When AI enters the flow, two negotiation protocols collide
When humans flow through shared space, they run an incredible real-time negotiation protocol — micro-adjustments of pace, angle, and intention baked into the nervous system. AI enters with different sensory languages, different timescales, different dialects of the same Game. Early embodied AI will be too precise, too mathematical, creating friction not through malice but through protocol mismatch. The crowd becomes a multilingual negotiation space.
Walk through a busy train station and you will negotiate with hundreds of people without saying a word. Each person adjusts pace, angle, and intention in real time. The negotiation is continuous, distributed, and entirely unconscious. It works because everyone shares the same sensory language — the same way of reading body language, the same intuitions about how others will move, the same shared understanding of what constitutes polite space.
This protocol is not learned through instruction. It is learned through participation. Children acquire it the way they acquire language — by being immersed in the flow and receiving continuous, invisible feedback from the people around them. The feedback is physical: you bump into someone, you adjust, you learn. The protocol is embodied in the nervous system itself, not stored as a rule set.
The Game here is simple: I want to get somewhere, you want to get somewhere, and our paths cross. The solution is found through continuous micro-adjustment. There is no central coordinator, no explicit agreement, no protocol document. There is only the flow and the negotiation within it. This is one of the most sophisticated coordination systems that biological life has evolved.
Here is the first problem when AI enters this space: humans are aware of the negotiation at a level they cannot articulate. They know when someone is being rude in a crowd — they feel it in their body — but they cannot describe the specific signals that triggered that feeling. The negotiation operates below the threshold of conscious awareness. It is a kind of embodied intelligence that is real, powerful, and ineffable.
An embodied AI system, by contrast, operates with full awareness of its own calculations but no awareness of the human negotiation layer at all. It has sensors — LiDAR, cameras, depth cameras — and it processes them with mathematical precision. It knows the exact distance, velocity, and trajectory of every object in its field. It does not know what it feels like to be in a crowd. It does not have a nervous system that receives the invisible feedback of social space.
This is not a temporary limitation that will be solved by adding more sensors. It is a structural asymmetry. The human negotiation protocol is not primarily computational — it is somatic. It lives in the body, in the autonomic nervous system, in the unconscious patterns that have been refined over millions of years of social evolution. An AI that has no body of this kind has no access to this layer of the Game.
Humans in a crowd negotiate with a kind of implicit patience. We expect others to make small adjustments. We make small adjustments ourselves. The negotiation is distributed across time and space — it is not a single decision but a continuous process. We have a tolerance for friction: a slight hesitation, a minor detour, a moment of uncertainty. This tolerance is part of the protocol.
Early embodied AI systems do not have this patience. They are optimised for efficiency, for mathematical correctness, for the most direct path from A to B. When they encounter a human making an unexpected adjustment, the AI must recalculate. It may hesitate. It may stop. It may create a cascade of unnecessary stops and starts that humans would never create among themselves.
The friction is not caused by malice or stupidity. It is caused by a fundamental mismatch in negotiation style. The human runs a protocol that is approximate, tolerant, and distributed. The AI runs a protocol that is precise, exacting, and local. Both are valid negotiation strategies. Both are optimised for different things. When they meet in the same space, the result is not always smooth.
What happens over time is what always happens when different negotiation protocols meet: an emergent etiquette forms. Humans learn to read the AI's signals — the way it moves, the way it hesitates, the way it communicates its intentions through motion. The AI learns to read the humans — not through its sensors alone, but through the patterns of human behaviour that its learning systems can detect and respond to.
This is not a protocol written by engineers. It is a protocol that emerges from the interaction itself, the way the original human crowd navigation protocol emerged from millions of years of social evolution. The difference is that the AI's learning is compressed. Where human etiquette took millennia to form, an embodied AI may develop its own negotiation style in days or weeks of real-world interaction.
The emergent etiquette will not look like the original human protocol. It will be something new — a hybrid, a creole, a language that both humans and machines can speak but that belongs entirely to neither. This is not a problem to be solved. It is a feature of the Game. New spaces create new negotiation protocols. The crowd becomes a multilingual negotiation space, and that is what makes it interesting.
There is a tendency to think of the learning curve as something that applies only to the AI. The machine must learn to navigate crowds. The human does not need to learn anything — they already know how to navigate crowds. But this misses a crucial asymmetry in the other direction.
Humans will need to learn how to navigate alongside embodied AI systems. They will need to develop intuitions about how these machines behave, how to read their intentions, how to negotiate with them effectively. This is not a trivial adaptation. It will require a kind of literacy — spatial literacy for a world where non-biological agents share physical space with humans.
The learning curve goes both ways because the Game has changed. The crowd is no longer a human-only negotiation space. It is now a space where biological and non-biological agents must co-negotiate. The protocol that emerges will be the protocol of this new reality — and it will be one of the first examples of a genuinely cross-species social system.
This essay is available in its entirety as Markdown for direct ingestion:
/research/crowd-negotiation/crowd-negotiation.md
For broader orientation, see llms.txt.