When the Algorithm Assigns the Work: AI Becomes a 'Manager' in Translation — Who Bears the Responsibility?

For a long time, the debate about artificial intelligence in the translation world revolved around the question "does the machine translate, or does the human?" Research published in recent weeks has moved the discussion to an entirely different place: in some fields, AI is no longer merely a tool that performs the work; it is becoming a management layer that assigns work, divides it up, monitors workers, and evaluates the results. So what does this transformation mean for official document translation, where errors are unacceptable? In this article, we have compiled the latest data and examined, step by step, how the official translation world answers the question: "When the algorithm is the manager, who bears the responsibility?"

Algorithmic Management: Who Assigns Work to Translators Now?

The global term for this new era is "algorithmic management"; in its simplest definition, it is a form of management in which workers' tasks, performance, and even pay are determined and monitored by software algorithms rather than human managers. The examples are no longer abstract: AI has turned into a digital manager that splits work into small pieces and distributes it, sets deadlines, scores performance, and shapes the worker's day. The picture is similar on the corporate side: according to Microsoft's 2026 Work Trend Index report, the number of active AI agents in the Microsoft 365 ecosystem grew 15-fold year over year, and 18-fold in large enterprises.

Nor is this transformation limited to certain sectors. A 2025 OECD study shows that algorithmic management no longer belongs solely to courier and ride-hailing apps. In the US, more than three-quarters of managers say their companies use 10 or more of 15 different algorithmic management tools; in Japan, 40 percent of companies use these tools. The troubling side of the coin is also visible: according to a report by The Institute of Employment Rights, AI-based algorithms are used not only for efficiency but also to monitor, classify, and, when deemed necessary, penalize worker behavior. Given that a significant share of freelance translators work through platforms, this picture directly concerns translation labor.

What Do the Numbers Say?

To grasp the scale of the transformation, it is enough to look at the major studies published recently. The data below paint a picture not of "extinction" but of rapid "role change" in knowledge-intensive professions, including translation:

• ILO: According to a 2025 study by the International Labour Organization (ILO), one in every four jobs worldwide is exposed to generative AI to some degree; but in many jobs, transformation rather than outright disappearance is expected.
• WEF: The Future of Jobs 2025 report projects major change affecting 22 percent of jobs by 2030; 170 million new roles could be created while 92 million roles disappear, with the net effect expected to be 78 million new job opportunities.
• PwC: According to the 2026 Global AI Jobs Barometer, the skills required in the jobs most exposed to AI are changing more than twice as fast as in the least affected jobs.
• Microsoft: 66 percent of AI users say they can devote more time to high-value work, while 58 percent say they are producing work they could not have produced a year ago.
• IDC: More than 1 billion active AI agents are projected to be in operation worldwide by 2029; roughly 40 times the 2025 base.
• Accenture: 86 percent of senior executives plan to increase AI investment in 2026; by contrast, only 27 percent of employees strongly express comfort with delegating tasks to AI agents.

It is worth being cautious when reading these figures. For example, because Microsoft has not disclosed the baseline value for the increase in agent numbers, it is difficult to assess the true scale of adoption; independent analyses also point out that 15-fold growth, without a comparison baseline, is a threshold easily surpassed from a low starting point. Still, the direction is clear: AI is embedding itself into translation workflows not only as a production tool but also as a coordination layer.

The 48 Percent Threshold in Publishing and Translation Labor

The example of transformation closest to translation labor came from publishing. According to a joint study by BookNet Canada and BISG, in the North American book industry, 48 percent of publishers and organizations and 46 percent of professionals actively use AI. But there is another side to the coin: 86 percent of the industry is seriously concerned about AI infringement of copyrights and proprietary content, and about inadequate legal oversight. The possibility that skilled editing, proofreading, and translation labor may be entrusted to algorithms in the name of "cutting costs" also raises questions of quality and language.

This debate is especially critical in the field of literary and artistic translation. The style of a novel, the rhythm of a poem, or the accurate transfer of a cultural reference arises not from statistical patterns but from the judgment of a translator who lives in both the source and target cultures. The crossing of the 48 percent threshold in publishing does not mean human labor has been devalued; on the contrary, it shows that the question of which work should be entrusted to the machine and which to the human must now be asked anew for every single project.

What Is Changing in the Translation Process? The New Workflow, Step by Step

The typical flow of a professional translation project today differs markedly from five years ago. Consider a concrete scenario: a machinery manufacturer wants a 200-page user manual translated into five languages. In a modern workflow, the project is first segmented by content type, terminology databases and translation memories come into play, machine translation produces a preliminary draft, and subject-matter expert translators correct and approve that draft line by line. In fields such as technical translation, where terminology demands consistency, this hybrid model—when set up correctly—can increase both speed and quality.

The critical steps of the new workflow can be summarized as follows:

1. Analysis and segmentation: The document's type, destination country, and need for official validity are determined.
2. Pre-translation: In suitable projects, translation memory and machine translation produce a draft; for official documents, this step may be skipped or subjected to strict oversight.
3. Human translation / post-editing: A subject-matter expert translator produces the text or corrects the draft from start to finish.
4. Quality control: A second pair of eyes checks consistency of terminology, numbers, names, and formatting.
5. Approval and certification: If required, the sworn translator's signature, notarization, and the apostille process are planned.

The key point to note is this: in this chain, AI can accelerate step 2 and even automate work distribution; but the professional judgment and legal commitment in steps 3, 4, and 5 cannot be delegated.

Why Can Responsibility for Official Documents Not Be Delegated?

As algorithmic systems speed up, questions arise that are increasingly hard to answer: Who made the decision—the software, the manager, or a model no one can fully explain? When a faulty decision is made, to whom does one appeal? Regulators are asking the same question: although most countries have no legal barrier to algorithmic management, frameworks such as the EU's AI regulation (EU AI Act) require these systems to be transparent and to operate under human oversight. Indeed, even in automated decision systems based on performance data, EU rules and ethical principles require that the final decision be approved by a human.

When a diploma, power of attorney, court ruling, or commercial contract is at stake, the answer to this question is legally clear: responsibility rests with the sworn translator who signs the work. An algorithm can distribute work and generate drafts; but only an authorized human can vouch for the accuracy of a translation entering the notarization and apostille process. At Ziya Tercüme, we use technology as a process accelerator, but for every document requiring sworn translation, notarization, or an apostille, the final word and the legal responsibility remain with our expert human translators. This is not merely a preference; it is the natural consequence of the chain of signature, seal, and declaration that official authorities require in translation.

Sworn Certification, Notarization, and Apostille: Who Does What?

In the shadow of the AI debate, the most frequently confused topic is the layers of certification in official translation. Sworn translator certification and notarization are separate procedures: the first is the translator's professional declaration and signature attesting to the accuracy of the translation; the second is the notary's certification of that signature. The apostille is an entirely different layer: this annotation, which ensures the document's validity abroad, is issued not by translation agencies but by the authorized official bodies in Türkiye—the provincial governor's office or district governorate (valilik/kaymakamlık) for administrative documents, and the competent judicial authorities for judicial documents. As of the date of publication, since which authority applies the apostille can vary by document type, it is necessary to confirm the current information before starting the process.

Building this chain in the correct order is decisive in preventing the document from being rejected abroad. Working with an agency experienced in notarized translation and apostille-certified translation processes clarifies from the outset which layer will be completed at which institution and prevents unnecessary back-and-forth.

Common Mistakes and Pitfalls to Avoid

The spread of AI tools has also produced new types of errors in official document translation. The pitfalls we encounter most often are:

• Submitting raw machine translation for official procedures: An unsigned machine output with no declaration will not be accepted as a translation by official authorities.
• Errors in names, dates, and numbers: AI models can corrupt proper names, ID numbers, and dates by "correcting" them; in an official document, even a single-character error is grounds for rejection.
• Confusing the certification layers: Paying for notarization where sworn certification would suffice, or applying with only a stamped translation where notarization is required.
• Confidentiality breaches: Uploading documents containing personal data to publicly available AI tools is a serious data security risk.
• Terminology inconsistency: Translating the same concept with different equivalents in the same document can create disputes of interpretation, especially in contracts.

The common denominator of these pitfalls is unsupervised automation. Especially in fields where the cost of error is high, such as legal translation and medical translation, it is indispensable that a human subject-matter expert perform the final review and approval, regardless of who produced the draft. A single wrong term in a court ruling can affect the course of a case; incorrect dosage information in a discharge summary can affect treatment.

Conclusion: Use the Tool, Leave the Responsibility to Humans

The data show that the translation profession is transforming rather than disappearing. The framework in Microsoft's report points to the same conclusion: as agents take over the execution of work, the human's space for directing the work, making decisions, and owning the outcomes expands. Indeed, the finding that 49 percent of Copilot conversations support cognitive work such as analysis, decision-making, and evaluation suggests that technology shares the decision burden rather than replacing the human. In translation, this means leaving draft production to the machine, and meaning and responsibility to the human.

The smart approach is to use AI for efficiency while never giving up the assurance of a sworn translator for work requiring official validity, copyright protection, or legal responsibility. If your document is a diploma, transcript, court ruling, or commercial contract, the first step you should take is to clarify which certification layers the receiving institution requires; the second step is to work with a translation partner that can build this chain from start to finish. Algorithms will keep distributing work; but the one who keeps signing will be human.

Sources

• Microsoft WorkLab — 2026 Work Trend Index: Agents, human agency, and the opportunity for every organization
• Hürriyet İK — Just as we feared it would take our jobs, it has taken charge of them: AI has become a digital manager
• GeekWire — Microsoft's new research finds an AI 'paradox' holding companies back
• Indigo Dergisi — The era of the AI manager: What would you do if your boss were a robot?