2025/02/08

機械が学術論文を書き、読み始めるとき──学術コミュニケーションを考え直す

 

人工知能が毎時間のように科学論文を量産し、あるアルゴリズムが別のアルゴリズムを参照し、以前に機械が生成した研究成果を引用する。その過程に人間の関与はほとんどない。そんな未来図を想像してほしい。荒唐無稽だろうか。そうかもしれない。だが、AI支援のライティング・ツールの普及は、学術コミュニケーションの今後をすでに示唆している。高度に洗練されたシステムは、説得力ある文献レビューを作成し、筋の通った主張を立て、それを学問的な体裁にまで仕上げることさえ可能だ。一方、従来は著者としての役割を担っていた人間は、いまや運転手というより傍観者に近い存在へと移り変わりつつあるように見える。

実は、高度なAIが一般化するずっと前に、ドイツの社会学者ニクラス・ルーマンは驚くべき考えを提唱していた。「コミュニケーションを行うのはコミュニケーションそのものである」という主張である。人が情報を交換していると考えるのが普通だが、ルーマンによれば、会話そのものは自律したオートポイエティック・システムであり、個々の精神とは大きく独立している。言葉はある発話から別の発話へ跳ね渡り、論文同士が相互に引用・応答を繰り返すことで、コミュニケーションの連鎖は自立的な生命を帯びる。ルーマンの視点では、人間はコミュニケーションの「中心的プレーヤー」ではなく、その環境の一部にすぎない。

そして、テキスト生成AIの登場によって、このルーマンの理論は一層説得力を増したように見える。誰、あるいは何が言葉を生み出しているかがさほど重要でないとすれば、人間の手がほとんど介在しないまま、自己言及的な論文が雪だるま式に増えていく学術世界が出現しても不思議ではない。あるAIが論文を書き、別のAIがそれを読み、引用し、さらに第三のAIがその議論を要約する――この連鎖は際限なく続くかもしれない。「コミュニケーション同士がコミュニケーションする」システムは、ほとんど自動運転のように動き、人間が追いつけない速度でテキストを生み出していく可能性がある。

しかし、結論を急ぎすぎるべきではない。現状の最先端AIであっても、真の理解を持たない。統計的パターンをもとに文章を組み立てているだけであり、深い意味や道徳的含意を熟考しているわけではない。一方、現代の学術コミュニティは、不毛な情報をふるい落とし、厳密性を確保し、うわさや臆測に流されないようにすることを理想としている。事実検証や証拠の解釈、疑わしい仮説への挑戦など、人間の厳格な目が学問の進歩を支えている。もしAIが膨大な文章を量産し続ける一方で、人間による十分な監視が行われなければ、学術の信頼基盤そのものが脆くなる危険がある。

とはいえ、AIが生み出すあらゆる成果を学問への脅威とみなすのは早計だ。膨大な研究資料を効率よく要約するツールとして活用すれば、人間の研究者はより創造的あるいは批判的な作業に時間を割くことができる。たとえばデータベースを延々と探索する代わりに、AIに文献をまとめさせ、その結果をもとに新たな問いを立てたり、解釈したりすることが可能だ。このように機械が人間の知的活動を補強する形で機能すれば、研究はむしろ促進されるだろう。

ここで大事なのは、学術コミュニケーションのあり方を根本から捉え直す必要があるという点だ。古典的な「人間の著者が、人間の読者に向けて知識を生産する」というモデルは、著者としても読者としても機械が加わった瞬間に崩れ去る可能性がある。今後は、人間が「どの問題に取り組むか」「どのように成果を応用するか」を決め、AIが膨大な情報を統合・要約するというパートナーシップに近づくかもしれない。ただし、AIが自らの引用を延々と再生産し、誰もその正否を確かめないまま、見かけだけもっともらしい情報を無限に増殖させる危険もある。

要は、学問を意味あるものにする根幹を守り抜くことが肝要だ。華々しい脚注の羅列だけでは何の価値もない。たとえルーマンが説く「コミュニケーションのみがコミュニケーションする」が正しいとしても、人間には好奇心、創造力、道徳的判断力、そして知識の正しさや価値を問い続ける姿勢がある。人間のリーダーシップがあれば、AIを使って研究を強化し、知見をさらに素早く明確に共有できる。逆に、そのリーダーシップを放棄すれば、学術の「会話」はエコーチャンバーに堕し、言葉は増殖しても知恵に至らない危険がある。

これは危機というより警鐘に近いだろう。AIが学術出版の景観を変革するなかで、研究者や学術機関、社会全体が「何が信用に値するコミュニケーションなのか」を改めて定義しなければならない。著者や責任に関する新たな基準、より強固な査読体制、機械生成の知識を扱う倫理面での幅広い議論が求められる。ルーマンの示すように、コミュニケーションが自己完結的に持続する可能性があるとしても、続くだけで自動的に価値があるわけではない。どの情報が発信や受容に値するのかを見極めるためには、人間の判断が不可欠なのだ。

この視点を取ると逆説的な事実が浮かび上がる。AIがコミュニケーションの量を急速に増やすほど、私たちはそもそも「何のために、どのようにコミュニケーションしているのか」をより深く省みざるを得なくなるという点である。その内省を通じて、学問の真の目的――紙幅を増やすのでなく、意義あるアイデアを発展させること――に近づくことができるかもしれない。機械が十数本の論文をいとも簡単に書き上げる世界では、知識は結局人間にこそ仕えるものであって、その逆ではないという原点を思い出す必要がある。むしろ、人間の知性と倫理観の重要性は従来にも増して高まる可能性がある。そしてその過程で、私たちは知識を共有する方法そのものを再発明することになるかもしれない。


*****

上の文章は、PCの前でお昼の弁当を食べながらほとんどAIによって作成したものです。私が行ったことは以下のことだけです。さすがに弁当を食べ終わる前には作業が完了しませんでしたが、いつもなら腹ごなしのためにとっている小休憩の時間でこの文章をこうしてブログに掲載しています。


1. OpenAI o1 pro の Deep Researchに以下の調査を命じる:Did Niklas Luhmann, a sociologist, say something like, "It's not humans but communications that communicates"? I want to extend an argument that when academic papers proliferate due to the exponentially enhanced productivity of artificial intelligence, it is the productions and interactions by artificial intelligence, rather than humans, that conduct academic communication. So, can you first investigate what Luhman said about the subject or agent of communication? Second, can you assess the validity of my argument above?
2. Deep Researchが調査を始めるにあたっての3つの質問に答える(詳細省略)。
3. Deep Researchが約4分後に出してきたレポート(英語)を読む。
4. OpenAI o1 pro に上のレポートに基づきエッセイを書くように指示する。エッセイの条件はいくつか提示(詳細省略)
5. OpenAI o1 pro がまもなく出力したエッセイを読み、もう少し読みやすくするように指示(詳細省略)
6. OpenAI o1 proが改訂したエッセイの2箇所だけ訂正して、日本語に翻訳するように指示。
7. OpenAI o1 proはやはり「です・ます」調の日本語翻訳を出力したので、それを「だ・である」調に変換するように指示。
8. OpenAI o1 proの出力を1箇所だけ訂正する。


ちなみに、上の日本語翻訳の元になった英文は以下のとおりです。

When Machines Write and Read Academic Papers: Rethinking How Knowledge Is Born

Picture a future in which artificial intelligence churns out scientific papers by the hour—one algorithm referencing another, manuscripts citing earlier machine-produced findings, all without much human intervention. Far-fetched? Possibly. But the proliferation of AI-assisted writing tools already reveals a glimpse of how scholarly communication may evolve. A sophisticated system can now generate a convincing literature review, propose plausible arguments, and format its findings with academic polish. Meanwhile, the human role in this process—traditionally that of the author—begins to look less like a driver and more like a bystander.

Long before advanced AI became mainstream, the German sociologist Niklas Luhmann proposed a startling idea: only communication communicates. In other words, we typically think of people as exchanging information, but Luhmann argued that conversation itself behaves as an autonomous system, technically known as an autopoietic system, largely independent of individual minds. Words bounce from one utterance to another, articles cite and respond to each other, and the chain of communication takes on a life of its own. Humans, in this view, become part of the environment of communication, rather than its central players.

Now that text-generating AI has entered the picture, Luhmann’s theory seems prescient. If it truly doesn’t matter who—or what—produces the words, we might soon witness an academic universe where self-referential papers proliferate with minimal human involvement. One AI writes a manuscript; another AI reads and cites it; a third AI summarizes the debate. And so on, seemingly ad infinitum. The system of “communications communicating” could continue largely on autopilot, producing text faster than any person could reasonably keep up.

However, we should pause before jumping to conclusions. Even the most advanced AI, at least so far, lacks genuine understanding. It generates prose by scanning statistical patterns, not by reflecting on deeper meanings or moral implications. The hive-mind of modern academia, by contrast, is supposed to sift insights from nonsense, champion rigor over rumor. Human oversight remains the bedrock of scholarly progress: verifying claims, interpreting evidence, and challenging questionable assumptions. If AI is left to churn out endless content without enough human scrutiny, the entire edifice of academic credibility could become a house of cards.

Yet it’s too simple to label all AI-generated work as a threat to scholarship. Tools that efficiently summarize massive amounts of research could free human thinkers for more creative or critical tasks. Rather than searching databases for hours, a researcher might let an AI compile relevant sources, then devote time to interpreting results and framing new questions. In that case, machines would support rather than supplant real intellectual activity.

And there, perhaps, lies the most important point: we need a new way of thinking about academic communication. If the classic model is “human authors produce knowledge for human readers,” that model breaks down once machines join the conversation, both as authors and readers. The future might better resemble a partnership, with humans deciding what questions are worth asking and how to apply the answers, while AI supercharges our ability to synthesize mountains of information. A key danger, of course, is a scenario in which AI begins referencing its own references, spinning out superficially convincing material with no one left to check if any of it is correct.

The guiding principle must be to preserve what makes scholarship meaningful. We want more than strings of footnotes impressively stacked atop one another. Even if Luhmann is right that “only communication communicates,” we humans still bring essential qualities to the table: curiosity, creativity, a capacity for moral reasoning, and a stake in whether our knowledge is good and true. With human leadership, AI might be harnessed to strengthen research and share insights with new speed and clarity. Without such leadership, the “conversation” of science risks becoming a chaotic echo chamber, where words multiply endlessly but rarely coalesce into wisdom.

In the end, this may be less a crisis than a wake-up call. As AI transforms the landscape of academic publishing, scholars, institutions, and the public will have to clarify what truly counts as trustworthy communication. We may need new standards of authorship and responsibility, more robust review processes, and broader debates about the ethical use of machine-generated knowledge. If Luhmann’s lesson is that communication can become self-sustaining, then our lesson must be that self-sustaining chatter is not automatically worthwhile. It takes human judgment to decide what is worth saying—and hearing.

By adopting this perspective, we face a paradox: as AI ramps up the sheer volume of communication, it forces us to reflect more deeply on why, and how, we communicate in the first place. Perhaps that reflection will bring us closer to the real purpose of scholarship: to advance ideas that matter, not just accumulate pages. In a realm where machines can spit out articles by the dozen, the rest of us might remember that knowledge ultimately serves people—not the other way around. We might discover that human intelligence and moral insight matter more than ever. And in the process, we just might reinvent how we share what we know.





機械が学術論文を書き、読み始めるとき──学術コミュニケーションを考え直す

  人工知能が毎時間のように科学論文を量産し、あるアルゴリズムが別のアルゴリズムを参照し、以前に機械が生成した研究成果を引用する。その過程に人間の関与はほとんどない。そんな未来図を想像してほしい。荒唐無稽だろうか。そうかもしれない。だが、AI支援のライティング・ツールの普及は、学...