As AI content grows, what should we expect?
When we can no longer tell AI-generated content apart.
Human or AI-written? As AI gets better and its use gets more common for written content, here's what it means for the rest of us.
Generative AI tools continue to improve in leaps and bounds. As it gets better, the day will come when we simply cannot tell them apart.
AI text watermarks
To address this problem, Google recently announced that it has open-sourced its SynthID text watermarking technology.
The technology works by:
- Altering text output to be slightly less probable.
- The modified text it outputs looks the same,
- But is detectable via specialised software.
LLM makers can incorporate this technology into their AI tools, making it possible to determine if a given content is generated via their AI models.
However, I personally deem it unlikely that many will incorporate this technology. I'm also not convinced that watermarking or AI-detection tools will work.
Easy to foil AI-detection tools
Today's AI-detection tools can be circumvented in many ways. For instance, one can:
- Run them through another LLM.
- Translate to another language; translate back.
- Substantially change AI-text to evade detection.
Moreover, many current AI-detection tools are prone to false positives. This means they falsely appraise samples as AI-generated when they are not.
This can be devastating to some, especially students who do not use AI tools but get identified wrongly:
- Neurodivergent might write in a generic manner.
- Nonnative speakers are more prone to be flagged.
Further complicating the situation are specialised rewriting services designed to make any written piece less prone to be flagged by AI-detection tools.
AI avoidance
For now, it is still quite possible to detect AI-written text. And I'm seeing AI-generated content everywhere:
- LinkedIn posts.
- Comments.
I don't know about you, any content I discern as AI-generated immediately drops multiple rungs in credibility.
Usually, I'll end up skimming or skip it entirely.
How to not write like AI
Regardless of your stance about AI use, this means it is important not to write like AI. In my opinion, you need to do the following:
a. Use AI tools more
You won't know how AI writes unless you use it heavily. As you notice the keywords, style, and tone, you can start writing differently.
b. Never use AI wholesale
Use AI for inspiration, rewrite small sections, or edit, but avoid generating and using large chunks of text - people who do a lot of (a) can tell.
Question: What do you do when you come across AI-generated content?