It’s long been known that mice can be trained to perform simple tasks in exchange for a reward. Bribe a hungry mouse with a morsel of food or a thirsty mouse with a drop of water and you can encourage it to navigate a maze or click a particular button. But sometimes, mice don’t act as expected, failing to complete the task at hand. Often, researchers have dismissed these actions as simple mistakes, resulting from inattention or disengagement. Yet, a study published April 26 in the journal Current Biology suggests, there’s more going on: mice can understand the rules of a task and still deviate in their behaviors, potentially testing their own hypotheses and attempting to learn more about their surroundings.
“These mice have a richer internal life than we probably give them credit for. They are not just stimulus response machines. They may have things like strategies.”
It appears that the decisions mice make during behavioral tests are more complicated than just basic reward-seeking choices. During human-imposed trials in the lab, mice may be continually exploring and re-testing the rules of their environment and performing their own small experiments.
[ Related: Mice may be able to recognize their own reflections ]
The findings expand our understanding of what’s happening inside rodent brains and indicate mice and other non-verbal animals might know more than they let on. The research could eventually help shed light on the neurological underpinnings of human behavior as well. “These mice have a richer internal life than we probably give them credit for,” says Kishore Kuchibhotla, senior study author and an assistant professor of neuroscience at Johns Hopkins University. “They are not just stimulus response machines. They may have things like strategies,” he adds.
Mice at the steering wheel
The work builds on previous research that tested mice on a simple licking task and adds a level of complexity with a two-choice test to parse mouse motivations. Kuchibhotla and lead study author Ziyi Zhu, a neuroscience PhD student, trained thirsty mice, restrained in place, to spin a wheel with their front legs in a certain direction in response to a sound. One tone corresponded with turning the wheel to the right, a second tone with spinning it to the left. If a mouse responded to either of the sounds with the correct action, it would get a tiny cup of water. If it spun the wheel the wrong way or didn’t spin it at all, nothing happened.
Throughout thousands of trials involving 13 mice, the researchers tracked mouse choice, response speed, and accuracy, and they noticed several patterns. For one, mice seemed to get more accurate in their decisions as the trials progressed, indicating they were mastering the task at hand. Individual mice also seemed to have quirks and preferences when it came to picking a wheel direction. And even when mice reached an expert level of wheel-steering competency, they would still display short bouts of wrong responses–often spinning the wheel in the same direction repeatedly, regardless of which sound was played.
[ Related: How video game tech, AI, and computer vision help decode animal pain and behavior ]
To better understand what was happening during these bouts, Kuchibhotla and Zhu instituted “probe” trials, where they temporarily stopped rewarding the mice for correct answers. Very quickly, mice changed course, stopped exploring, and began to respond to the right and left sound cues more accurately, in accordance with the pattern they’d been trained on–indicating the mice understood what they were supposed to do to get the cup of water, and had been purposefully forsaking reward.
“As behavioral neuroscientists who work in animal models, the onus is on us to come up with more clever and rich ways to extract meaning from nonverbal animals’ [actions],” says Dr. Brian Sweis PhD, a neuroscientist and psychiatrist at Mount Sinai who conducts animal behavior research but was not involved in the new study. “I think this paper did a really nice job of that… it was a beautiful deep dive into a behavioral analysis,” he adds–pointing especially to the follow-up examination of the initial trial data and the ways the researchers varied their experiments.
“It may look like the animal is making a ton of errors, but during those errors, it’s actually getting smarter.”
Using a computational model, Zhu and Kuchibhotla assessed how each trial outcome related to the ones before and after it and what factors seemed to be influencing mouse behavior. They found that reward played a big role, but so did a bias towards rotating the wheel in a preferred direction, which differed from mouse to mouse. Yet this bias wasn’t fixed–mice would switch it up, spinning to both sides over the course of many trials and when the researchers presented mice with sound prompts for exclusively their preferred direction, the mice would exhibit more periods of rotating the wheel to their non-preferred side. Taken altogether, these observations show a dynamic choice bias that the researchers hypothesize is a learning strategy.
Learning without language
“Mice are surprisingly using higher-order approaches to learn even simple tasks, which may seem maladaptive. It may look like the animal is making a ton of errors, but during those errors, it’s actually getting smarter,” says Kuchibhotla. “We put these animals in these bizarre situations. They don’t know when the environment may change. They don’t know when we may change the rules on them. There’s value in having this sort of continuous exploration.”
Where humans can rely on language to understand an assignment, non-verbal animals have to find out for themselves what the rules of a particular situation are. Kuchibhotla suggests this difference could account for why mice take on this continually shifting approach to a task. “Verbal or written instructions collapse the mental space of exploration. Once you know what you’re supposed to do, there’s no need to explore. That’s one of the hypotheses we have–that in the absence of instructions humans will [also] engage in continuous exploration.” He’s currently conducting follow-up research in human behavioral trials to determine if that’s true.
Finding lessons in mouse mistakes
Other follow-up work includes tracking the mice’s neural activity as they engage in the wheel spinning task, training and testing the mice on multiple tasks at once to see how strategies change, and running cognitively impaired mice through similar tests which could ultimately reveal underlying patterns in human neurological diseases like Alzheimer’s.
There are limitations to what this single study proves. Despite all the researchers’ carefulness, something simpler than strategizing could still be at play, offers Sweis. For instance, maybe the mice changed the direction they spun the wheel in every so often because their front legs got tired. “I don’t think that negates anything [the study authors are] showing here, but physical factors could be a driver,” says Sweis. “We have to understand the brian in the context of the whole body.”
Still though, examining the choice process, he explains, “gives us insight into the many different ways the brain can work” and can clarify what’s happening when things start to go wrong. He suggests another follow-up project could look at how aging influences the exploration process and if task flexibility shifts or declines with age. There are many possible offshoots for interesting research, and the study “serves as a rich foundation for us to understand biology a little bit more.”
It reframes decades of rodent behavior results, where errors were dismissed as uninteresting failures. “Animals need to make mistakes to learn,” says Sweis–and there’s lots we can learn from them too.
