The integration of Artificial Intelligence (AI) into various business operations is a hot topic across industries and the technical writing departments are no exception.
While generative AI has made significant strides, its application in regulated industries raises questions about quality, security, and traceability.
This article explores viable avenues for implementing AI in a Component Content Management System (CCMS) for technical documentation, and more specifically in regulated industries.
About the Author
Hi, I’m Jean-Luc Borie, and I’m excited to share the insights and research presented in this article. The inspiration to create this content was the DITA Molière Conférence 2023, which took place on the 31st of October 2023.
I’m actively involved in the DITA Molière association, which aims to bring together the community of DITA users in French-speaking countries and develop the use of DITA.
If you share an interest in DITA and would like to become a part of this community, feel free to visit the DITA Molière association website, where you can find more information about upcoming events and other news.
If you have any questions or projects to embed AI in your technical documentation production process, contact us!
What type of AI can be used for technical documentation?
Understanding the types of AI in the market is very important. While generative AI has been making headlines, it’s important to emphasize that there are different types of AI. Focus on those you are already using or may consider to improve the technical documentation process:
Natural Language Processing (NLP)
Refers to artificial intelligence systems and applications that utilize natural language processing to understand, interpret, and generate human language, enabling interactions and tasks related to language and text.
This subfield of AI focuses on enabling computers to understand, interpret, and respond to human language in a valuable way. Indeed, NLP has a longstanding history of application in technical documentation. It has been employed to enhance translation within Translation Management Software, maintain precise terminology in terminology management systems, and aid in extracting pertinent information from existing documents and other aspects. As demonstrated below:
- Information Retrieval: Organizations can use AI systems to rapidly search for technical information in extensive documentation databases, simplifying document maintenance and updates.
- Content Generation: Generative AI, such as language models, can be used to create technical descriptions, usage instructions, and other content based on specific requirements. The most well-known example, of course, is ChatGPT.
- Content Classification: AI can assist in automatically categorizing technical documents based on their content, facilitating documentation management and organization.
Non-Natural Language Processing (Non-NLP)
Represents artificial intelligence applications and techniques that are not primarily focused on natural language processing tasks but encompass a wide range of non-language-related applications.
- Data Analysis: AI techniques can be used to analyze complex technical data, generate graphs and visualizations, and detect trends or anomalies.
- Process Automation: AI-based chatbots can be integrated into technical support systems to answer common user questions and guide users through technical procedures.
- Predictive Maintenance: AI can be used to predict hardware failures or potential technical issues, allowing for proactive maintenance planning.
The fundamental equation of AI
Artificial Intelligence (AI) is a broad field with various sub-disciplines, each offering unique capabilities and challenges. Below is an equation that has helped me better understand what artificial intelligence is. We will revisit this equation a bit later.
AI = Model + Data + Infrastructure + Skills
- Model: This is the algorithmic aspect of AI, which can range from simple linear regression models to complex neural networks.
- Data: AI models are only as good as the data they are trained on. Quality data is essential for effective AI implementation.
- Infrastructure: This involves the hardware and software that support the AI model, including servers, GPUs, and cloud services.
- Skills: Human expertise is required to design, train, fine-tune and maintain AI models. This includes data scientists, machine learning engineers, and domain experts.
If you have any questions or projects to embed AI in your technical documentation production process, contact us!
What is the potential of AI in a CCMS?
Integrating AI into a Component Content Management System (CCMS) is beneficial, particularly in large industries with substantial technical content. AI assists in managing an overwhelming volume of content components.
When you combine the composite nature of content produced with documentation standards like DITA XML or S1000D with the vast volume of technical content maintained by large industries, you end up with a number of content components that are absolutely impossible to manage on a human scale, without the assistance of algorithms.
It then becomes quite clear how AI can bring significant potential to the context of technical documentation within large industries, as you can see in the examples below:
Smart Tagging for Automated Classification
AI algorithms can automatically classify DITA files and associated resources based on different criterias, such as:
- Content-Based: AI can recognize the type of content within a file, whether it’s text, image, or video, and tag it accordingly.
- Link-Based: If a diagram is referenced in a topic categorized under “A,” the AI can recommend tagging the diagram as “A” as well.
Suggestions for Content Reuse
One of the most time-consuming aspects of technical writing is creating new content that already exists in some form within the organization. AI can Identify similarities in existing content and suggest opportunities for reuse, thereby enhancing both efficiency and consistency.
Context-Based Filtering of Reusable Components
In a CCMS, multiple components like text blocks, images, or code snippets can be reused across various documents. AI can Intelligently filter these components based on the context in which they are to be used and offer suggestions, making the documentation process more streamlined and efficient.
What are the challenges of using Generative and “Black Box” AI models in regulated industries?
The feedback from our clients in regulated industries is nearly unanimous: most of the experiences with using AI for technical documentation are on hold because they raise too many concerns about security, legal aspects, liability, and even simply the risk of non-compliance.
In reality, when we delve into the matter, we realize that it primarily pertains to experiences related to the use of Generative AI and more specifically, it concerns ‘black-box’ GEN AI, like ChatGPT, whose functioning (Model) and the content on which it has been trained (Data) are not precisely known, and this presents of course numerous challenges for regulated industries.
But there are other types of AI, some of which have been used quietly for a long time in regulated industries, such as for translation and terminology. Regulated industries express concerns about using generative AI models due to security, legal, and compliance issues. “Black box” Generative AI models, like ChatGPT, pose specific challenges which we can analyze below;
Reliability Concerns
The accuracy and reliability of information is strongly questioned. How can we ensure that the technical information generated by ‘black-box’ Generative AI is both accurate and reliable?
Unfortunately, we cannot do so because generative AIs, like ChatGPT, undergo training with an unknown algorithm and unknown data. The results are therefore unpredictable, and far too risky to use for producing technical documentation for products that could pose risks to public health.
Non-Compliance Risks
In the context of using “black-box” Generative AI come from various factors:
- Ensuring that the content generated by such AI models adheres to the stringent quality standards mandated by regulatory bodies presents a considerable challenge.
- Potential for data protection issues, as the AI may inadvertently incorporate sensitive or personal information into the documentation, potentially leading to violations of data protection laws.
- Ethical considerations surrounding the use of “black-box” Generative AI in documentation remain not entirely clear, which could introduce additional compliance risks.
Transparency and Auditability
Particularly when dealing with Black Box Models. The inherent lack of transparency in how these models generate content can present a significant challenge, especially when subjected to audits and scrutiny.
In addition, establishing accountability for AI-generated content can pose a complex challenge, leading to a gray area in regulated industries where defining responsibility for such content is not straightforward.
Data Security
Concerning the confidentiality and protection of sensitive information, raises valid concerns when considering the use of “black-box” Generative AI.
These concerns collectively underscore the unsuitability of “black-box” Generative AI for application in regulated industries and, more broadly, in the field of technical documentation.
Practical Steps to Leverage AI in Technical Documentation While Maintaining Compliance
Recall the AI equation?
AI = Model + Data + Infrastructure + Skills
There’s a strategic approach to master all the terms of the equation and comprehensively address transparency, data security, and reliability concerns – leveraging an open-source AI system.
Consider Open Source AI
And if you are still asking “why should I consider open source AI?”, read some of the reasons below:
Open Source
Opting for open-source AI models comes with the significant advantage of transparency. This transparency is pivotal, especially when it comes to audits, as it allows us to gain a full understanding of how the model operates. Additionally, it offers the flexibility to tailor the AI to meet our precise requirements.
Your Data
Regarding data, a crucial process involves fine-tuning the AI model. Initially, the model can undergo pre-training on a general open-source dataset. It can be fine-tuned using your specific data, a step that guarantees relevance and accuracy in line with your needs.
Your Hosting
When you choose to host the AI on your infrastructure, it provides an effective solution to many data security concerns. This approach ensures that sensitive information stays firmly under your control, offering an added layer of protection.
Limited Scope
When it comes to technical documentation, particularly within regulated environments, it’s advisable to steer clear of generative AI models. This cautious approach helps mitigate potential risks and ensures compliance with regulations.
Experiment!
Give Hugging Face a try! We strongly recommend exploring the capabilities of a leading open-source AI platform. You can swiftly set up internal Proof-of-Concepts to assess AI’s applicability within confined areas using Hugging Face.
To embark on this journey, we recommend putting together a skilled project team. This team should consist of an information architect from your documentation team, someone who understands your content intricacies, and a Python developer with a solid background in Deep Learning, particularly with frameworks like PyTorch or TensorFlow.
By bringing together this team, you’re not just preparing for success: but also ensuring that your exploration of AI is founded on a strong technical foundation.
We firmly believe that enterprise AI, especially in the field of technical documentation, has the potential to excel through focused and controlled projects. These projects have the ability to deliver substantial value to your organization. They operate within specialized models and controlled environments, using your internal infrastructure and data, leading to reliability, security, and cost-effectiveness.
This approach sets the stage for a successful integration of AI into your technical documentation processes.
