Biological neurons have more control over their environment than previously thought, a new study suggests, and that’s something that could be reproduced in artificial neural networks used in machine learning to improve the artificial intelligence (IA).
A team at the Flatiron Institute (USA) has developed a new model of neurons that considers them as small “controllers,” an engineering term for devices that can influence their environment based on information collected about that environment.
“Neuroscience has advanced a lot in the last 60 years and we now recognize that Previous models of neurons are quite rudimentary” said Dmitri Chklovskii of the Flatiron Institute and lead author of the paper published in PNAS.
The new model suggests that this decades-old approach does not capture all the computational capabilities that real neurons possess and that “This old model is potentially holding back AI development”says a statement from the research center.
The study postulates that Individual neurons exert more control over their environment than previously thought and that updated model could ultimately lead to more powerful artificial neural networks that “better capture the powers of our brains,” the researchers say.
Almost all neural networks using modern artificial intelligence tools, such as ChatGPT, They are based on a computational model of a living neuron from the 1960s.
Artificial neural networks aim to imitate the way the human brain processes information and makes decisions, although in a much more simplified way.
These networks are built from ordered layers of “nodes.” The network starts with an input layer of nodes that receives information, then has intermediate layers of nodes that process the information, and ends with an output layer that sends the results.
Typically, a node only passes information to the next layer if the total input it receives from those in the previous layer exceeds a certain threshold.
When current artificial neural networks are trained, information passes through a node in only one direction and there is no way for them to influence the information they receive from those up the chain.
However, It is possible that our neurons are not mere passive transmitters of information but rather that they control the status of others, the study adds.
The proposed new model treats neurons as tiny “controllers,” and Chklovskii believes it “could be an important step toward improving the performance and efficiency of many machine learning applications.”
The neuron-as-controller model is inspired by what scientists know about large-scale brain circuits made up of many neurons.
The team found that a novel form of control, known as direct data-driven control, is simple and effective enough to make it biologically plausible for it to take place in individual cells.
Chklovskii next plans to look at types of neurons that don’t fit into his new model, such as those in the retina that receive direct input from the visual environment.
These neurons may not be able to control their inputs in the same way as neurons deeper in the brain, but they could use some of the same principles, being able to predict their inputs, even if they cannot influence them.