Martin Schrimpf, a neuroscientist, aims to build a “digital twin” of the brain using artificial neural networks to understand human intelligence. He created an open-source platform called Brain-Score, which contains nearly a hundred human neural and behavioral data sets, allowing researchers to test thousands of AI models against human data since 2017.
Schrimpf’s approach involves testing people on tasks related to language or vision, comparing the observed behavior or brain activity with results from AI models built to do the same things. He then fine-tunes his models to create increasingly humanlike AI. The process has been successful in mimicking human behavior, especially in vision and language systems.
While artificial neural networks have a neuron-level similarity to the neuronal processing units in the brain, Schrimpf emphasizes that there is still much to be explored in understanding human intelligence. He believes that current models are useful for brain science and that the ability to predict human neural responses suggests that human behavior can be reduced to computational processes.
However, Schrimpf also acknowledges the need for caution when developing AI models that can influence thought. He recognizes that this is an “ethical minefield” and emphasizes the importance of working with experts to develop responsible models. Despite concerns about timelines, Schrimpf remains optimistic that we can create a digital twin of the brain within a few decades.
Ultimately, Schrimpf views human intelligence as a pattern of information processing that can arise elsewhere. While this perspective raises questions about what it means to be human, he believes that it is an opportunity for us to learn more about ourselves and our experiences.
Source: https://www.quantamagazine.org/how-ai-models-are-helping-to-understand-and-control-the-brain-20250618