Understanding how AI can facilitate collaboration in a technical communication context, which can be important when using tools such as ChatGPT to plan and generate content.
By Filipp Sapienza | Member
Artificial intelligence or “AI” is expected to have a significant impact on the field of technical communication by automating many tasks, such as writing and editing technical documents, creating instructional videos, and providing customer service. Additionally, AI-powered tools will be able to analyze large amounts of data and identify patterns, which can be used to improve the effectiveness of technical communication by tailoring it to the specific needs of the audience. AI will also enable the creation of more interactive and personalized technical communication experiences, such as virtual assistants that can provide real-time support. Overall, AI is likely to significantly improve the efficiency and effectiveness of technical communication, making it more accessible and user friendly for both technical and non-technical audiences.
This response was produced by ChatGPT (Chat Generative Pre-trained Transformer), an artificial intelligence chatbot launched by the firm OpenAI in November 2022. ChatGPT responds to natural language queries submitted in a web form. ChatGPT differs from a search engine in that, while the search engine retrieves existing content, ChatGPT both retrieves and produces content as natural speech.
For example, when I asked Google, “How does one groom a Jackchi dog?”, it displayed a set of links to information about JackChi dogs and a couple of how-to sentences.
When the same question is posed to the ChatGPT interface, the results were more elaborate.
Core Competencies of Technical Communication
According to Rainey, Turner, and Dayton (2005), the core competencies of technical communication encompass the following:
- Skills in collaborating with both subject matter experts (SMEs) and coworkers
- Ability to write clearly for specific audiences directed by clearly defined purposes
- Ability to assess and analyze user needs
- Ability to learn to use technologies
- Ability to take the initiative (be a self-starter) and to evaluate one’s own work and the work of others.
What do AI applications like ChatGPT portend for the future of technical communication? Let’s start by focusing on each of these core competencies.
Collaboration with SMEs and Coworkers
The ChatGPT interface can facilitate collaboration by connecting coworkers and SMEs with common knowledge repositories, from which to discuss, debate, and reach consensus on the content generated by the chatbot. This feature is necessary because the chatbot generates a lot of verbiage. In fact, OpenAI acknowledges that ChatGPT “sometimes writes plausible-sounding but incorrect or nonsensical answers.” This behavior is common to large language models and is called hallucination.
Generating extra textual content brings up questions of information integrity. When asked to comment on the accuracy of ChatGPT’s response, “JackChi person” Cory Sapienza told me, “Brushing a JackChi weekly is not an incredibly realistic goal, because their fur will shed at incredibly different rates throughout the year. Besides, it’s more important to take care of their nails and ears.”
Another potential drawback is that AI may prevent collaboration if technical communicators rely on it alone rather than consult with live SMEs.
AI models require “training” from real human users. Some questions about training might include:
- What kind of SMEs will train the chatbot?
- How “expert” will the SMEs be?
- What technical skills are necessary to train a chatbot?
- How do we account for sociocultural biases that human SMEs might embed in the AI?
Clear Writing for Diverse Audiences and Specific Purposes
Unlike static documents, AI-generated content allows users to ask for clarification of a particular step. Upon confusion about cleaning a Jackchi’s ears, I prompted ChatGPT to elaborate on that subject.
A user can also input casual language, and ChatGPT will reply in a friendly voice. When I wrote, “Oops! I forgot step 4,” ChatGPT replied, “No worries! Here is the updated step-by-step guide to cleaning your Jackchi’s ears, including step 4…”
Sometimes a follow-up request, if re-requested, will generate slightly different content from the original; other times the AI will repeat the content from the initial prompt, thus sidestepping the request for clarification. There can also be a lack of specificity in the provided content. For example, ChatGPT doesn’t really distinguish between dog breeds. When I substituted “JackChi” with “Labrador Retriever,” I received almost the exact same content.
Provision of tailored content is not fully consistent in the ChatGPT system. While it sometimes provides nearly identical advice for different dog breeds , in other ways, it does customize the format of the advice for different groups. Consider the following:
- AI can adapt text for different audiences. When asked to rewrite the instructions for a child, ChatGPT produced the following:
- ChatGPT can also employ different voices. When asked to reproduce the instructions in the voice of Geoffrey Chaucer, it rendered the following:
- ChatGPT can produce translated content on demand. When asked to write the instructions in Spanish, it gave the following response:
You might notice that it produced the children’s (and not original adult) Spanish version of the instructions. ChatGPT remembers only the immediate context of the discussion. In this respect, ChatGPT is capable of producing not only an interactive form of technical documentation but also one that is a product of producer-user collaboration.
Ability to Assess and Analyze User Needs
Usability is a core task of the technical communication field. Being a text-generator, ChatGPT’s capabilities in usability are somewhat limited. Other products on the market, such as odaptos, perform many user research functions, including reading emotional responses of users, System Usability Scale (SUS) scoring, heatmaps, automatic transcription of audio interviews, and expedited analysis of user metrics. Future releases of ChatGPT are planned to include more robust graphical components and UI analysis.
ChatGPT can provide How-To’s and answers to some user research tasks. ChatGPT was prompted to explain the SUS lostness score and produced the following:
When asked to design a usability test, ChatGPT produced a more general response.
In its current manifestation, I do not have the impression that the AI interface is capable of designing an adequate research design. Intermediate steps in the response (such as number of users to test) require increasingly specific follow-up questions (asking it to specify how many for a particular type of test). Even then, I found some of the responses not completely accurate (for example, the frequent response to “number of users” to test was 16–20, when according to Nielsen and Norman, 5–8 users is acceptable for most user tests).
ChatGPT will produce working HTML and CSS code for a basic responsive website that conforms to user interface standards. ChatGPT produces the computer code for you to copy to your clipboard and paste into an HTML editor, save as an HTML file, and load into a browser. As pictured in Figure 9, the results work well.
The left part shows the website squeezed to below 600 pixels wide, while the right shows the website when expanded beyond 600 pixels.
ChatGPT also produces errors when writing JavaScript and Python code. I asked ChatGPT to write JavaScript code that would track where a user clicks on a web page. It produced code that adds an eventListener() to the web page and saves where users click to a JSON file, but when I pasted this code into my editor and loaded it in the browser, it didn’t work. I told ChatGPT, “It doesn’t work. Please help.” The answers from ChatGPT were very general and not particularly useful.
On the positive side, ChatGPT does produce well-commented code. Future iterations of AI-assisted coding should be more firmly embedded in the programmer’s system ecology, where it would be able to debug code with knowledge of the system(s) on which the code will run.
ChatGPT will also produce DITA, but success depends on how the question is asked. When I asked, “Structure the above JackChi grooming process in DITA,” it said, “I apologize, but as a text-based AI model, I am not able to structure the instructions in DITA format.” When I reworded the question to, “provide a CONREF example from the JackChi instructions,” it gave a robust and helpful response that includes not just code but additional explanatory text about the feature.
Future AI models are expected to be multimodal, meaning having the capacity to accept and produce images and sounds that may assist user interface designers. We may be in for an AI that converts big data inputs into near real-time user interfaces.
Ability to Learn Technologies
The current ChatGPT AI is “trained” by real people using a variety of learning methods. The model also solicits user feedback to correct erroneous information. When I searched for references written about a specific aspect of the Russian theorist Mikhail Bakhtin (a philosopher often mentioned in technical writing education), the answer was not only erroneous, but included references that didn’t exist.
I posed the issue of learning technologies to ChatGPT, and it gave the following response:
As an AI language model, ChatGPT does not “learn” new technologies in the traditional sense. Instead, its capabilities are based on the data it was trained on and the algorithms used to process that data. However, updates and improvements to the algorithms and training data can be used to enhance ChatGPT’s abilities and keep it up to date with new technologies. Additionally, human experts can provide feedback and input to help refine and improve the model.
Ability to Take Initiative and Evaluate Work
Of all the core competencies, the ability and willingness to take initiative may be the most important skill a technical communicator needs to work with AI. For example, practitioners will likely need to learn some data science if they are going to work with multimodal AI that can produce near real-time interfaces using big data. Coding skills may become more important to configure systems.
It’s unclear what AI will look like in the distant future, but in the near term, there will be a need for technical communicators to evaluate what it produces, and to assess the accuracy, wording, and graphics used to train the systems. Since taking initiative is already built in for technical communicators, the profession is well suited to adapting to an AI future.
Will AI Take Jobs in Technical Communication?
I posed this question to ChatGPT, and this is the answer.
AI has the potential to automate many tasks in technical communication, such as writing and editing technical documents, creating instructional videos, and providing customer service. This could lead to some job loss in the field, as some tasks may no longer be performed by humans. However, it’s important to note that AI is not likely to replace all jobs in technical communication, but rather it will change the nature of these jobs and create new opportunities.
AI can assist and augment the work of technical communicator, by taking over repetitive, low-value tasks that are time-consuming and allowing technical communicator to focus on more complex, high-value tasks such as strategy, research, and engagement. AI can also help to improve the efficiency and effectiveness of technical communication, by providing insights and enabling the creation of more personalized and interactive content.
In addition, AI-powered tools and technologies will also create new job opportunities for technical communicator. For example, the development and maintenance of AI-powered tools will require people with specialized skills and knowledge, such as data scientists, machine learning engineers, and AI-powered content developers, who will work closely with technical communicator.
Overall, it’s likely that AI will have a significant impact on the field of technical communication, but it’s not likely to replace all jobs in the field. Instead, AI will change the nature of these jobs and create new opportunities. Technical communicator will need to adapt to this new environment and develop new skills to stay relevant in the field.
In a nutshell, the answer is yes and no, and the outcome is likely going to be determined by how well technical communication practitioners are willing to take initiative and adapt to a changing profession. At present, ChatGPT is akin to a “stochastic parrot,” and I am not aware of parrots who also work as technical communicators, but stranger things have happened (Bender, et. al., 2021).
References
Personal communication. February 7, 2023.
Bender, Emily, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” in FAccT ’21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, https://dl.acm.org/doi/proceedings/10.1145/3442188.
Rainey, Ken, Roy Turner, and David Dayton. “Core Competencies for Technical Communicators,” Technical Communication 52, no. 3 (March 2005): 323.
Filipp Sapienza is a usability researcher and data analyst in Colorado. He can be reached at urbann_animals@hotmail.com
