Living in San Francisco, I don’t have enough fingers to count the number of people who have told me they wanted to blog more.

Regular blogging is another item in the pile of should-dos with healthy diet, consistent exercise, and limited social media use.

How can I fight the force that stops me from blogging more? After all, I have a list of topics that might make good posts. There are people who, by some power, push through.

How does someone who blogs regularly approach public writing?

This was a question I posed to Paul Christiano during a talk of his on AI Alignment that I moderated last Thursday. Paul is one of the most prolific writers in the AI Safety community: he blogs on Medium at ai-alignment.com, posts regularly on the Agent Foundations forums, and coauthors a handful of academic papers per year.

Paul’s answer on his writing philosophy was that he writes, first and foremost, in order to help himself think more clearly. His motive is not to build an audience. Paul admits that some of his writing is difficult for interested people to understand, and while he’d like to improve there, general comprehensibility is a secondary goal. He communicates that the ideas he is publishing aren’t solidified— the epistemic uncertainty is high.

My immediate internal reaction was something like, “But what if readers think your writing is bad?” and then, looking into the eyes of this well-respected researcher, a surprising realization hit.

It’s okay to publish incomplete, even incoherent ideas if you communicate you’re still learning and thinking things through. No one expects your writing to have all the answers all the time.

The rest of the Q&A focused on technical details in Paul’s research agenda, perhaps a topic for another blog post. I’ll link the video recording here when it’s done editing.

In the days following I pondered, ‘why was the idea of communicating uncertainty so unintuitive to me?’ The feeling that surfaced is that part of me had never unlearned the proverbial Writing Teacher — a cold, vigilant grader who looked for robust, persuasive essays and punished ambiguities. Under that frame it makes sense to not blog, as after all, what then is a blog audience but an unbounded set of such predators.

The force that blocks me from writing more is a fear of judgement, but it turns out my notion of my audience was outdated. In the adult world, no one who matters negatively judges thinkers who reveal their ideation in its early stages. The reality that Paul helped me see is that self-fulfilling, humble technical writing is safe.

I think ;)

AI Alignment