Despite the continuous advances made by the state of the art in machine learning and artificial intelligence (AI), the human brain still differs from that of other animals in its ability to connect the dots and derive information, that supports problem solving in situations that are inherently uncertain.
It does remarkably well despite sparse, incomplete, and almost always less than perfect data. On the contrary, it is very difficult for machines to derive new knowledge and generalize it beyond what they have been specifically trained to do, also lit or exposed.
How the brain represents and navigates space
As early as 1948, scientists suggested that the brain forms a “cognitive map”, a flexible and adaptable internal model of the external spatial world that can dynamically update itself when new external information arrives. The biological and physiological mechanisms responsible for the creation and maintenance of such cognitive maps were unknown, it was hypothesized that this map inferred paths and routes through its physical spatial environment for the brain regardless of how the brain does so and can navigate. Abbreviations and paths that you have not been physically exposed to. It does this based on what it has learned from partial information.
More recent work is discovering the cells and parts of the brain that are involved in creating such cognitive maps. In particular, it has been suggested that “place cells” in the hippocampus, an important structure involved in learning and memory, encode the current location in the brain of an individual in their physical spatial environment. They also predict future spatial positions in the context of navigation and route selection decisions.
“Lattice cells” in another part of the brain, the medial entorhinal cortex, located in the temporal lobes of the brain, are believed to be responsible for generalizing how the brain encodes spatial information in order to make navigational and route decisions in new situations that the entorhinal cortex acts as a relay or interface between the hippocampus (which lies deeper in an older evolutionary part of the brain) and the rest of the cortex.
These ideas are supported by theoretical (mathematical) and computer models of the brain that go one step further and suggest how neurons in the entorhinal cortex create a two-dimensional grid code. This code is able to derive new paths and shortcuts through spatial environments and in situations the brain has never encountered. In addition, analog neuron-based grid-based models were tested in artificial neural networks.
Beyond Space: Representing and Deriving Abstract Concepts
Other recent works have expanded the concept of cognitive maps beyond the spatial representation of the physical environment. Cognitive maps of continuous representations of sound frequencies and odor concentrations have also been shown experimentally demonstrated more than a month ago that in addition to encoding representations of abstract cognitive maps of non-spatial information, the brain is able to use such maps to infer abstract relationships “paths” through maps to which it was not directly exposed. They need information that goes beyond what they have directly experienced and learned.
The researchers used functional magnetic resonance imaging (fMRI) to infer brain activity from blood flow measurements in subjects performing various cognitive tasks. Specifically, they were able to show that the brain creates a learned abstract two-dimensional grid (not spatial or physical). Navigation system for traversing and prioritizing learned social hierarchies, relationships of social rankings between different people.
It is important that the social hierarchy was not presented directly to the test subjects, in other words, they were never directly exposed to all relationships that connect the various individuals in the data set, but was reconstructed in an organized manner by the given brain. by pairs of social information between pairs of individuals.
Then, through carefully constructed and controlled task decision experiments, they were also able to show that the brain uses these cognitive maps of the social hierarchy to derive various “connections” between individuals to aid decision-making problems deduce socially important connections and paths in your abstract internal model of what social hierarchies look like. It is the equivalent of the brain that derives routes and navigates through physical space from A to B even though it has not been exposed to a direct path from A to B.
Because the researchers used fMRI, they were also able to trace which brain regions are involved in the construction of such abstract cognitive maps. These are the same regions of the brain that have been observed to play a role in the construction of spatial cognitive maps. This enables neuroscientists to understand the underlying physiology that makes these skills possible.
As with any work, there are of course limitations in the experimental methodology and, as a result, how far the interpretations of the results can be taken. In this special case, the scenarios and the amount of data presented and the decision-making tasks required by the test subjects. What made the resulting cognitive maps two-dimensional was the fact that there were only two classes of traits for the subjects to consider, which was both the cognitive burden associated with learning and developing relationships and the difficulty of the decision problems they were having were asked to solve simplified.
It is not obvious how this can be scaled internally in the brain into more complex multi-dimensional features and data. Or how to scale when the number of learned instances required increases significantly. To be fair, what the researchers were able to show in this article was an experimental tour force on itself. But it does raise questions that need to be answered in future research.
However, it is likely that this is not limited to just social hierarchies and spatial navigation, but represents a generalized ability of the brain to take in all sorts of diverse and abstract information and to make associations and relational internal models. In view of the often scarce and almost always incomplete or imperfect information, it uses these cognitive maps of the internal model to derive ways that support decision-making tasks.
How the brain evolved to achieve these abilities and the underlying “algorithms” that make them possible is still not well understood. Developing and researching mathematical models that will lead to a deep understanding of what the brain does and how it is not yet mature and remains a very active area of research.
But no matter how the brain does it, it is a skill that machines and AI have not (yet) been able to seriously recreate. While spatial navigation through machine learning systems has come a long way and can be impressive, it still requires a lot of training and can fail miserably when unpredictable scenarios arise.
But where machines really can’t compete with the brain is making reasonably educated guesses and decisions in the face of incomplete, sparse, and noisy abstract situations where problems must be solved through inference, and there is much to be learned from the brain for artificial intelligence.
