Background:
Neuroscience studies the nervous system, which includes both the central and the peripheral nervous system. The central nervous system includes the brain and the spinal cord (this article focuses on the brain), while the peripheral nervous system includes the nerves that extend to the extremities of our bodies, allowing the signals from our central nervous system to send signals throughout our bodies.
Artificial intelligence is formally defined as “a set of technologies that allow computers to perform advanced functions including the ability to see, understand and translate spoken and written language, analyze data, make recommendations, and more.” In short, it’s basically advanced technology that allows computers to perform human functions. For example, there’s ChatGPT that we all know and love, that can answer our questions, have a conversation as if it were human, and so much more.
How is AI Modeled After Our Brain?
-
Neurons and Artificial Neurons
-
Neurons (nerve cells) are the basic building blocks of our brain. Each neuron is made up of 3 main parts: the cell body (contains the organelles), dendrites (receives the signals), and axons (sends the signals). Fun fact! The longest cell in our body is a nerve cell that extends down our spinal cord, and in most individuals, it is 1-1.5 meters long! The axons are myelinated (surrounded by a layer of special fats and proteins) in order to help the signal travel faster.
-
Artificial neurons are the basic units of artificial neural networks, which is an artificial model of the complex networks of neurons in our brains. This artificial neural network allows computers/other technologies to mimic human behavior, like decision making or learning. Similarly to a neuron, these artificial neurons are arranged into three layers: the input layer (data enters), the hidden layers (data processing), and the output layer (final product). The hidden layers are the most complex, because they have such a wide variety of functions, which range from pattern recognition to data processing.
-
-
Neuroplasticity
-
Neuroplasticity is the capacity for our brains to change throughout our life, in response to things like new experiences or injury. There are a few different types of neuroplasticity, the two main groups being structural or functional. Structural neuroplasticity involves physical changes in the structure of the brain, which can be caused by things like the formation of new synapses or neurons, or axonal growth. Functional neuroplasticity includes more of changes in the brain’s neural circuits or pattern, rather than changing the structure of the brain itself, caused by things like recovery after injury, or learning. Maladaptive neuroplasticity is a type of neuroplasticity where the changes in the brain might lead to more negative effects, which can be caused by things like damage to certain neurons.
-
Plasticity in AI refers to the ability for an artificial neural network to change its outputs in response to changing inputs. In order to make that change, the hidden layers will undergo some type of transformation.
-
-
Synapses and Weights
-
Synapses are the gaps between an axon and a dendrite, where a signal is transmitted from one cell to another. To do so, the brain will use chemicals known as neurotransmitters; each neurotransmitter creates a different response in the postsynaptic neuron (receiving end). With repeated activation, a synapse can become stronger, with the signal creating a larger response in the postsynaptic cell, a phenomena known as synaptic plasticity. Another process known as synaptogenesis refers to the formation of new synapses in response to learning new information. This allows the brain to store the new information learned, and this process is especially important to the hippocampus. More broadly, there is a concept called Hebbian Learning, which refers to the strengthening of an entire pathway or group of neurons in response to an action, rather than just a singular synapse.
-
Weights in neural networks are numbers that designate the strength of certain connections between artificial neurons. The higher the number, the stronger the connection, which also relates to the relative importance of a connection when it comes to a certain process.
-
-
Memory
-
In the human brain, there’s a region of the brain called the hippocampus that stores our memories. It takes newly learned information and turns it into long term memory.
-
There’s a type of recurrent neural network called Long Short Term Memory that is designed to handle long sequences of data, and it’s able to take in new data without forgetting what happened earlier in the sequence. There’s a special part called the Forget Gate, that chooses what data/information to discard because it’s either irrelevant or outdated.
-
-
Decision Making
-
Humans make decisions based on both inputs from the prefrontal cortex (rational or logical input) and the amygdala (emotional input).
-
AI makes decisions through its algorithms that help it evaluate the potential outcomes of a decision
-
AI’s application in Neuroscience
-
Neuroimaging:
-
AI Enhanced Brain Imaging: Brain imaging includes scans like MRIs, PET Scans, and EEGs. Artificial intelligence is able to detect patterns that us humans might not be able to see with our naked eye. Or, it’s able to enhance the image so that humans can have an easier time reading the scans.
-
Neural Connectivity Maps: these maps are a detailed representation of the synapses in the brain, which help with the visualization of the ways regions of the brain communicate with each other. There’s a project called the Human Connectome Project that aims to map every single neural connection in the human brain. Processes like Diffusion Tensor Imaging uses AI to track the movement of water molecules along myelinated axons in the brain and recreate a model of brain connections based on its track of the water movement.
-
-
Brain Computer Interfaces: AI translates brain signals into commands to control external devices:
-
Restores Mobility: Allows for people who have prosthetics or paralysis to control robotic limbs or wheelchairs. Or, there’s a technology where AI interprets signals from our brains to translate thoughts into texts/speech for people who cannot communicate.
-
Future: In the future, BCIs might work to improve human cognitive abilities like memory retention, focus, or learning speed. For example, hippocampal BCIs might one day work to restore memory function in individuals with Alzheimer’s.
-
-
Neurological conditions
-
Diagnosis: AI is able to recognize patterns that are more subtle than what the human eye might be able to recognize. AI can detect changes in the activity of the prefrontal cortex in patients with depression, or in epilepsy patients, it can predict the onset of a seizure based on EEG information. Or, in ADHD, it can detect differences in the communication patterns in the brain of someone with ADHD and compare it to someone who doesn’t.
-
Treatment: AI is able to personalize treatments to an individual by analyzing their neural activity and analyzing their clinical history. For example, in neuromodulation therapies (electrical, chemical, or magnetic stimulus to interact with nerve function), AI can optimize the treatment by finding the best location in the brain to stimulate, or change the intensity of the stimuli. Or, AI is able to use a patient’s genetic data, brain imaging, and clinical history to predict a reaction to a specific drug.
-
