Novak Djokovic strategically positions himself on the tennis court in advance of his opponent's shot. He anticipates the server's intentions and reacts with precision long before the ball, traveling at 130 mph, reaches his side of the court.
The brain simply cannot afford to only react to stimuli - it will be too late. Rather, it needs to be proactive in expecting what will happen, such as what the ace Tennis players do.
In 1960, Hermann von Helmholtz argued that rather than seeing the world directly, the brain might be making some of it up or ‘filling in the blanks’ where inputs are noisy. Helmohtz's key idea was to view perception as an "unconscious inference" where the brain makes the best guess at what is causing the incoming signals instead of relying solely on sensory inputs.
For this, the brain needs to learn from experience and based on past experience, be able to predict events in the real world. Neuroscientists embraced the idea of the brain as a ‘prediction machine’ in the early 20th century and since then, there has been growing evidence in favour of predictive processing arising from experiments in mammals. However, much is not clear regarding the neural mechanisms of how the brain generates predictions and how it uses them. It is also not clear how widespread the ability to predict is within the animal kingdom, or how early in development one can observe it.
A recent study by researchers at the National Centre for Biological Sciences (NCBS) shows evidence of predictive processing in the larvae of zebrafish that are only a few days old. The study elaborates on the role of the cerebellum in creating and updating predictions of the real world to fine-tune motor behaviour.
“This study is the culmination of several years of work combined with innovation and perseverance. We were pleasantly surprised by the results since it is quite unexpected that the larval zebrafish brain, with only about a hundred thousand neurons, is capable of such exquisite computations!”, says Prof. Vatsala Thirumalai, principal investigator of the study.
“We were looking at how networks of neurons learn a pattern they experience in the real world. The set of rules and connectivity of neurons that allows learning, is fundamental to all of us, ranging from an ant navigating its way to its nest to a dragonfly chasing its prey to humans in their daily life” says Sriram, lead author of the study.
Sriram and colleagues designed an interactive ‘video game’ for larval zebrafish. They fixed the head of the larva in a thin respirable drop of transparent agarose gel while the tail was left free to wiggle. Then, they placed the fish on top of a projection screen to create a virtual treadmill for the fish - patterns projected on the screen drifted forward slowly; in response, the larvae wiggled their tails as if to swim in the direction of the drift. This is a reflex that larvae exhibit called the ‘optomotor response’ or OMR in short. The experimenters then digitally coupled the extent of tail wiggle to the drifting patterns such that more wiggle caused the patterns to drift backwards giving the larvae a perception that they were indeed moving forward. With this behavioural response in place, the researchers then presented repeating pulses of forward flow, eliciting forward swim responses for every pulse. With greater repetitions of the pulsed optic flow, larvae produced swims with lesser lag time. This suggests that the fish brain forms a representation of what to expect from their world which helps them respond faster when the world remains consistent, says Sriram.
Then, they tricked the larvae by introducing a weak backward drift instead of a forward drift of the pattern. The altered stimulus surprised the larva as its expectations did not match its prediction. When this was followed by forward drift, larvae were slow to respond, as if they were doubtful of what stimulus would be given next. To the larva, any discrepancy in the pattern meant an unexpected event took place, and the pre-planned behavioural response was unsuitable, says Sriram.
Once the researchers had established that the larva relied on its prior belief, they set out to find how the brain forms an expectation and updates it when prediction errors occur. They recorded neuronal activity in the cerebellum- in humans, this is a fist-size region at the base of the brain and is implicated in motor coordination, balance and cognition. In larval zebrafish, the cerebellum is a much smaller structure containing fewer neurons compared to humans. Yet, it resembles the human cerebellum in many ways. The NCBS researchers marked a specific group of neurons in the cerebellum called Purkinje neurons, with a fluorescent sensor, whose intensity indicated the level of electrical activity. Using this method, they were able to show signals corresponding to stimulus expectation and prediction error in Purkinje neurons. For predictions to be learned and used both expectations and error signals will be needed. The group further showed that the strength of these signals correlated well with the readiness to swim. Thus, when Purkinje neurons showed strong signals for expecting the forward drift stimulus, fish were more ready to swim and produced a response quickly and vice versa.
The investigators then asked where the signals in Purkinje neurons came from. They were able to show that Purkinje neurons combined signals of expectation from granule cells and signals of error from the inferior olive neurons to evaluate whether predictions are true and swim with lower latency when what is expected is true.
This work helps us lay down a circuit pathway in the cerebellum directly involved in acquiring experience from the environment, updating predictions when they are incorrect and planning a motor response, says Sriram.
This study has opened the door to many more exciting questions for us to follow up on in the coming years. "One particular aspect that we are currently studying is how development changes prediction processing and how that impacts behaviours. Unlike humans, the larval zebrafish has to survive on its own from when it is only a few days old. Its brain is not yet fully developed but it is capable of quite a few maneuvers. This means that brains can produce function even when they are immature. Studies of these mechanisms could potentially help us find ways to restore function in human brains that have not developed normally.” says Prof. Thirumalai.
GIF Credit: Vatsala Thirumalai Lab, NCBS