Extensive research has shown that the ways in which we perceive the world are in fact influenced by our own expectations.
It is called “prior beliefs” and helps us make sense of what we are experiencing in the present, based on similar experiences of the past.
Consider, for example, how a patient’s X-ray image containing a shadow can be disregarded by an intern with little experience, but will instantly catch the attention of a professional physician. The seasoned physician’s previous experience helps him arrive at the most likely interpretation of a weak signal.
By combining previous knowledge with uncertain evidence our perceptions, thoughts, and actions are impacted in major ways; this is known as Bayesian integration.
Recently, MIT neuroscientists have found distinctive brain signals which encode these prior beliefs. In addition, they have discovered that our brains use signals to make judicious decisions when faced with uncertainty.
“How these beliefs come to influence brain activity and bias our perceptions was the question we wanted to answer,” said Mehrdad Jazayeri, the Robert A. Swanson Career Development Professor of Life Sciences, a member of MIT’s McGovern Institute for Brain Research, and lead author of the study.
The scientists trained animals to perform a timed task in which they needed to reproduce various time intervals. This task is highly challenging as our sense of time is not perfect and can often go too slow or too fast. But when intervals are within constant fixed range, the most efficient strategy is to bias responses toward the middle of the range. And that is precisely what animals did. In fact, recording from frontal cortex neurons showed a simple mechanism for Bayesian integration: Previous experience warped the depiction of time in the brain so that neural activity patterns associated with different intervals formed a bias toward those that were within the range of expectancy.
For centuries, statisticians have understood that Bayesian integration is the best strategy for handling uncertain information. We automatically rely on prior experiences to optimize behavior when we are faced with uncertainty.
“If you can’t quite tell what something is, but from your prior experience you have some expectation of what it ought to be, then you will use that information to guide your judgment,” said Jazayeri. “We do this all the time.”
In the latest study, Jazayeri and his team wanted to learn how our brains encode prior beliefs and put them to use in the control of behavior.
They trained animals to reproduce a certain time interval by giving them a task called “ready-set-go.” This task requires animals to measure the time between the two “ready and set” flashes of light after which the “go” is generated by making a delayed response when the same amount of time has passed.
The animals were trained to perform this task in two contexts. Intervals varied between 480 and 800 milliseconds in the “short” scenario, while they were between 800 and 1200 in the “long” context. At the first stages of the task, the animals were given information about the context through a visual cue. Hence, they knew intervals can be expected from either the shorter or longer range.
In previous cases, Jazayeri had demonstrated that humans performing this task tend to be biased toward the middle range. In this case, they found that animals tend to do the same. For instance, if animals were expecting a short interval, and were given an 800 milliseconds interval, the interval they produced was a little below 800. In turn, if they believed the interval would be longer and were given the same interval of 800, they produced an interval a bit higher than 800.
“Trials that were identical in almost every possible way, except the animal’s belief led to different behaviors,” said Jazayeri. “That was compelling experimental evidence that the animal is relying on its own belief.”
As soon as they understood that the animals relied heavily on prior beliefs, the scientists embarked on a mission to find out how the brain encodes prior beliefs to guide behavior.
Neuron activity from around 1400 neurons in a frontal cortex region was recorded. This was previously shown to involve timing.
During the “ready-set” timeframe, each neuron’s activity profile evolved in its own way, and around 60% of neurons had different patterns of activity depending on the short/long context. In order to make sense of the signals, the researchers took a closer look at the evolution of neural activity for the whole population over time and discovered that prior beliefs actually bias behavioral responses by warping the neural representation of time toward the middle of the range of expectancy.
“We have never seen such a concrete example of how the brain uses prior experience to modify the neural dynamics by which it generates sequences of neural activities, to correct for its own imprecision. This is the unique strength of this paper: bringing together perception, neural dynamics, and Bayesian computation into a coherent framework, supported by both theory and measurements of behavior and neural activities,” said Mate Lengyel, professor of computational neuroscience at Cambridge University (Mr Lengyel was not involved in the study).
Scientists believe that previous experiences change the power of connections between neurons.
The power of these connections, also known as synapses, determines how neurons act upon each other and limits the activity patterns that a network of interconnected neurons is able to generate. The discovery that prior experiences warp neural activity patterns opens a window which shows us how experience alters synaptic connections.
“The brain seems to embed prior experiences into synaptic connections so that patterns of brain activity are appropriately biased,” saod Jazayeri.
The researchers used a new computer model consisting of a network of neurons that could perform the abovementioned “ready-set-go” task. By using techniques borrowed from machine learning, they were able to accurately modify the synaptic connections and create a model which behaved like the animals.
These models provide a substrate for in-depth analysis of the underlying mechanisms (a procedure known as “reverse-engineering”), which makes them incredibly valuable.
Astonishingly, reverse-engineering the model showed that it solved the given task in the same way as the monkeys’ brain did. In addition, the model had a warped representation of time according to previous experience.
The researchers used the computer model to further dissect the underlying mechanisms using perturbation experiments that are not possible to do in the brain at present.
By using this method, they managed to show that the unboxing of the neural representations removes the behavioral bias. This crucial discovery validated the vital role of warping in Bayesian integration of prior knowledge.
The researchers are now planning to study the way in which the brain builds up and gradually fine-tunes the synaptic connections that encode prior beliefs while an animal is figuring out how to perform the timing task.
Please share and comment in the section below if you found this article of interest to you.