There is a lot of excitement about ChatGPT these days. As a managing director and employer, I have of course also asked myself the question: How are intelligent systems changing our working environment? Time for a personal assessment.
ChatGPT: The internet phenomenon
4 million users after one week, over 100 million after only two months.
The whole world is excited – and all because of an artificial intelligence that can write short stories, term papers, blog articles and screenplays in a matter of seconds.
‘ChatGPT: Write a Java error message as a limerick.’
‘What was before the Big Bang?’
‘Develop a marketing strategy for a tech startup.’
The AI finds an answer to (almost) every question. It does this unerringly and – at least at first glance – in a linguistically flawless manner. On the internet, the release of ChatGPT has sparked a new fundamental discussion about how AI systems will change our professional lives. Depending on one's point of view, reactions range from euphoria to a sense of doom.
What is remarkable is that, unlike previous technological advances, the coming upheavals will primarily affect highly qualified and intellectual workers.
What impact will AI have on their professions? Or, to put it more specifically from the perspective of my industry: Will AI systems soon make the jobs of technical writers redundant? If I had to answer this question in just one sentence, it would probably start with the words ‘No, but...’ But let us take it one step at a time.
Why ChatGPT is not (yet) suitable for special tasks
My initial conclusion after the first rounds of testing the talkative chatbot:
The AI's answers are often amazingly good. But it also often remains vague and some of the answers are peppered with false assertions. So, apart from the fact that you should not rely on ChatGPT in terms of content, how can the tool be used in the day-to-day work of a technical writer?
To put it briefly: The text AI is in its current form not yet suitable for specialised tasks. ChatGPT has been trained with a huge database of articles, wikis and other internet texts. However, the system knows nothing about subject-specific or internal organisational data sets and can therefore only access knowledge that is freely available on the internet.
So when it comes to specific requests such as ‘Write a guide to product XY’, a general AI such as ChatGPT quickly reaches its limits. It simply lacks the necessary data to solve such specialised tasks – which makes perfect sense.
A smart assistant for routine tasks
However, the tool can already be used in daily work across numerous industries. For example, the AI is good at summarising basic information on a topic in an organised and comprehensible way. ChatGPT is particularly useful for brainstorming ideas and as a source of inspiration for blog posts, for example. The chatbot can also be helpful for routine writing tasks, such as composing and translating emails or cover letters.
As mentioned earlier, although the AI texts are linguistically convincing, the chatbot occasionally fibs and mixes facts with fiction. The tricky thing is that you cannot distinguish between wrong and correct information at first glance. You have to be well-versed in the subject matter to be able to correctly classify the answer and not fall for the AI's line.
And the chatbot has some other weaknesses: Stylistically, the AI texts appear sometimes quite stilted and some phrases are repeated regularly. Also, the AI is not capable of properly penetrating a topic argumentatively. This results in the main aspect of a text receiving just as much linguistic attention as the much less important secondary argument.
AI-generated content should therefore be treated with caution and should not be passed on unchecked, even in internal communication.
What does the future hold?
ChatGPT as a smart assistant for routine tasks? To be honest, that does not sound like the big AI revolution yet. Things will only become really exciting when AI tools are trained on a large scale with companies' specific data sets. AI systems will take on more and more tasks, especially in areas where standard processes are in place and a comprehensive database is available. This development will, of course, also influence and change the work of technical documentation.
The tools will be an important aid in searching through and organising large volumes of content. I also see potential in the writing process itself – for example, in pre-phrasing and compiling standard content.
But the key question is: What can AI do when it comes to creating something new and specific? Will it be able to describe a machine from A to Z that has not yet been documented? Can it develop a concept topic for the online help of a new software?
I am convinced that when it comes to creating new ideas and individual solutions, people will stay ahead – even in the long term.
Correlation instead of causation: The weaknesses of AI
This is primarily because, although AI can analyse huge amounts of information it does not create any new knowledge. Put simply, it works with statistical probabilities without having any insight into actual connections. Solving complex problems requires more than reproducing what is already known. Even with ever-larger amounts of data and better testing mechanisms, this lack of human experience will probably not be completely eliminated.
Technical editors in particular know what this means. To get to the really important information, it is not enough to just evaluate the existing design and development data. Whether it is the right way to change the oil in a machine or hints for operating software, a lot of what makes a good and comprehensible manual has to be researched in a personal conversation with the designer or developer of a product. On this aspect, I do not see artificial intelligence on par with humans for a long time to come.
Not least, the comprehensibility of AI-generated content is a major problem. The results are based primarily on statistical clusters. The AI provides plausible-sounding and coherent answers without being able to assess what is right or wrong. It lacks the critical thinking skills to question its own statements.
Moreover, the decision-making of new deep learning algorithms is often no longer technically comprehensible so that it remains unclear how exactly the AI arrived at a particular result. Especially in such a sensitive area as technical documentation, humans will always have to check and adjust the machine's results.
My conclusion: Promising opportunities but no job killer
Even when the furore surrounding ChatGPT dies down, I am certain that intelligent systems will soon be providing us with valuable support in more and more areas of our professional lives. And yet I do not believe that AI will automatically make entire industries and jobs redundant.
On the contrary, if AI systems take over more and more routine tasks, people can devote themselves more productively and efficiently to specialised tasks. And it is precisely this type of task that will increase rather than decrease in the future. In fact, I even see significant potential for technical writing here: As highly individualised work becomes more important, the demand for qualified information brokers will increase. By optimally preparing and making knowledge accessible, they enable people to take on tasks with a higher degree of specialisation.
This also applies to the interaction between humans and AI. The ability to ask the system the right questions is becoming increasingly important and will be the soft skill of the future.
Separating the important from the unimportant, focusing on the target group, asking critical questions – these are all skills that already characterise technical writers today. I am sure that they will continue to be key to how companies transfer their knowledge in the future.
