Humanizing AI-generated content: why it is not about mistaking

One lecturer presented a slide with a typo and then proceeded to say, “At least you know this presentation was not made with AI”.

People are deleting “—” and adding mistakes, typos and slangs to their content (presentations, cover letters, posts) to “humanize it”.

So often I’m asking myself: What makes the content “feel” machine or human-made? What is the point in creating with AI using very elaborate prompts and then proceeding to downgrade the quality of it by adding mistakes?

This post is not a tip on how to better humanize AI-content, but a reflection on how authenticity presents itself and how to keep our voice.

A few years ago, adding typos in social media content was a (questionable) way to generate engagement. Now we are adding mistakes to our content to sound human. This worries me even more in the academic field, because until very recent past a mistake was a sign that the writer hadn’t bothered putting much effort into the manuscript. Soon it can be seen as you care, because you took the time to write it yourself. I can’t imagine a worst ending for scientific writing.

This goes beyond text. Everyone exposed enough to AI-content “feels” when something is AI-made. A website, dashboard, figure, post. Everything follows the same machinery-style, so what is this thing that we “feel” but can’t name (far beyond the —)?

The first thing is the machine cadence. Every single phrase or element in a design follows the same number. It could be three sentences, three bullet points, three pictures or diagrams, or similar/same length for every phrase. Who writes every sentence with exactly 80-100 characters? Are you counting characters while organically writing to make your text sound better?

Long and convincing, still saying nothing.  The other day I was reviewing a manuscript that had a very good research question. It was very well written until I got to the discussion. It had extremely long paragraphs, quoting other valid research papers, but that entire section said nothing! All the phrases presented explanations about the general topic of research but did not connect or explain the results obtained by that specific paper. When I finished reading, it was like I was down scrolling in my phone: I saw a lot and learned nothing. While doing research to write this post I came across a concept that summarizes this feeling: the Uncanny Valley, a discomfort people feel when something artificial, like a robot, looks and acts almost human but has subtle, eerie flaws.

So how to authentically express yourself?

Here I want to focus on scientific writing, but I think you can go beyond (presentations, speeches, code, figures and diagrams...).

Read non-AI. Remember when people say that you are what you eat? Similarly, you write what you read. So read more non-AI content. It is hard nowadays to be 100% sure that something was not touched by AI, but instead of first reading what AI summarized about a topic, read the source first. Let’s say you want to discuss the effect of contamination in plant survival rate: you can use AI tools for searching literature but read the papers. Start from the abstracts to select the ones worth a full read.

Write. That sounds ridiculous but that’s it. You might think that your scientific skills are like riding a bike: you do once, you do forever. I don’t know how to ride a bike (yes, I know, I’ll learn!), but we need constant practice in writing in everything. If you code, do you remember where to put that if, that comma, that semi-colon? We once produced and corrected our texts, our codes, our speeches and we learned by doing it. Now we are leaving the knowledge on the table when we delegate all the hard process to AI.

Be specific. Put effort into bringing context, numbers, what you reasoned about the topic you are writing about. Like in the last paragraph, I thought about an example of how to read more non-AI content. When I was writing I had to think about an effect and an outcome for my example. I wrote and re-wrote many times to get an example that makes sense for me, so I hope it makes sense to you too. Refine, curate, and inject your experience into what you are writing.

Ultimately, my biggest concern is not that AI is getting better than humans... it is us - humans - getting used to the absence of humanity and fingerprints in what we consume as content and in how and what we learn as scientists.