
I have had an increasing sense of something absent in the reading and experiments I have been doing on using AI generatively.
Coherence
1550s, "harmonious;" 1570s, "sticking together," also "connected, consistent" (of speech, thought, etc.),
Etymyonline
I was using perplexity.ai to explore different types and structures of conversation (you can see the results here) as part of a project for a client using the Better Conversations Foundation framework.
The results were impressive, and they pointed me to what was missing. It lacks coherence. The parts do not fit together well. They make for a “clunky” conversation unless thoughtfully modified.
The overall scaffolding of the results is good, but the references it uses to bring them together are less so. It is an assembly of, I suspect, the easiest sources to find, and it reminds me of something built of cheap parts—looking superficially convincing but unlikely to stand up to rigorous use.
That is not to deny its usefulness—it is a good scaffolding and gives me the opportunity to build something inside it from sources I choose rather than the one it provides.
It raises important questions. I could have asked Chat GPT to turn this into a training course on conversations, and it would do a good job of bringing average material together, and I’m sure there are those doing just that out there to fill gaps in presentations and training proposals.
In the end, however, just as we become the average of the five people we most associate with, something similar is true of the material we use to shape our thinking.
Craft matters. And conversation is a craft.
Reminder
Our first “open” session for OTW paid subscribers will take place tomorrow, Tuesday, June 4th, on Zoom. We will discuss the last few posts on observation and orientation and what we are noticing in the changing world of work.
Link below
Keep reading with a 7-day free trial
Subscribe to Outside the Walls to keep reading this post and get 7 days of free access to the full post archives.