In the AI era, it's easy to understand how AI tools can enhance tech writing. But have you ever considered how tech writing can enhance AI? Let’s explore the possibilities of this relationship!
Prompt Engineering, a Brand of Tech Writing?
The similarities shared by the disciplines of technical writing and prompt engineering are striking. Common skills they require include the ability to formulate clear and unambiguous instructions, a solid grasp of language and grammar, and an understanding of user needs. Additionally, the fundamental principle of technical writing to understand and write for your audience can be applied to prompt engineering, where the chatbot is the prompt engineers' audience. As a prompt engineer it is essential to develop an understanding of how chatbots learn, just like tech writers must learn how to optimally present information for each audience type.
Imagine a chatbot that teaches elementary school students about plants. Custom instructions tell the bot how to behave and present information to the young students. Answers should be fun, bite-sized, uncomplicated, and captivating. It's the prompt engineers' job to translate this behavior into words, detailing guidelines and constraints. A tech writer will be more likely to provide clear instructions like "Limit your answer to a maximum of 50 words" instead of the more ambiguous "Don't make the answer too long".
So, tech writers seem to be perfectly positioned to adopt prompt engineering as an extension of their expertise. Let’s see what other opportunities we can identify to leverage the skills of technical writers in AI chatbot enhancement.
Garbage in = Garbage out
Chatbots used by enterprises must often answer questions about domain-specific or enterprise-specific knowledge such as user manuals, employee data, compliance guidelines... To achieve this, they are grounded on knowledge bases. Grounding a chatbot means providing it with knowledge that was not part of its original training set. The most used grounding technique is Retrieval Augmented Generation (RAG). With RAG, enterprise-specific information relevant to the user prompt is used as context for the chatbots' answer. This increases the accuracy of answers, while leveraging the strengths of the language model to generate text and present the information in the desired tone or format.
One pitfall is that the quality of the answers depends heavily on the quality of the information used for grounding. If the knowledge base is poorly structured, incomplete or uses ambiguous language, this is reflected in the output. Tech writers can play a crucial role here. As experts in creating high-quality knowledge bases, they ensure well-organized, complete, and clear information that enhances the quality of answers and the value of custom chatbots.
Multi-Source Data Integration Provides New Opportunities
Previously, a significant limitation of grounded chatbots was their inability to integrate different pieces of information. In response to a question, the model would select the most relevant chunks of information and generate an output without integrating them. For example, imagine a knowledge base with one file listing people's names and their business unit, and another file containing access rights per business unit. If you asked a bot trained on this knowledge base whether "John Doe has access to environment X", the bot would have been unable to provide a correct answer due to the lack of data integration.
However, multi-source data integration has changed this scenario. Some chatbots can now select relevant information from multiple sources and integrate it to achieve a deeper understanding, leading to more accurate answers. This development opens up new opportunities for chatbot enhancement by leveraging the expertise of technical writers. Tech writers can create additional content, such as terminology lists, ontologies, and FAQ lists to provide context for the chatbot. This results in smarter bots that can answer a wider variety of user questions more efficiently.
Underlining the Importance of Proximity to End Users
As mentioned earlier, a technical writer needs to think like their audience. This becomes increasingly important when chatbots are queried by the end users. When a tech writer has a good understanding of the challenges end users face, they can leverage this knowledge to test the chatbots’ performance on questions about those concepts.
For example, through regular conversations with end users of a software application, a technical writer discovers that the documentation about user settings is frequently consulted. The tech writer should make sure that the application chatbot correctly explains all user settings when prompted and adjust the documentation if needed.
Utterances also play a role in the performance of the bot; a good tech writer must be aware of or willing to investigate how its audience asks questions. Chatlogs can play an important role here.
Points to ponder
Despite the range of opportunities, AI is still in its infancy, and there are some limitations to keep in mind.
Custom instructions enable you to obtain a certain level of control over the chatbots’ behavior, but experience teaches us that chatbots are rather stubborn and sometimes completely ignore instructions. Moreover, rigorous testing can identify potential improvements and boost confidence in the chatbot’s capabilities, but AI is often unpredictable. The quality of answers to the same query often varies, indicating that control over answers is limited and the robustness of these tools remains a significant challenge.
Another limitation is that you don’t have control over user prompts. At most, tech writers can provide instructions and guidelines for writing good prompts, but it is up to the user whether they choose to follow them. And as we all know, humans can be just as stubborn as chatbots!
Conclusions
Technical writers are perfectly positioned to enhance the performance of AI chatbots. Not only can they make sure the information on which the LLMs are grounded is structured and of high quality, they can also utilize their skills to formulate custom instructions for the chatbot's persona and behavior. Additionally, tech writers are terminology and ontology experts in their domain. They can leverage this expertise to enhance an AI chatbot thanks to multi-source data integration. However, AI is still a relatively young domain with its own growing pains, and anyone using these technologies should be aware of this.
As the landscape of AI continues to evolve, so will the role of tech writers in the development of AI chatbots. Are you as curious as we are to see what other opportunities will emerge?