For Martin Schrimpf, the promise of artificial intelligence is not in the tasks that it can achieve. That lies in what AI might reveal about human intelligence.
He is working to build a “digital twin” of the brain using artificial neural networks. The AI model is loosely inspired by how neurons communicate with each other.
Its final goal sounds almost slick and epic, but his approach is simple. First, he and his colleagues test people with language and vision-related tasks. Next, we compare observed behaviors or brain activity with results from AI models constructed to do the same thing. Finally, we use the data to fine-tune the model to create an increasingly human AI.
Because this process works best with more data and more models, Schrimpf has built an open source platform called Brain Score, which includes nearly 100 human neural and behavioral data sets. Since Schrimpf first developed the platform in 2017, when he was still in graduate school, researchers have tested thousands of AI models against human data.
Schrimpf originally planned to work in the technology industry, but after co-founding a software startup during his early academic background, he felt unfulfilled. “I thought I could ask neuroscientists how the brain works. That would help me build better AI,” he said. “But I realized there was a big opportunity in the opposite direction: Prototyping ideas in Silico [on a computer] Explain the brain using AI models. ”
He moved from his hometown of Germany to the United States and earned a PhD in Brain and Cognitive Sciences from the Massachusetts Institute of Technology. In 2023 he returned to Europe and started the Neuroai Lab at Lausanne, the Federal Institute of Technology, Switzerland.
