Part 3: Utilizing Generative AI in Medical Translation — Collaboration Models Between Humans and AI and Key Points for Implementation

1. The key to utilizing generative AI in medical translation is "collaboration," not "full automation"</h>

From the previous verification, it has become clear that the current use of generative AI is not about completely replacing human translators but realistically utilizing it as a support tool. In quality-prioritized fields like medical translation, it is premature to pursue the ideal of "dramatically increasing translation speed by leaving it entirely to AI." Rather, it is important to build a collaborative model where humans and AI each take on their areas of expertise.

In Human Science's demonstration, we also considered a workflow where ChatGPT was entrusted with full-text translation and humans checked and corrected it. However, the result revealed that many errors and unnatural parts still remained in the translations generated by AI, and removing them completely could require more effort and time than the usual translation process. In fact, when humans reviewed the draft translations created by AI, multiple mistakes that even the AI itself failed to detect were found, highlighting the risks of carelessly "outsourcing translation entirely to AI."

In short, meeting the high-quality standards of medical translation requires human expertise and meticulous attention, and AI remains limited to a supporting role in that process. So, what specific collaboration models can be considered?

As shown in the timeline above, the collaboration between generative AI and human translators is currently still evolving. In the future, as AI performance improves and becomes optimized for specialized fields, human involvement may gradually decrease.

For example, AI might instantly translate standardized sentences and repetitive parts, while translators focus on difficult expressions and subtle nuance adjustments, making such a division of roles more common.

However, in our view, no matter how much AI advances, "human involvement will never be reduced to zero." The act of translation requires contextual judgment, consideration for the reader, and creative ingenuity backed by specialized knowledge, making the human eye indispensable for ensuring final quality. What is important is the question of "who" and "how" the evolving technology is mastered, and the collaborative model that leverages the strengths of both humans and AI is the direction we should aim for.

2. Utilizing Generative AI: Key Points for Implementation and Operational Design</h>

So, what points should we pay attention to when utilizing generative AI in practical work? Let's organize best practices for collaborating with AI in medical translation.

2-1. Use AI Appropriately According to the Task</h>

Generative AI has tasks it excels at and tasks it struggles with. For example, AI is effective for translating fixed expressions, checking terminology consistency, and detecting simple errors. On the other hand, translation requiring advanced specialized knowledge and adjustments considering the overall flow of the text remain human roles. It is important to clearly separate the parts entrusted to AI and those handled by humans, enabling efficient collaboration.

2-2. Ensure Security and Compliance</h>

Since medical information is involved, security measures are essential when introducing AI into the internal environment. By implementing mechanisms to prevent data misuse through OpenAI API integration as introduced here, or by using AI tools that operate on internal servers, you can utilize AI while protecting confidential information. It is also indispensable to establish operational rules that comply with terms of use and legal regulations.

2-3. Do Not Take AI Suggestions at Face Value</h>

It is important to always review the outputs generated by AI. Take a moment to verify the translations and correction suggestions provided by AI, and only adopt those that are appropriate. Especially with models like ChatGPT, which can confidently present incorrect information, the rule of thumb is that the "final decision rests with humans."

2-4. Devise Prompts and Settings</h>

To get the desired results from AI, it is necessary to provide appropriate instructions. For example, you can instruct it to "leave this term untranslated in English" or request it to "adjust the style at the end of sentences," thereby controlling the output by adjusting the prompt (instruction text). Accumulating prompt know-how within the company and preparing reusable templates is effective.

2-5. Conduct phased introduction tests</h>

Instead of applying it to all projects at once, it is better to start with internal verification and trials in some projects. For projects with strict quality requirements, refer only to the proposals, and for projects with more flexibility, try adopting AI translation experimentally to accumulate experience while minimizing risks. Reflect feedback in the AI utilization process and gradually optimize it.

By keeping these points in mind, generative AI can become a powerful assistant. In fact, within Human Science, we have confirmed that the introduction of AI proofreading tools has reduced omissions and terminology inconsistencies, contributing to an overall improvement in translation quality. The key is to skillfully use AI to reduce the human workload while ensuring that the final quality is responsibly guaranteed by humans, maintaining this balance.

Free Download

Five Challenges and Solutions in Medical Translation

Introducing common challenges faced by many companies in medical translation and specific solutions.
Additionally, we present multiple success stories of cost and delivery time reduction.