a profession. Even now, the job-to-applicant ratio for ‘artists, designers, etc.’ is around 0.18, with only 1 in 5 people able to find work.

After characterizing our AI-driven economy, E05’s sentence on ‘making it difficult to pursue art as a profession’ was an excellent translation, and a stylistically effective way to introduce art into a discussion on macroeconomics. However, the second sentence here could be more reader-friendly. The ‘1 in 5 people’ at the end of the sentence seems as if it could refer to the whole population, rather than people applying for jobs in the field of art and design (and many journals might expect to see ‘one in five’ spelled out). Quote marks could be used here, because the author was citing an officially defined category of the Japanese Labour Force Statistics (full title: 美術家,デザイナー,写真家,映像撮影者; for once, we know exactly what 等 means!), but readers then probably need some acknowledgment of this, lest they misunderstand the quotation marks to indicate irony. The alternative might be a broad description like ‘the field of art and design (in Japan)’. On a smaller point, ‘around’ might logically go with the one-in-five (for the original ごらい), whereas the ratio of 0.18 appears to have been rounded.

It is not only artists that will be unable to escape the effects of generative AI, but almost all white collar jobs: professions such as administrative staff, accountants, tax consultants, and teachers and researchers like myself

E05 has covered all the main points here, but the sentence has become a bit ungainly. I feel like a ‘It is not only …’ structure may need another verb in its second half. Also (very pedantically), I think the colon is a bit problematic. With ‘almost all’ before the colon, it is hard to satisfy the convention that what comes after a colon should explain what comes before it. (Adding ‘including’ somewhere before the colon might help.)

A generative AI like ChatGPT is, as it were, an ‘academic super-elite’, but also a ‘human awaiting instruction’.

…. ‘.. akin to a ‘super indoor type’’

I agree with E05 on ignoring ‘deviation’ for the translation of スーパー偏差値エリート, and ‘academic’ could be part of a good solution here, but after that, I start to disagree with some of the choices made. ‘Academic super elite’ could become a red herring for readers; would they imagine a top professor from a leading university (who presumably can form intentions, even if they are not good ones)? The original Japanese term is used quite a lot in discussion of socioeconomic status, but it makes more sense here if we are going back to its original definition, of candidates whose entrance marks are in the top percentile separated from the mean by several standard deviations. After all, ChatGPT is good at answering questions, but does not get any socioeconomic status for doing so. Describing 「指示待ち人間」 as a ‘human awaiting instruction’ may literally work, but it is very stilted for a passage that is so lively in the original. I feel that ‘ultra’ or ‘the ultimate’ might work better in front of ‘indoor type’ than ‘super’.

… it cannot create a story or a film based on its own experiences.

I think E05 made an interesting choice of ‘story’ for 小説 here. It could be a prudent choice, considering all the controversy over Al and novels, and also gets closer to the heart of what AI cannot do (failing to imagine any fiction at all, not just its own version of War and Peace).

Furthermore, unable to make its own value judgements on aesthetics or morality, it is confined to mimicking human judgement.

I thought this was an excellent translation for the 自分で美醜や善悪などの「価値判断」を行うことができず sentence. It hit the mark nicely for 美醜や善悪, and presented AI’s failure here very fluently.

Therefore, unlike humans, it cannot create a work by thinking up a number of novel ideas and selecting the most desirable ones.

As mentioned in my general comments, I was looking for two things in this sentence: a) making sure readers understand that sentence follows from all of AI’s failings rather than just the lack of judgment, and b) making it clear that AI fails at every point where we succeed. I fear E05’s proposed ‘therefore’ leads to dropped points on the first test, by locking the readers into the view that this is part of a standalone paragraph on judgment. On the second point, I would award points for putting ‘unlike humans’ (‘people’ or ‘us’ would have been even better) at the front, but I think some ambiguity remains; does AI fail at thinking up novel ideas, selecting desirable ones, creating works, or some combination of all three? Maybe, a two-clause approach with these words would have worked better (for example, ‘AI is unlike people; it can neither … nor ….’).

If you were to prompt an image-generation AI to ‘draw Tokyo Tower in the style of van Gogh’, it would produce an image in that style for you. But, it could not create said style from scratch.

E05 then nearly pulls off a brilliant translation with ‘could not create said style from scratch’. The problem is, I do not think readers will take ‘said’ in the way E05 means. They will probably assume it means something like ‘aforementioned’, and refers back to van Gogh’s style. Even so, I am bound to say thar ‘from scratch’ was a particularly inspired choice.

If you are asking whether that is not difficult even for humans, I would have to agree it is true of older media such as paintings.

Replacing ‘true of older media’ with ‘true in the case of older media’ might make this sentence a bit clearer, but readers might still need to read it two or three times to make sure they got the meaning. It is worth remembering that English is supposed to be a reader-orientated language! On closer inspection, it looks like もはや has slipped through the net, and that word is quite important in establishing the idea of a media’s life cycle for the next section.

E05 handles the last five sentences pretty well (but please see my general comments). My only real quibbles would be with ‘one kind of media’ (when there are actually two kinds) and the resurrection of quote marks in the penultimate sentence. However, I thought ‘harnesses technology’ was a nice touch for 活躍 in the same sentence, and ‘must undergo drastic reform’ was a strong finish at the end.


E20 has also produced a very accurate translation and showed some imaginative use of English while taking great paints to think about the source text. My one main recommendation would relate to register, producing language fitting for a literary journal. With a bit more reading of similar articles that have appeared in English-language journals and newspapers (and there are many to choose from!), I think E20 could have been even more confident about some of his word choices.

AI Enables “Instant Materialization” Economy

With the 「アイデア即プロダクト」の経済 of the title, E20 took the path I urged in my general comments, and avoided retroconverting katakana. Although I am happy with that approach, I am afraid I was not convinced by the choice of ‘materialization’. To be fair, I did find some corporate web sites that used ‘materialization’ in this way, but when I think about phrases like ‘materialize out of thin air’, I worry that ‘アイデア’ has not been rendered properly.

ChatGPT, a language-generation AI drawing people’s attention, is capable of not only answering any questions but also generating theses, novels or programs. Similarly, image-generation AIs such as StableDiffusion can generate pictures or photos of decent quality if you just enter words like ‘a couple flirting on the beach’ for example.

I explained in my general comments how an accurate translation might not make a fully coherent thread at the opening, and I fear E20’s translation is something of a case in point. To some readers, it might come across as telling them things they already know to no apparent purpose. Furthermore, ‘drawing people’s attention’ doesn’t seem as exciting as a 今話題 should be. From the readers’ perspective, I also see a problem with the sequence-disrupting AI/今話題side comment coming right after ChatGPT. At the risk of sounding like an English teacher, I would argue that ‘not only …. but also’ structures look unnatural in English when they are introduced into a subject-verb-object clause/sentence within which grammatical complexity already exists (whereas there are no such constraints on だけでなくin Japanese). I recommend that E20 look at the alternatives proposed by E51 or think about how to make a sentence with AI as the opening 主語 (which is what I prefer).

In the second sentence, I can see E20 has used ‘decent quality’ for 相応する, but I think that will leave readers wondering what the yardstick for this decent quality is. I suggest ‘… can make a decent attempt at generating pictures to match …’ might cover what 相応するis doing here, if we want to use the word ‘decent’.

Eventually, people will be able to fashion things like small items and furniture swiftly on their own using generative AIs in combination with devices such as 3D printers.

I thought this was an excellent translation. There is so much to like here: the use of ‘fashion’ as a verb, reinstating AI as one of the stated elements, using ‘combine’ rather than ‘connect,’ and ‘eventually’ all hit just the right note.

AI enables what we could call ‘instant materialization’ economy, where you could easily have your ideas materialized …

I have already given my opinion on ‘materialized,’ and I think defining ‘materialization economy’ economy with the word ‘materialized’ does not work well stylistically. However, I really approve of using ‘where’ to link the original definition with the そこで sentence.

On the other hand, more professional artists would find it harder to make a living because ordinary levels of proficiency wouldn’t be enough to compete with AI. Even now, the jobs-to-applicants ratio for “artists and designers” stands at merely 0.18, indicating that only one out of five applicants can land a job.

Qualifying ‘artists’ with ‘professional’ is one way to keep the focus on economics, but I think readers will be unsure if ‘more professional artists’ means artists who have greater professionalism than the creators in the previous paragraph, or a greater number of people who have the profession of artist. Maybe this problem could have been avoided by getting closer to the original 「アーティストが職業として」, where the word is used as it might be on a job description, rather than as a term by which people define themselves.

Having used the simple present tense in the previous sentences, E20 goes with ‘would’ here (as a conditional?), which looked a little bit like a hedge on tense selection.

I really liked E20’s “ratio …. stands at merely 0.18” and “land a job”; this is the sort of imaginative use of English that readers of a highbrow journal would appreciate. As I mentioned to E05, I believe that the category name is actually a quoted title in Japanese Labour Force statistics and should be identified as such if quote marks are used or given a broader title that acknowledges the Japanese nature of the source data if they are not. On a minor point, I think we need ‘around’ or ‘approximately’ before the one-in-five figure.

There are three things the present AI is void of; they are “will”, “real experience” and “value judgment”.

I think E20 made a good choice to go with ‘void of’ for 「欠けている」 to indicate we are talking about a total absence here (but I wonder, 何となく, if ‘devoid of’ would seem more natural in modern English). English quote marks are not needed here, and I explained my reservation about ‘value judgment’ as an uncountable property in the general comments.

Generative AIs like ChatGPT are, as it were, super-elites, but they are also like reactive human beings.

E20’s ‘super-elite’ may get us too far into socioeconomic territory and away from the entrance-exam-passing prowess that works best for 「スーパー偏差値エリート」in this context. I think ‘reactive’ has some promise for 待ち人間, but maybe we need to follow through a bit more to make it work fully: ‘merely reactive’ could be even better and gives the right negative nuance, especially if we can stretch 能動的に to ‘proactive’ for contrast in the next sentence, as a sort of common collocative pairing.

Never having experienced activities like clamming or campfires, they never create novels or movies based on their real experience.

The ‘never ... never’ structure worked well here, and the alliteration of ‘clamming’ and ‘campfires’ is a nice touch that our imaginary readers would appreciate.

Furthermore, their inability to make a subjective value judgment on things like beauty and ugliness or good and evil prevents themselves from doing more than following judgments made by humans. Therefore, they cannot create works as humans do by sifting through a number of innovative ideas that germinate and choosing what they think is best.

I think a lot of this paragraph is on the right lines, but it could do with a bit of polishing. We could only use ‘themselves’ as the (reflexive) object of the verb ‘prevent’ if ‘they’ were the subject, but ‘their inability’ actually has that grammatical role, so ‘them’ should replace ‘themselves’. I also thinks ‘follows’ does not quite work, although I can see what E20 means. Technically, the model is following an algorithm, and ‘follows judgments made by humans’ sounds more like waiting for someone to input the decision; staying close to ‘imitate’ might have been better.

The choice of ‘therefore’ will suggest to many readers that the inability to create works flows from the failure to make value judgments mentioned in the same paragraph, rather than all three ‘voids’ covered over three (Japanese-style) paragraphs. ‘Germinate’ could work nicely, but I think it needs to be followed by ‘in the mind’ or something like that. I think ‘cannot create… as humans do’ makes a nice start to contrasting our abilities with AI’s total inability, but maybe there is still some ambiguity over whether AI might manage one out of three for sifting, germinating, and creating, when its actual score is zero.

In short, generative AIs cannot originate works. If you instruct an image-generation AI to create a painting of the Tokyo Tower in a van Gogh style, it will generate an image accordingly. However, they can never create a style of their own.

Having ‘generative AI’ as the subject is a very good choice, and ‘cannot originate’ is a good way to convey the nuance of 難しい here. This may be a slightly unusual collocation in modern English, so ‘cannot be the originator’ might be preferable. E20 navigated the Tokyo Tower sentences well enough; ‘create a painting’ may solve one dilemma around 書いて but as the distinction between ‘create’ and ‘generate’ is so crucial to this piece, I might have chosen a different verb here, to avoid confusion.

If asked whether I find it difficult even for humans to create an original style in this age, I would have to say yes when it comes to conventional media like paintings.

This sentence became a little too convoluted. The expression ‘If asked whether I find it difficult even for humans to create …’ suggests the author is an art teacher not having much luck with getting his students to paint. Replacing ‘find’ with ‘think’ could help a little bit, but the sentence would still be very unnatural to most readers. The ‘if’ clause is just too crowded; ‘If asked whether people would find it difficult to create …’ might give us a better shot at a natural sentence. ‘Conventional’ could also become a bit of a red herring, as it is not really consistent with the explore>exhaust>innovate cycle in the next few sentences, in the way that ‘ancient’ or ‘time-honoured’ would be.

E20 handles the home straight well (but I will not reproduce it in in full here). In the final sentence on education, I think the ‘would’ and ‘apparently’ could do with a bit of tidying up, but there were some highlights before then. I particularly liked ‘creativity that capitalizes on technologies’ for 活躍 (a nice verb choice for an economist!). Also, I thought that ‘will not disappear’ had a touch of ‘the-dream-shall-never-die’ eloquence about it. Given the author’s passionate advocacy of a universal benefit income in a technology-driven society, that eloquence may not be out of place.


E42 has coupled an ability to translate with an ability to write and has come up with a very nice text. As a general comment, I think E42 could do a little more to create a consistent and authentic author’s voice. For example, there was a mix of quite casual sentences (where ‘me’ and ‘you’ were used) and quite formal ones (‘aforementioned’). On a rather minor point, I also found E42’s treatment of AI as a plural noun a bit eccentric; I think ‘AIs are ...’ would be conventionally acceptable, but not ‘AI are …’.

The AI-driven economy of ideas and instant products

E42’s title certainly gets the message across to readers clearly in plain English, which is often a good move. However, just on this occasion, it may not have been the right call, as we know the author is in the business of devising catchy titles for his economic concepts. Furthermore, although I thought ‘AI-driven’ worked well stylistically, I would question whether it fully encapsulates the sense of future possibility in可能にする.

ChatGPT, the language-generative AI oneveryone’s lips, can not only answer any question you ask it but can also generate essays, novels and computer programs among other things. On the other hand, just by typing in a group of words like ‘couple flirting on the beach’, image-generative AI like Stable Diffusion are able to generate photos or illustrations that coincide with that input.

I think the phrase ‘on everyone's lips’ is what an 今話題 should be like, so that was a good choice. However, as I commented to E20, cramming a side comment (subordinate clause) and a ‘not only …. but also’ structure so close together in the very first sentence is a bit more grammatical complexity than readers would welcome. The message of this paragraph is not about the contrast between these two AI models, so ‘meanwhile’ would be a more coherent choice for 一方 than ‘on the other hand’.

E42 has chosen ‘coincide with input’ for 相応する. I believe many readers may take this to mean coinciding in time, as if Stable Diffusion has some ESP-like power to produce output as the user is typing input. It does not help that the English teacher’s dreaded dangling modifier, ‘just by typing’, might indicate that Stable Diffusion provides its own input. Such an interpretation might seem absurd, but remember, we should never underestimate the propensity of readers to latch onto absurd meanings in a context-poor language like English.  (Continued to Part 2)