How I use AI to improve my explanations
Pros and cons of getting artificial help.
This newsletter is about how I use AI to improve my language when explaining things.
🇸🇪 Klicka här för att läsa den svenska versionen.
This morning, when I used artificial intelligence to write and edit, I made some discoveries that spoke to its strengths and weaknesses. that I wanted to share with you.
Writing is thinking, and it's natural for your thoughts to wander when you're trying to produce something of a reasonable academic standard. In my texts, this sometimes makes the reasoning unnecessarily complicated. I often add subordinate clauses in my attempts to get all the nuances across, and the result is that the points are packed into a box that is in a box that is in a box.
To edit is to crop. You can highlight a paragraph with Grammarly and ask for a simplified version. The AI then analyses the paragraph and suggests something that is always more understandable and better, which is much easier than trying to sort out your own jumbled thoughts. It goes without saying how good this is.
It is also possible to ask for a version in a different tone, which reveals a weakness. Unfortunately, it seems to be an American tone of voice, which makes it unusable for me, who has to write in European English. These proposals also contain various upbeat and cheery versions, filled with worn-out phrases that sound like they were taken from self-help books. This is a sign that Grammarly's AI is trained on large amounts of American texts.
As a non-native English speaker, one advantage is that Grammarly analyses my sentences on the fly, highlighting misspellings and suggesting improvements. Unlike the simplification function, I can control the improvement function. I have set my language to be confident and informal, prioritising active phrasing over passive, and based on British spelling. When I finish a sentence, it highlights different words and phrases in different colours and gives suggestions on how to improve the text. (Something that also improved this text!)
One example of how an AI often consists of different, independent parts with different functions is that Grammarly then often corrects itself in my texts when it encounters paragraphs that it previously simplified.
While AI helps me become a better writer in English, there is always the risk that it makes me a less interesting writer. When you train AI on large amounts of text, the norm becomes an average of the quality of that large amount of text, which conflicts with the personal tone that I want to characterise my texts. The longer I use AI, the more cautious I become about this, which is why I no longer accept as many suggestions for rewording as I did in the beginning. This is not unique to Grammarly. All text suggestion services have average language, often characterised by clunky wording. If you've tried to generate any text related to entrepreneurship or sustainability, you've already discovered this.
The problem is likely to be transient. Soon, you'll be able to practice this kind of tool in your own texts and in your own tone of voice, and it will become even more useful. Do you agree?
