AI models mimic animal behavior in complex task performance

Scientists have developed a new approach to training artificial intelligence systems by mimicking how humans learn complex skills: starting with the basics. This “kindergarten curriculum learning” helps recurrent neural networks (RNNs) develop more rat-like decision-making capabilities when solving complex cognitive tasks. The innovation addresses a fundamental challenge in AI development—how to effectively teach neural networks to perform sophisticated cognitive functions that integrate multiple mental processes, similar to how animals naturally approach complex problems.

The big picture: Researchers have created a more effective way to train neural networks by breaking complex cognitive tasks into simpler subtasks, significantly improving AI’s ability to mimic animal behavior patterns.

• The approach, dubbed “kindergarten curriculum learning,” focuses on teaching AI systems fundamental cognitive skills before combining them into more complex behaviors.
• Traditional training methods often fail to capture important aspects of animal cognition, particularly when tasks require integration of multiple cognitive functions over extended time periods.

Key details: The study focused on a temporal wagering task previously studied in rats, where the AI had to learn to make value-based decisions using long-timescale inference.

• The researchers identified essential subcomputations needed for the task and designed simpler “kindergarten” training exercises focusing on those fundamentals.
• This pretraining method proved crucial for RNNs to develop similar problem-solving strategies as rats, including the ability to infer hidden states over extended periods.

Why this matters: The research demonstrates how structured learning approaches from human education can improve artificial intelligence capabilities, potentially bridging the gap between AI and biological cognition.

• The findings could lead to more biologically plausible AI models that better capture the nuanced decision-making processes observed in living organisms.
• Such approaches may help develop AI systems that can handle increasingly complex cognitive tasks requiring multiple integrated skills.

In plain English: Just as children learn arithmetic before tackling calculus, these researchers found that neural networks perform better when first taught basic cognitive skills before attempting complex tasks—creating AI that thinks more like animals.

The mechanism: The pretraining approach specifically helped neural networks develop slow dynamical systems features necessary for both inference and decision-making.

• These features allow the networks to maintain information over longer periods and integrate different cognitive functions—capabilities that conventional training methods struggle to produce.
• The researchers’ approach effectively builds relevant “inductive biases” into the networks, guiding them toward solutions that resemble biological cognition.

Behind the numbers: The study relied on previously collected rat behavioral data, with the research team making their code and model files publicly available through multiple repositories.

• The implementation is accessible through GitHub (https://github.com/Savin-Lab-Code/kind_cl) and CodeOcean for broader scientific reproducibility.
• The original rat behavioral data and neural network files are available through Zenodo repositories for other researchers to build upon.