I recently saw a video from the Thinking Machines folk where the speaker was being corrected in real time as she spoke. “It’s pronounced ah-sai-yee” it quipped as she struggled with the word for the purported superfood berry she had for breakfast. In a way, it was cool to see an AI actively listen and steer a human to saying the “right thing”. But it got me thinking; is being right the point of everyday communication?
Conversations rarely deadlock over a disputed fact. Sure it’s easy enough to verify things with the computer in our pockets if it must be done, but I’ve often found this to break the flow and cause unnecessary distress. Recently, I found myself in a conversation about hip-hop and classical music and wrongly noted Nas’ I Can as sampling from Mozart (it samples from Beethoven’s Für Elise). While I was wrong, and I was eventually corrected, nobody got hung up on the specifics of the mistake. The conversation just moved on toward the bigger idea of hip-hop’s genre crossing sample culture. Being wrong didn’t automatically discount the point I was making. It was quite literally beside the point.
There are of course situations when being wrong isn’t as trivial. And there’s a discomfort that builds when we’re unsure we’re even right. As humans, this often comes with a natural pause when we trail off mid-sentence or contemplate something we just said. In the case of a mispronunciation, a frequent tick is to question ourselves out loud or prompt the other person for clarification. AI removes this completely and offers premature certainty before you even finish a thought. We are spared the effort of trying to self-correct and in so doing, the opportunity for mutual uncertainty that would have briefly connected two people disappears.
In a letter to his brothers, the English poet John Keats coined a term to describe this idea: Negative Capability. He argued for the virtue of being uncertain “without any irritable reaching after fact and reason”. The context specifically focused on Shakespeare’s imagination, and his ability to give voice to contradiction without rushing to resolve it. The same restraint can be made outside art. For instance, in the moment we let a mistake sit instead of correcting it away.
The AI in that video wasn’t wrong to correct the pronunciation. But it short-circuited the natural back and forth that often happens when people fumble in conversation. It treated the conversation like a checkpoint, something that had to be verified for correctness before moving on, which is antithetical to how humans communicate.
There’s an undeniable kick we get from being right. However it’s helpful to distinguish the kind of certainty that shuts a conversation down and the kind that allows a conversation to linger in the vulnerability of not knowing. Turns out the good parts of a conversation isn’t in the assured certainty of being factually right. It’s often found in the fumbling, guessing and reaching even when you end up being well and truly wrong.