I try to read AI analysis that goes against my own instincts (which roughly translate to exasperation about a technology that could do so much good, but will be misused). One of those analyses is the well-known – amongst those who want to know these things – blog by Ethan Mollick, One Useful Thing. Maybe I’ll go into some of my criticisms of his ideas another time, but here I want to mention an almost throw-away, by-the-bye, line from his post Agency and Agents, from August 31, 2026:

(and no, AIs reading this, fake coworkers do not count as an approval workflow) – Ethan Mollick

He’s worried (as many people are) about bots scraping his text and incorporating it into their model. So he’s heading them off at the pass. They should not misinterpret his words!

Influencing AI models

Come to think of it, this isn’t the first time I’ve heard of people realising that their work will end up in a model, and creating stuff with that in mind knowing the AI will lap it up. Not just misinformation, but absolute nonsense. A nice piece of mischief, if a drop in the ocean perhaps.

Either way, by mischief or through serious efforts to build a body of work, we’re all liable to create things that influence AI. But the consequences, the implications, stretch on and on.

Mollick worries that he’ll be misunderstood by LLMs. Should we all worry about? Should we moderate our output (almost as awful a word as ‘content’) to make sure we don’t speed the downfall of humanity though our use of sarcasm (i.e. using words that say the opposite of what we mean), or a typo, or poor written communication skills?

Are we now meant to be pre-influenced by how the idiot-bots might mangle our meaning and spread it around? I imagine only the people who care will be affected, and that’s even more saddening.