AI Translation And Data Security

AI tools and cloud-based services are revolutionizing the publishing world, together with many other industries. For authors, editors, translators, and publishers, both AI and the cloud can deliver huge savings in terms of time and cost.

But the use of largely unregulated AI technology also raises concerns around the security of intellectual property (IP) – particularly prior to publication, and especially concerning the use of AI translation services built on Large Language Models (LLM).

Should the publishing industry be concerned? We spoke to SNAP Innovation’s Sales and Project Manager, Robert Janoska, to discover more.

Your data is the real cost of using ‘free’ AI services

There’s an important distinction to be made between ‘free’ AI tools – such as some versions ChatGPT – and professional B2B AI translation services that require a fee. No AI tool is truly free. So if you’re not paying to use an AI service, then the fee is essentially the data you share with it. Data is hugely valuable because an LLM only gets better by being trained on large volumes of data.

Let’s say you have an as yet unpublished book that needs translating into different languages. If you use a free AI translation tool, this content is then in the public domain. Not only can it be used to train an LLM, and it could even be sold to third parties. There’s also no control over where the information is stored. It’s like using an unsecured network to access the Internet – your data is left vulnerable to misuse.

Pay to ensure your data remains your data

Paid-for AI services are different. Services like AI translation and image recognition from providers such as SNAP Innovation are built on professional B2B services. There’s obviously a cost attached to this. But the advantages and peace of mind are priceless. For example, you can choose where your data is stored. That means you can select data centers in the EU only, so they fall under the General Data Protection Regulation (GDPR).

Using a professional paid-for AI service has other benefits as well. Your data remains your data. It’s not stored and it’s not sold to third-parties. When it comes to translation, AI doesn’t always capture the nuances of language – particularly in fields such as science, law, and finance. That means you need to invest time and money in training your LLM. What’s important here is that with paid AI translation services, your training information also remains your own – no-one else has access to it.

The other benefit is that all the data generated by AI – such as the translation of a book into another language – remains yours because you’re paying for it. And, if your agreement with the AI service ends, the data – including your Intellectual Property – is still yours.

AI translations still need the human touch

Right now, AI translation works brilliantly and reliably for smaller, less complex content – such as image descriptions. It can also work for longer texts, but the quality of the translation depends on the subject and AI’s training or dictionary.

AI translation services do a great job, but like a lot of AI writing tools, they sometimes still require a human editor. It’s in more complex cases that a human translator can help proof and edit AI-generated copy.

Nuances in meaning, the subtleties of language, and specific terminology in complex fields such as law mean skilled humans must iron out errors in translation, and give AI-generated copy the correct look and feel.

However, with the right training and datasets, it’s increasingly possible to train AI to adapt translations to a particular field. Similarly, a terminology dictionary can help AI to translate specialized terms correctly.

Keep your data safe in the cloud

Using cloud-based services from big techs like Microsoft, Google, and AWS also wraps another layer of security around your content. In fact, they give everyone enterprise-level security features, to keep your data safe.

PublishOne – integrated with AI Translation services from SNAP Innovation – is built on Microsoft Azure. Microsoft won’t reuse customer data submitted to the service or output content to train or improve the Azure OpenAI foundation models. Not only that, Microsoft is not the owner of their customers’ output content – so your translations remain your property.

Discover more about how the world’s friendly publishing platform – PublishOne – and our service partner – SNAP Innovation – work together to keep your data safe. Contact us