New research seeks to answer what AI models can teach us about the human brain

AI and language processing: A new frontier in neuroscience research: The intersection of artificial intelligence and language processing has become a hot topic in neuroscience, sparking debates about the effectiveness of using Large Language Models (LLMs) to understand human brain function.

• Researchers are exploring ways to link LLM outputs to brain activity during various language tasks, as discussed at a recent symposium of the Society for the Neurobiology of Language.
• LLMs, such as ChatGPT, possess an experiential equivalent of nearly 400 years, far surpassing the average human lifespan.
• However, questions arise about whether these AI models can provide meaningful insights into the biological and evolutionary aspects of human language processing.

Historical context of human language: The development of human language spans a much longer timeframe than the existence of AI, raising doubts about the applicability of LLMs in understanding our cognitive processes.

• Modern humans have existed for over 40,000 years, with ancestors spending thousands more years developing communication skills.
• The process of language acquisition in children adds another layer of complexity to human language development.
• These factors contribute to the unique cognitive abilities that humans possess, particularly in language use.

Limitations of AI in language research: Critics argue that AI models may overlook crucial biological and evolutionary information essential to understanding human language processing.

• Elizabeth Bates, a renowned cognitive scientist, emphasized the need for networks to “get a body and get a life,” highlighting the importance of physical and experiential aspects in language development.
• The debate on the usefulness of LLMs in understanding human language processing intensified during the symposium’s coffee break, revealing a divide among researchers in the field.

An election prediction analogy: To illustrate the debate surrounding AI and language research, an interesting parallel can be drawn with different approaches to predicting US presidential elections.

• Allan Lichtman’s “13 keys to the White House” model, developed in collaboration with Vladimir Keilis-Borok, uses historical factors to predict election outcomes.
• Lichtman’s approach has successfully predicted numerous elections, including unexpected results like Donald Trump’s 2016 victory and subsequent loss in 2020.
• In contrast, Nate Silver’s model relies on sophisticated mathematical and statistical analysis of polling data from various states.

Simplicity vs. complexity in predictive models: The comparison between Lichtman’s and Silver’s election prediction models raises questions about the most effective approach to understanding complex systems like human language or political outcomes.

• Lichtman’s model uses 13 simple yes/no questions based on historical patterns, while Silver’s model employs advanced statistics and a wide array of data points.
• The success of Lichtman’s simpler model in accurately predicting election results challenges the assumption that more complex models are always superior.

Balancing new technologies with traditional methods: The debate surrounding AI in language research reflects a broader question about the role of advanced technologies in scientific inquiry.

• While LLMs and other AI tools offer new perspectives and capabilities, they should not be seen as inherently superior to traditional research methods.
• Researchers are encouraged to consider the value of historical and biological context when studying human language processing.
• A balanced approach that combines new technologies with established methods may yield the most comprehensive understanding of complex phenomena like language.

Looking ahead: Integrating AI and traditional approaches: As the field of neurolinguistics evolves, researchers face the challenge of effectively integrating AI tools with traditional research methods.

• The use of LLMs and other AI technologies in language research is likely to continue, but with a more critical eye towards their limitations and potential biases.
• Future studies may focus on developing hybrid approaches that leverage the strengths of both AI models and traditional neuroscientific methods.
• Continued debate and collaboration among researchers from various disciplines will be crucial in advancing our understanding of human language processing.