Humans and other animals are better off learning by conjecture, using the information we have to figure out things that we cannot observe directly. New research from the Center for Mind and Brain at the University of California, Davis shows how our brain does this by creating cognitive maps.
“The work suggests a new framework for learning in a structured environment that goes beyond the association’s growing and empirical learning,” said Eri Burman, UC Davis and assistant professor of psychology at the Center for Mind and Brain and lead author of the article.
In structural environments, individual elements are regularly related to each other because they are often in the real world. Study insights can be used to improve educational strategies that promote the use of a cognitive map for accelerated learning through hypotheses and, potentially, approaches to accelerate learning transfer in machine learning. In artificial intelligence, Burman said.
Learning by conjecture vs. association
Most learning studies have focused on learning by association – how animals learn to associate one thing with another through trial and error. The difference between what was expected and what actually happened inspires learning in such cases.
When there is a hidden structure behind these associations, you can use direct observation to predict indirect and unseen results directly from the association chain.
For example, knowing that the quality of seasonal foods is controlled by climate change allows you to guess which foods are better to eat based on cooked foods in the same season, Burman said. Observing ripe apples, we can assume that pears must also be ripe, but not strawberries. It is important to know this type of structure when making decisions.
Another example is an investor who speculates that a technical bubble could be blamed for the fall in Facebook shares, suggesting that Microsoft shares will also fall soon.
“Knowing this hidden relationship means you can learn much faster,” Burman said.
Learning to experiment in a structural system
To investigate how people can use cognitive maps to learn information, graduate student Philip Witkowski, project scientist Seongmin Park and Burman have created a work. In a series of trials, volunteers were asked to choose two of four abstract sizes that could be two different gift cards (e.g., Starbucks or iTunes). The volunteers chose them based on two facts: their estimates of the probability of each size being directed toward a particular gift card, and the randomly assigned payments for each gift card.
The size is divided into two pairs. In each pair, the probability of one shape leading to a certain result was the opposite of the other size. For example, if there was a 70% chance that Form A would lead to result 1, there was a 30% chance that Form B would lead to the same result and vice versa. Thus, information on topics can be obtained by estimating the probability of a result, such as shares from Microsoft Facebook shares. The pair of shapes were not attached, so subjects could not learn anything about the results of choosing C or D shapes from the results of choosing A or B.
Researchers have tracked how subjects learned about the system by observing their progress through various experiments. After analyzing the results, they found that the volunteers were using speculative learning to decide which sizes to choose.
Some volunteers were re-invited for the second part of the experiment, while their brain activity was measured by effective magnetic resonance imaging while performing the same task. When there is a difference between your previous knowledge and what you have just achieved, learning manifests itself as an explosion of activity in the brain, an “update of faith.” Activity-related activity has been found in the prefrontal cortex and midbrain areas where the neurotransmitter dopamine is secreted.
At the same time, researchers have found a representation of organizations that control the hidden (or latent) potential for A and B in the prefrontal cortex.
The FMRI results show that the brains represent different results against each other, Burman said. This performance allows for those “aha” moments.
Conventional thinking holds that the progressive learning of reward from direct experience is enhanced by the release of dopamine into the brain. The new study also includes dopamine but for speculative learning.
“Our work suggests a more general role for dopamine signals in updating beliefs through hypotheses,” Boerman said.