Can my dog predict the future ?

When I throw a stick, my dog starts moving before the stick even leaves my hand. How is that possible ? Does my dog know the future ? Of course, beyond the joke and the clickbaity title, this should not seem like magic. We usually admit that animals can anticipate by predicting the future, at least in some weak sense. I suppose we think of it like this: my dog has observed me throwing toys for years, so he has a mental model of where the toy will go depending on my movements. When I start moving, his brain uses the present information, puts it in the model, makes a few computations, predicts the trajectory of the stick, and starts moving according to this prediction. And sure, this is far from being Nostradamus. I can profit from this behaviour to easily fool my dog. I simply don’t release the stick, and my dog starts moving to the wrong place.

We have a tendency to say that dogs predict the trajectory of their toy, even though they can make bad predictions from time to time. And this seems necessary, even obvious, because dogs need to be at a certain place in the future to be able to catch the toy. But this future-position-of-the-toy is not available in the present, at the time when the dog needs to start moving. Prediction is necessary, then, because dogs can only use the information available in the present, and obviously this is not sufficient for knowing the future. It is a typical example of what Andy Clark and Josefa Toribio^[1] have called “representation-hungry” problems: it seems like it can't be solved without using mental representations. But because there are regularities in the physics of toy-throwing, it is possible to have a model that relates present information to future toy positions.

Anticipation without prediction #

But is it that obvious ? Let’s change the perspective a bit. Instead of doing an overhead throw, let’s swing the stick from side to side. My dog still makes the same kind of anticipation: when I swing left, he moves left. When I swing right, he moves to the right. But this allows to recognize that maybe he’s not making a prediction at all. He is just coupled to the stick, always following it when I swing it. If I don’t release the stick, he ends up being wrong, but not because of his lack of predictive skills, simply because the information to which he adjusts his movement has changed. There is no need here for a complex mental model simulating toy-ballistics. In fact, no knowledge about the future is required, just the visual information available in the present.

Surely this is an exception, right ? If I really throw something, my dog needs to be able to predict where it’s going to land. Well, not according to research. In 2004, researchers^[2] attached cameras to the heads of two dogs (Lilly and Romeo) in order to observe their strategies while catching Frisbees. If they were using a predictive strategy with a mental model, they should run in a straight line to the landing point, and then wait if they’re early enough. In fact, they use a strategy focusing on constantly adjusting their movement to the flight of the Frisbee. Basically they try to maintain an optical image of the Frisbee so that it seems to be flying in a straight line with constant speed. This is called a prospective strategy, rather than a predictive one. The details are a bit complex to put into words, so go read the paper if you want to know more. I’ll just try to give you a sense of how that can work. Imagine the parabolic trajectory of a ball moving towards a dog. From his point of view, it seems to go up, slow down, then speed back down. If the dog moves in a way that cancels this perceived acceleration, he’ll be at the right place at the right time. Add other tricks analogous to this one, depending on the situation, and dogs are capable of catching objects without using prediction.

At a first glance, maybe we attribute predictive capacities to dogs because we’re guilty of anthropomorphism. We attribute our “higher” cognitive skills and behaviours to animals. This probably happens more often than not^[3]. But in fact we are not that superior, and in this case we would be as wrong if we attributed prediction to humans. There is a large body of research, centred on the “outfielder problem”^[4], that shows that we use the same strategies as Lilly and Romeo in order to catch objects. This is something we should celebrate. By “using the world as its own model”^[5], we can avoid using a superfluous, imperfect, time- and energy-consuming internal model. The couplings solutions we use have a huge advantage, because they’re always up to date, they can be applied to balls as well as less predictable objects like Frisbees, and work within a large range of conditions, taking wind or friction into account. My dog can’t predict the future, but he doesn’t need to.

Keeping them coupled #

Dogs and humans can anticipate the future without using prediction, by coupling their movements to available information in the environment. This has strong implications for the way we teach sports. When teaching skills that require some form of anticipation, the goal should not be to create internal models that allow predicting outcomes, e.g. the trajectory of the ball. What we want is rather to create functional couplings between the body and the ball. At some point, the teaching methods used for these goals will diverge: the second option requires the athlete to move and adapt in real time. No decontextualized and repetitive drill or watching of an opponent on a screen will suffice. Once again, we need to keep perception and action together.