Originally Published: September 6, 2022

Writing & AI

Writing has always been a way for me to explore and think through high-impact topics. Years ago, writing came naturally; crafting an essay was easy and fun. Ideas flowed from my mind to my fingers with ease. Then, at some point, the tap seized. The well was flush, but the machinery was failing. Transferring my thoughts into cogent, coherent essays became increasingly difficult.

This failure was multi-faceted.

• My ego wanted to be seen as an exceptional thinker, so the focus of my writing shifted to fewer topics and novel ideas.
  
• I was trying to write for imagined external audiences of serious, critical thinkers who would dismiss anything imperfect.
  
• I didn’t know exactly what I wanted to write about anymore. My interests were diverging from UX and product to societal concerns and geopolitics.
  
• Inspiration wore off quickly — I would eagerly sit down to write, only to feel the motivation drain away within an hour.
  
• Often, the idea that inspired me wasn’t all that interesting after some probing.

Even right now, as I type this sentence, I can feel the writing engine in my mind sputtering, struggling to continue.

That brings me to what I really wanted to talk about — writing with the assistance of artificial intelligence (AI).

Over the last few years, a series of language models have shown increasing promise in generating human-looking writing. You prompt the model with something to nom on, and it outputs text that looks like something a professional writer put together.

For me, the biggest elephant in the room is that AI-assisted writing is going to destroy the searchable web. Google search results are already overflowing with low-quality pages, stuffed with keywords and designed to get viewers to click affiliate links. AI-assisted writing removes the only real guardrail constraining these websites, which is that it required effort to produce content. Now, you can have your own AI-powered blog in seconds, producing endless volumes of content.

Short of blacklisting websites that abuse AI content, I don’t think we can prevent the destruction of the searchable web. Enjoy the closing moments of signal-rich Google searches before the noise overtakes them. And hope that whatever rises from the ashes of Google search will be somewhat serviceable.

Next to that larger elephant is a smaller elephant, and this one may be more of a personal hang-up — I get distressed anytime I am mislead in a meaningful way. If someone uses AI to generate an article and then slaps their own name on it, are they misleading me? The answer comes down to how substantially they were assisted. If they are using AI to overcome writer’s block by prompting for ideas on where to take an essay next, that seems okay. If they are broadly copy/pasting from AI and claiming those ideas as their own, that feels like it’s not okay.

Unfortunately, we don’t get to see behind the curtains of what we read. We only have their final product to judge, and that final product may not be an accurate reflection of a writer’s skill or lack thereof.

It’s obvious that AI-assisted writing models like GPT-3 will eventually become valuable tools for authors and writers. While writing this essay, I used OpenAI’s GPT-3 model a few times to suggest what could come next. I didn’t directly use any of the output text, but what I saw did influence the direction of my thoughts. Not in a manipulative way, but as something more natural, as if I would have arrived at those thoughts whether or not the AI participated.

I’m excited for the arrival of AI-assisted writing. It’s coming whether we like it or not, so it’s better to be optimistic and thinking about how we can shape the technology to best service both writers and readers.