This newsletter is a hand-written reflection on an in-person conversation. You can watch the video or listen to the audio here.
A few weeks ago I was exchanging messages with a long-time friend, and teed him up for a funny joke. He responded with a witty quip and a smiley emoji, then quickly confessed that "Chat helped me write that one."
Why did that make the joke feel different?
It was still funny, and my friend is a funny guy, but somehow knowing that the humor was outsourced to an LLM, even in part, changed the way I viewed it.
AI is changing our perception of writing
Another friend of mine told me that he was reading a management book written by Peter Drucker decades ago, and he was shocked at how shocked he was at the number of em dashes, especially because he had read the book before.
Em dashes have been used in writing for centuries, and in a few short years, AI’s prolific inclusion of them in generated copy makes us suspicious.
Here’s the full text of the post:
AI slop has fundamentally changed how I experience reading.
I now notice the tells everywhere. The short, punchy sentences. The relentless groups of three. The em dashes—the unnecessary little asides.
And once you see it, you can’t unsee it.
The worst part? My brain now flags perfectly innocent human writing as AI slop too. AI hasn’t just made me enjoy bad writing less. It’s made me suspicious of all writing.
That’s a real loss.
If you picked up on the subtle irony in the tweet, you’ve likely already questioned how the use of AI in writing will impact the future. If you didn’t, the joke is that the author clearly used AI to generate the post to further reinforce the point he is making.
If we do become increasingly suspicious of all writing, that is, indeed, a loss.
The spectrum of trust
We often frame hard questions as having binary answers because it is much easier to understand our world when we can paint the human experience in black and white, but that is rarely how the practical work of day-to-day living plays out.
The reality is that AI is here, and it will be used to generate words, and we will read those words.
The question we need to ask is where we should read those words, and where we shouldn’t.
A simple example is a user manual for a product. Accuracy and clarity are paramount, and the words are a means to an end (assembling using the product). LLMs can be used very effectively to generate and fact-check objective information.
But what about a rationale for economic policy that impacts people at different levels of society in different ways? We become much more uncomfortable with the idea of non-deterministic machines communicate fundamentally human ideas, which have fundamentally human consequences.
And, of course, we all must grapple with the decision of whether or not to let AI help us write our jokes.
Empathy as the end game
Questions abound as we navigate this uncharted territory, but the most helpful ones are simple:
Am I using AI in a way that increases my capacity for human communication, or diminishes it?
Am I becoming better at wrestling to make my words clear, or am I outsourcing that to the path of least resistance?
How are the tools I use shaping me?
Working out the answers is messy, but you’ll know if your communication reveals more or less empathy over time.
Watch or listen
You can watch or listen to the conversation that inspired this newsletter. Watch the video below, or check out the audio on Spotify or Apple Podcasts.