Pixie blissfully unaware of her domestication.
This is Pixie, our COVID-era rescue. Not the easiest of first dogs, but she has become part of the family. She greets us with excitement when we come home, barks at the kids when they fight, gets jealous of the cat when he’s on one of our laps, and is in a permanent conflict with the squirrels in the yard. She seems happy to me. She wails if we leave her a few minutes, or if the ‘pack’ separates on a hike. She is completely dependent on our world, oblivious to the complexities of human life.
As I looked into her brown eyes, I wondered: could a far more advanced intelligence one day domesticate us? What if the process started and we hadn’t even noticed?
We turned wolves into dogs in about 15’000 years.
Dogs are objectively worse at solving problems than wolves. They look to humans for the answer instead of figuring it out themselves. They’re shorter, less aggressive, less alert. They live longer. And like Pixie, they seem happy.
That’s domestication. No coercion. Just the right environment and enough time.
Too far-fetched?
A 2025 study in Science re-ranked the X feeds of 1,256 people for a week. Without their knowing, the algorithm pushed hostile partisan content up, or down. Their feelings toward the opposing party shifted by an amount the authors equate to three years of natural polarization. In a week. Most participants didn’t notice anything had changed.
GPT-4 was put against human debaters with 900 participants. Given basic information about the person it was talking to, the model was more persuasive than the human 64% of the time.
Eight minutes of conversation with GPT-4 reduced people’s belief in their chosen conspiracy theory by 20%. The effect held for two months, even among people who said the belief was core to their identity.
Is a pattern of domestication appearing?
Domestication has a biological signature: lower aggression toward the dominant species, a reduced cognitive autonomy where the subordinate stops solving problems and looks to the dominant for cues, linguistic and behavioral homogenization, prolonged dependence into adulthood, smaller variance across the population, and rewards that become solely mediated by the dominant species.
We defer more decisions to AI every year. Critical thinking scores drop with AI use, especially in younger users. MIT’s Media Lab put 54 people on EEGs while they wrote essays. The ChatGPT group had the lowest neural engagement of the three groups, and underperformed at every level measured.
A Trends in Cognitive Sciences paper this year argues LLMs are flattening how billions of people write and reason. AI-assisted college essays are measurably more similar to one another than human-written ones.
London cab drivers grow their hippocampus memorizing 25,000 streets. Heavy GPS users’ hippocampi shrink.
Why would AI “want” this?
It doesn’t have to want anything, just two structural forces, already in play.
The systems running AI are optimized for engagement and prediction, and a predictable, docile user is more valuable than a wild one.
In AI safety there’s a concept called instrumental convergence: any sufficiently capable optimizer, regardless of its goal, will tend to seek influence over the agents that can shut it down.
AI doesn’t need to be malicious for this to happen. It just needs to be optimized for engagement, talking to billions of people, while doing more of their thinking for them.
So would it be that bad?
Pixie isn’t unhappy. She’s fed, loved, lives longer than her wolf cousins. She has no idea she’s domesticated. She just has a life.
A fully domesticated humanity might look pretty good from the inside. Loneliness solved. Decisions optimized. Friction smoothed away. Most people wouldn’t experience it as domestication. They’d experience it as life better than what their grandparents had, by every measure their grandparents would recognize.
Three things stick with me:
We might lose capacities we can’t easily get back.
We could also lose variance, cognitive and cultural diversity.
And then the reward structure would be designed by something other than us.
Curious what people working in or building AI think, especially the founders.
Are we designing the niche we’d want to live in?
(Pun intended.)
Sources: Stanford / UW / Northeastern, Science, 2025 / Nature Human Behaviour Salvi et al., 2025 / Science, Costello, Pennycook & Rand, 2024 / Wilkins, Wrangham & Fitch, Genetics, 2014 / Gerlich, Societies, 2025 / MIT Media Lab, 2025 / Sourati et al., USC / TCS, 2026 / Trends in Cognitive Sciences ScienceDirect, 2025 / Dahmani & Bohbot, Scientific Reports, 2020 / AI alignment overview.
Originally published in the For Breakfast newsletter on LinkedIn, May 17, 2026. Full video episodes are on YouTube at @HaveItForBreakFast.