Neuroscience is the study of the structure and function of the nervous system, which includes the brain, spinal cord, and nerves. Neuroscience inspired the emergence of a new branch of science – Artificial Intelligence (AI). AI is a branch of computer science that deals with the simulation of human intelligence processes by machines.
“AI is not even near its full potential; it’s just in its infancy. We haven’t seen anything yet
Harari, Y. N. (2016). Homo Deus. Harvill Secker.
This is a quote from the book ’Homo deus’ by Yuval Noah Harari, a well-known historian.
We, as a society, are heading towards Artificial General Intelligence. If one wants to accomplish artificial general intelligence, it needs a coalition of both AI and neuroscience. In other words, AI and neuroscience drive each other forward. However, the link between AI and neuroscience is not a new concept.
AI and neuroscience: a historical perspective:
The correlation between AI and neuroscience dates back to the 1940s.
1940’s: Artificial Neural Network
Researchers developed artificial neural networks in order to investigate neural computation. They had the capacity to compute logical functions.
1940’s: Deep learning
An advancement that revolutionized the field of AI through dramatic advancements is Deep learning. With this, one can understand the different convoluted layers of the cerebral cortex and understand how they perform basic functions like visual processing, memory, and motor control.
1950’s: Perceptrons
An attempt to artificially generate models that demonstrate the mechanism of processing information by neurons, Rosenblatt developed a model of neuronal signalling called perceptrons. It had individual nodes which received weighted inputs and could produce a binary output if they reach a threshold of inputs. He proposed that neurons might learn through a supervisory feedback mechanism.
1980’s: Multi-Layered Perceptrons
An advanced version of perceptrons arised in the 1980’s. They could perform versatile functions and identify and correct errors in their functions. This model used the back-propagation algorithm, an algorithm to test for errors by working it’s way backwards from the output nodes to input nodes.
1980’s: Recurrent Neural Networks
Recurrent Neural Networks (RNNs) are a model that shows the recurrent connections between the neurons of the cerebral cortex. RNNs are a type of deep learning. They could reportedly perform a wide range of cognitive functions, especially memory. Scientists used this model to test how the brain controls working memory.
This fell under the banner of Parallel Distributed Processing (PDP).
1980’s: Convoluted Neural Networks
Building up on this PDP to biological computation led us to Convoluted Neural Networks (CNNs). The mammalian cortex again inspired this.
Slowly, as deep learning began its evolution out of PDP and into a core area of AI, new ideas emerged through deep belief networks.
In parallel to deep learning, neuroscience also erected a second branch of contemporary AI- Reinforcement Learning (RL). It is an algorithm used in machine learning which learns to reward a desired behaviour and punish an undesired behaviour. It is able to perceive and interpret its environment and act accordingly.
How Neuroscience Benefits from AI:
AI based neural circuits to help neuroscientists test their hypothesis
By studying the nervous system, one could develop AI machines that mimic the neural circuit. Artificial Neural Networks (ANN) not only help neuroscientists test their hypothesis but also helps in the early diagnosis of psychiatric disorders. Developing an AI-based neural circuit can also help analyse data from large neuroimaging datasets in an unbiased way. The ability of AI to work with complex data and extract hidden patterns makes it a go-to choice for neuroscientists for their data analysis.
AI based neural circuits to understand and decode how the human mind works
AI also helps us understand the human mind better by decoding how neurons in our brain communicate effectively and how cognition works. Moreover, building intelligent machines help us in understanding the intelligence of human brains and the brains of other animals. AI models have helped us understand the different functional aspects of the brain like listening and speaking. The model could identify which areas of the brain light up when thinking and could use this for helping people with speech disorders. ANNs can help us identify how the brain processes information.
How AI Benefits from Neuroscience:
Human brain as a source of inspiration for designing ANN’s
Neuroscience strongly benefits from AI through diagnosis, treatments, and even data analysis. However, there are instances where AI benefits from neuroscience. Primarily, neuroscience helps the designing and development of different kinds of ANN. It acts as a rich source of inspiration for the development of new algorithms, eventually helping in making more intelligent machines. Secondly, it helps in validating the current AI-based models. Neuroscience has also inspired the development of attention-based AI.
Taking inspiration from attention
Attention is usually perceived as an orienting mechanism of perceiving things. However, it can also be internalized towards the contents of memory. This idea has inspired some work in AI. Since attentional mechanisms can select the information it needs from internal memory, it has also led to important developments in the field of machine learning. Neuroscience has inspired a method called transfer learning. it is a process in which knowledge obtained from a given task is reused when performing a related task/ problem. This boosts the performance time, the performance and increases task efficiency. Transfer learning has greatly improved the performance of ANNs.
Applications of AI in Neuroscience:
Computational psychiatry:
Applying the advances of neuroscience to the benefit of patients is tricky because of the involvement of a complex organ – the brain. But, bridging this gap is highly imperative to diagnose and treat psychiatric and behavioral disorders. Computational psychiatry combines different levels and types of computation with many types of data. This can help improve the understanding, prediction, and treatment of mental illnesses.
Computer-assisted therapy
Assisted therapy in the form of chatbots can help with psychiatry and behavioral treatments. It involves the use of computers to deliver aspects of psychotherapy to patients in need. It is used as a first line of treatment for depression in some countries like the United Kingdom. The complexities of such therapy programs range from a simple text-based format (like reading out of a brochure) to highly sophisticated virtual reality.
One of the key strengths of this form of therapy is the user’s convenience. A patient can access it in the comfort of their homes at any time without having to wait for appointments. Also, this will help preserve the privacy of the patients as some patients may not be willing to disclose details of their identity. However, an important point to be noted is the fact that it can never replace psychiatrists. It merely acts as an extender of treatment for psychiatric and behavioral disorders.
AI-assisted CT scan
It is almost equivalent to a radiologist. Although it is almost equivalent to a radiologist, a medical radiologist can never be replaced and he plays a prime role in the medical diagnosis. The radiologist along with AI join hands together to give efficient medical imaging. Medical imaging is the prime source of information about a patient’s well-being. But, the efficient interpretation of highly staggered data in the form of X-rays, CT scans, and MRI scans is a daunting task for physicians, which is why they look towards AI for help.
AI in medical imaging
The use cases of AI in medical imaging are very interesting. AI can help identify changes in the thickness of certain muscle structures or changes in the flow rate of blood through the heart. Certain fractures, injuries, and dislocations which are hard to visualize through the eyes of a radiologist can be identified by AI. Furthermore, the application of AI on large-scale medical imaging data can help us identify trends, patterns, and risk factors associated with the disease.
Robotic limbs
Robotic limbs are based on the brain-computer/machine interface (BCI). BCI is a direct communication between the brain and an external device. The device acquires the signals from the brain, analyses them, and converts them into specific commands to produce the intended action. In this case, the action is hand or leg movement. This application helps improve the quality of life in patients with stroke, cerebral palsy, spinal cord injuries, and other neurological conditions.
Reinforcement learning
Reinforcement learning, as mentioned previously is an application of machine learning that aims to address the maximization of the reward by mapping the states in the environment to their respective actions. This application is especially useful in learning animal behavior. In fact, the base for reinforcement learning is animal psychology. Furthermore, reinforcement learning has resulted in the proposal of a new theory on dopaminergic signalling, a crucial pathway responsible for the functioning of central and peripheral nervous system of the body.
Boons of AI and neuroscience:
- It reduces human touchpoint
- It helps neuroscientists test their hypothesis
- When scientists want to do behavioral studies, they look to AI for help. Several deep learning software toolboxes help in the 3D marker-less pose estimations across several species and several behaviors
- Before the emergence of AI, the diagnosis of psychiatric and developmental disorders was solely based on the patient’s subjective behavioral symptoms, making it ineffective. However, the power of computing and the ability to analyze large amounts of neuro-imaging datasets has been able to bridge the gap, thereby leading to better diagnoses and treatments of psychiatric disorders
Banes of AI and neuroscience:
- Ethical challenges- loss of identity, loss of privacy
- Although AI can generate accurate predictions, it doesn’t come up with an understanding of how the inputs and outputs relate to each other
- Even though AI originated from neuroscience, the biological possibility of a modern AI is still questionable
- Although ANNs are used to test a hypothesis, it is not brain-like. It cannot fully reciprocate the complex processes of the brain
Use cases of AI in neuroscience:
- Chethan Pandarinath aimed to help people with paralyzed limbs with robotic arms so they could use their arms naturally as they would with their own. He achieved this goal by first identifying the electrical activity in certain neutrons involved in the movement of limbs and fed the instruction to a robotic prosthetic
- The diagnosis of neurological infections like meningitis is tedious due to its complex symptoms. However, AI-based models have helped get the cerebrospinal fluid cell counts and their ratios and accurately predict the type of meningitis
- CNN’s with optical imaging accurately diagnose brain tumours in <150 seconds. Brain tumours are one of the most misdiagnosed illness in neuroscience .AI can assist in treating neurovascular disorders that impact brain oxygen supply. Stroke is a common and serious neuro-vascular disorder that can lead to paralysis. One company which is working on this is BrainQ .
- Traumatic brain injuries due to accidents are very common. AI can accurately detect brain injuries and protect patients from radiation.
- Neurosurgery is very risky and complicated medical procedure. It is more prone to mistakes and even a tiny error can make a difference between life and death. In these cases, AI can help neurosurgeons during brain surgery.
Despite the fast paced research in the field of AI and neuroscience, a lot of work needs to be done to bridge the gap between machines and human intelligence. Let me know what you think in the comments section.