Are Horses Smarter Than We Thought?

The study conducted by Louise Evans, Heather Cameron-Whytock, and Carrie Ijichi:

When we think of intelligence in animals, we often picture apes or dolphins—but new research on horses is changing that. A study conducted in 2024 on model-based learning in horses suggests they’re far more conscious and thoughtful than we ever thought.​

This study looked at how horses make decisions and learn from consequences, comparing two types of learning: model-free and model-based:

Model-free learning is habit formation. It’s trial and error—the repetition of what’s worked before without much thought.

Model-based learning, on the other hand, involves foresight. It’s when one uses an internal “map” of cause and effect. This involves predicting outcomes and choosing actions based on expected consequences.​ In other words, model-based learning is usually a measure of higher-level intelligence in animals.

To test this, researchers used a stop-signal task. Horses were first trained with positive reinforcement—rewarded for touching a target. Then a new twist was added: a bright LED light told them to stop touching the target. If the horse did touch the target, they just didn’t receive a treat and the session continued.​

After three full sessions using only positive food rewards, none of the horses learned to obey the stop signal. They just kept going for the reward, ignoring the light cue completely.​ The horses’ lack of understanding suggested model-free learning. The horses simply learned a pattern: they touched the target and – most of the time – they got a treat.

Then, the researchers introduced something new: a small cost for mistakes. When horses touched the target during the stop light, they got a 10-second “time-out”. No treats, no attention, and no progress. Their accuracy immediately increased. This wasn’t a gradual improvement. It showed that the horses already understood the meaning of the light but were just now adapting strategically, understanding that their actions now came with consequences.​

Blue: Expected response if horses never learned, Green: Potential response if horses learned from introduction of consequence, Orange: Actual response

That immediate improvement suggested model-based learning—the horses weren’t simply reacting and ignoring the light. They were evaluating, predicting, and adjusting their behavior based on feedback. It’s not that the horses didn’t understand the meaning of the light when positive reinforcement was used, they just didn’t care. When they did not experience a consequence, they weighed the costs and benefits to make a decision – it was not worth responding to the light when they received a reward majority of the time. But, when they had a consequence, they re-weighed their options – it became objectively “bad” to touch the target when the light is on. In other words, the horses formed abstract expectations, made cost-benefit calculations, and even exerted self-control when properly motivated.​

My Take – Why This Matters in Training:

Now, what does this mean in the context of horse training? This study doesn’t just disagree with purely reward-based training; it also shows the importance of consequence in shaping a horse’s understanding. It’s not about force or fear. It’s about clarity and communication. How are we supposed to tell our horses what to do if we’re not speaking a language they can understand?

Over the last decade, positive reinforcement has become almost gospel in horse training circles. And don't get me wrong—positive reinforcement is important and effective. The issue isn't the use of rewards in training, it's the lack of consequences and overall boundary-setting in horse training. Purely positive reinforcement has its place—it builds motivation and trust. But without boundaries, it can leave a horse uncertain about what behaviors are acceptable.

A horse that never experiences pressure or correction isn’t really safe. It might appear compliant until something triggers a reaction it doesn’t understand how to control. If you spend any amount of time on the “horsey side” of the internet, you've probably come across videos of owners “allowing their horses to make a choice” or “playing with their horses in the pastures.” The issue with these videos is that horses don’t understand human boundaries, they understand horse boundaries. Horse “play” is much more dangerous than it can seem. It is important for you, as an owner and/or trainer, to teach your horse about boundaries and, as supported by the study, this can only be done effectively if consequences are incorporated into training.

You, of course, cannot train a horse for every situation, but you can teach them how to self-regulate no matter the situation. To do this, though, they need a real-world equivalent of the “stop cue.”

That might sound abstract—how can you teach a horse to handle a situation without putting them directly in it? But we actually do this all the time without realizing it. Think about one of the very first exercises trainers do when starting a young horse: yielding the hindquarters. This isn’t something you ask for under saddle, and it’s not a maneuver used in daily riding. Yet it’s one of the most fundamental skills in groundwork—and for good reason.

When a horse rears or bucks, it’s using its hind end to generate power. If that hind end disengages, the horse loses both balance and leverage, and the behavior de-escalates. So, teaching a horse to yield its hindquarters isn’t about domination—it’s about giving the horse and handler a way to safely reset when tension spikes. Ideally, you never have to use it in a crisis, but it’s there as a calm, learned reflex for when emotions run high.

To teach a horse to yield its hind-end, you cannot rely solely on rewarding every time they move their legs. You must use physical corrections to show the horse what you’re asking. Imagine if you’re in math class and your teacher praises you and gives you $5 when you answer a question correctly. And if you answer incorrectly, the teacher doesn’t tell you that you got it wrong and you will still receive the $5 every time you get a question right. Because of the way human brains use model-based learning, you will most likely rush through challenging problems because you still get praised when you get questions wrong and you will also still receive the $5 when you get the easier problem right.

This is the same thing for a horse, the horse is willing to risk not getting a treat for not doing something when it gets the “stop cue” (in this example, a vocal cue to dis-engage the hind-end) knowing it will not be put in the vulnerable position of yielding the hind-end, it will not have a consequence for ignoring the cue, and it will still get a treat for moving its hind legs in a different, more comfortable direction. Continuing with the teacher analogy, if the teacher takes away $1 every time you get a question wrong, you will prioritize getting questions right so you don’t lose money – you actually learn. The same thing goes for the horse, if you introduce a consequence (in this example, it can be a whip tapping on the hind legs when the horse moves in the wrong way), the horse will weigh the consequences and benefits and be more likely to yield its hind legs correctly and learn to relax.

Let’s be clear: this isn’t about punishment in a harsh or abusive way. It’s about structure, boundaries, and feedback. When a horse rears, bolts, or crowds your space, redirecting them—like yielding their hindquarters—isn’t cruelty. It’s communication, if they can’t understand you, they will be less likely to relax.

The study proves that horses are capable of model-based reasoning—they can consider outcomes before they act. But that only works if they experience both sides of the equation: reward and repercussion. Purely positive reinforcement training (+R) didn’t teach the stop response. It wasn’t until negative punishment—the removal of reward—was introduced that the horses connected cause and effect.​

That’s the same principle behind effective pressure-and-release training. Pressure signals a boundary or a request; release reinforces the correct choice. Without pressure, the release loses meaning. The horse never develops that mental model where actions have consequences.

Think of it as helping the horse engage its frontal lobe, not just relying on instincts. Structure builds confidence because the horse knows exactly what to expect and how to succeed. Boundaries aren’t rejection—they're clarity.

If you're training with rewards only and wondering why your horse isn't consistent, it may be because your horse is using model-based learning. The moment you add a small, fair consequence—not a punishment, but a consequence—you'll likely see a dramatic shift, just like those researchers did.

That's what really hit me about this study. It shows that horses are smarter than we give them credit for. They need both sides of the equation to make informed decisions about their behavior.

And that’s what I personally believe marks the difference between just any equestrian, and a talented horse trainer – the ability to train horses in the horses’ language. This is what communication between horse and rider is, not necessarily teaching the horse our language, but instead teaching them a skill in their own language.

I highly recommend reading the article: https://www.sciencedirect.com/science/article/pii/S0168159124001874

Similar article: https://journals.plos.org/plosone/articleid=10.1371/journal.pone.0258672#sec007