Part One: Cognitive Neuroscience

With the recent developments in artificial intelligence research, the real-world applications AI can bring are particularly exciting. Perhaps most exciting of all is the impact AI has on the neuroscience field.

But first, a short introduction on AI. AI must learn from large datasets (this is called “training”). After this period of learning, the models will analyze new inputs and perform the desired task, improving its performance as it continues to predict (this is called “testing”). On the other hand, neuroscience has continually been centered on understanding how billions of neurons are able to communicate and control behaviors. Thus, the both of them are able to support and build on top of one another: engineers study brain architecture to build better AI, while computational neuroscientists train neural networks to simulate tasks performed by specific parts of the brain. In this article, we will discuss the impacts of AI on cognitive neuroscience.

Cognitive neuroscience is the study of biological processes and underlying neural mechanisms, focusing on how brain structure and activity produce mental processes like memory, attention, and perception. By mapping mental functions to physical brain circuits, neuroscience and psychology are combined.

Often, studies on cognitive neuroscience focus on decision-making, memory, language processing, and attention. Brain activity is measured in various ways, including with fMRI (functional magnetic resonance imaging), EEG (electroencephalography) and TMS (transcranial magnetic stimulation).

Through tests, imaging, and scans, biomarkers can be found for neurological disorders (including Alzheimer’s and Parkinson’s) [1]. Additionally, evidence-based learning and memory retention strategies can be adjusted based on brain mapping data and how the brain absorbs information, assisting with improvements in education [2]. Besides this, the origins of atypical development and conditions such as dyslexia allow for intervention early on [3].

On its own, cognitive neuroscience provides strong benefits for healthcare and education. But how could AI benefit this field? The answer lies in computational modeling and algorithms.

Reinforcement learning, or RL, is a type of machine learning process where agents learn to make decisions by interacting with their environment [4]. These agents go through various stages of trial and error, addressing decision-making in uncertain environments.

Interestingly, cognitive neuroscience is the backbone of such AI systems. Conversely, using AI, we are able to decode and visually represent what we see and imagine. Reinforcement-learning algorithms have been implemented to analyze complex neuroscience data, since AI is able to find hidden patterns and build complex applications [5]. Additionally, AI can analyze brain scans and sometimes figure out the meaning of what a person is hearing, reading, or thinking, almost like decoding language from brain activity.

LLMs (large language models) are a category of deep learning models trained on immense amounts of data [6]. They are capable of understanding and generating language and content to perform specific tasks. They work as prediction machines, learning patterns in text and generating language that follow such patterns. Some examples of LLMs you may know are Claude, ChatGPT, and Gemini.

In cognitive neuroscience, ChatGPT has made strong strides, particularly in human behavioral simulations, standardized neuroimaging data analysis, and neurotheoretical validation (using biological data to confirm the accuracy of mathematical models or conceptual theories). A previous study has found that LLMs are able to surpass experts in predicting experimental outcomes: when LLMs indicated high confidence in their predictions, their responses were more likely to be correct [7]. Perhaps, our future will be one where LLMs assist us with making discoveries.

Despite the strong promise AI brings, it’s important to recognize that developments in AI models also depend on our current knowledge of neural architecture. Thus, it’s imperative to gain a further understanding of cognitive neuroscience in order to advance and optimize such frameworks.

However, challenges still remain in bridging the gap in knowledge between cognitive neuroscience and AI. Additionally, cognitive neuroscience and AI may not always share the same goals and priorities: while research on cognitive neuroscience aims to understand how the brain works, research on AI focuses on improving model performance. A prominent problem (”black-box”) is clear: AI models are also unable to demonstrate how they arrive at certain conclusions: there is no way to verify their answers, and an incorrect prediction without an explanation makes it difficult to trace the source of the error. So, models that show their reasoning (”white-box”) should be built to understand the why [8]. Current medical models may not perform well across certain populations and settings, making real-world application less accessible. Finally, as AI continues to grow, laws and scientific consensus must keep up: the issue of neuroethics has yet to be explored in-depth.

Play this custom Connections game to check your understanding of cognitive neuroscience!