A behavior-environment information loop drives sensory navigation

Imagine you are trying to find the best pizza in a new city. You have two ways to do it. The first way is passive: you stand still, wait for a smell to drift by, and if it smells good, you walk toward it. If the smell stops, you stop. This is like a robot that only reacts to what it senses right now. The second way is active: you realize that standing still won't help, so you start walking in a zig-zag pattern. By moving, you change what you smell next. You are actively shaping your own experience to gather better information. This is how most living things, from tiny bacteria to humans, actually navigate. They don't just wait for the world to tell them where to go; they move to find out.

Scientists have long studied how animals use their senses to make decisions, but they mostly looked at the one-way street: how a smell turns into a turn. They rarely asked the other question: how does the act of turning change the smell you will encounter next? This paper steps into that gap. It uses a branch of science called information theory, which is basically the math of how much "surprise" or "newness" is in a message. The authors want to measure the two-way conversation between an animal's actions and the world around it. They ask: Is the animal just reacting to the world, or is it actively poking the world to get a reaction? Understanding this loop is crucial because it explains how creatures find food, mates, or safety in a chaotic, changing world, and it might even help us build better robots that can navigate without getting lost.

---

The Great Information Loop: How Moving Changes What You Know

Think of navigating the world like playing a game of "Hot and Cold" with a friend who is hiding a treasure. In the old way of thinking, scientists believed the player was just a detective listening for clues. If the friend said "warmer," the player moved closer. If they said "colder," the player moved away. This is a reactive strategy: the environment sends a signal, and the player reacts.

But this paper argues that the real magic happens when the player realizes they can make the friend talk. If the player starts running in circles, they might accidentally bump into the treasure or trigger a new clue. This is an active strategy: the player's actions change the environment, which then sends back new, better information. The authors, a team from Yale University, call this a "behavior-environment information loop." They wanted to measure exactly how much of the navigation comes from reacting versus how much comes from actively shaping the experience.

To do this, they invented a new math tool based on something called transfer entropy. Imagine you are trying to guess the next move of a video game character. If you only look at the game screen (the sensory input), you can make a decent guess. But if you also know what the character just did (the behavioral output), your guess gets even better. Transfer entropy measures that "better guess" factor. It splits the total information flow into two directions:

1. Reactive Flow ($\dot{T}_{s \to a}$): How much the current smell or sight tells the animal what to do next.
2. Active Flow ($\dot{T}_{a \to s}$): How much the animal's current move changes what it will smell or see next.

The Secret Formula for Success

The team started by building a super-simple computer model of a bacterium. They gave it just two choices: run straight or tumble (spin around) to change direction. They also gave it two sensory states: smelling "up" the gradient (toward food) or "down" the gradient (away from food).

When they crunched the numbers, they found a surprising pattern. The bacterium's success at finding food didn't depend on just one of these flows. It depended on the geometric mean of both. Think of it like a bicycle: if you have a great engine (reactive flow) but no steering wheel (active flow), you just drive in circles. If you have a great steering wheel but no engine, you go nowhere. You need both working together.

They created a single number, which they call $\Phi$ (Phi), to represent this perfect balance. When they tested this number against real-world data, it worked like a charm. They looked at:

• Bacteria swimming toward food.
• Worms crawling through chemical landscapes.
• Flies tracking invisible, swirling odor plumes in the wind.
• AI agents (computer programs) learning to navigate a virtual world.

In every single case, the higher the value of $\Phi$, the better the navigation performance. The math predicted the success rate with incredible accuracy (a correlation of 0.95, which is almost perfect in science). This suggests that whether you are a tiny microbe or a smart robot, the secret to finding your way is maintaining a tight feedback loop between what you sense and how you move.

One Size Does Not Fit All

Here is where it gets really interesting. The paper shows that there isn't just one way to be a good navigator. Because the success depends on the balance of the two flows, different creatures can achieve the same goal using completely different strategies.

Imagine two bacteria finding food at the exact same speed.

• Bacterium A is a "Reactive Master." It smells the food, gets excited, and runs straight. If it smells less, it tumbles immediately to correct its course. It relies heavily on reacting to the signal.
• Bacterium B is an "Active Explorer." It doesn't just react; it keeps moving in a way that guarantees it will keep smelling the food, even if the signal is weak. It shapes its own path to keep the information flowing.

Both get the same result, but their internal "personality" is totally different. The authors found this "degeneracy" (where different paths lead to the same destination) in bacteria, worms, and even in the AI agents they studied. This means that just watching how fast an animal moves isn't enough to understand how it thinks. You have to look at the information loop to see if it's a reactive follower or an active shaper.

The "Goldilocks" Zone of Movement

The paper also discovered a twist when things get too predictable. In a very stable, calm environment, you might think that being super-deterministic (always doing the exact same thing when you smell food) would be best. But the math suggests otherwise.

If an animal becomes too predictable in its movements, it stops generating new information. It's like a robot that walks in a straight line forever; eventually, it stops learning anything new about the world. The authors found that the best performance often happens at an intermediate level of active flow. The animal needs to move enough to keep the world interesting and informative, but not so much that it loses its way.

They tested this with a computer game called a "two-armed bandit" (a slot machine with two levers). As the AI learned to play, it didn't just maximize its rewards; it adjusted its information flow. When the game was easy, the AI became very reactive. When the game was hard, it needed to be more active to explore. But if it got too confident and stopped exploring, its performance actually dropped. The best strategy was a "Goldilocks" balance: just enough active movement to stay curious, but enough reactive control to stay on track.

Why This Matters

This research changes how we look at navigation. It tells us that movement isn't just a way to get from point A to point B; movement is a way to ask questions of the environment.

The authors didn't just simulate this; they measured it in real bacteria, worms, and flies, and the pattern held up. They didn't find a single "best" way to navigate. Instead, they found a universal rule: successful navigation requires a dance between reacting to the world and poking the world to see what happens.

So, the next time you see a fly zig-zagging through the air or a dog sniffing the ground, remember: they aren't just following their noses. They are actively conducting a conversation with the air, the wind, and the ground, using their own movements to keep the conversation going. And according to this paper, that conversation is the key to finding their way home.