summary: Although decision-making often involves trial and error, traditional models assume that they always act optimally based on past experience. A new study reveals how humans and animals actually make choices using small, interpretable artificial neural networks.
These models predicted individual decisions more accurately than traditional theories by reflecting imperfect behavior in the real world. This task can change the way you understand cognitive strategies and coordinate mental health and behavioral interventions.
Important facts:
- Realistic insights: Small AI models revealed that decision-making strategies are often optimal but systematic.
- Individual differences: The model predicted individual behavior better than the optimality-based framework.
- Wideer impact: Findings may inform mental health approaches by mapping cognitive diversity.
sauce: NYU
Researchers have long been interested in how humans and animals make decisions by focusing on the behavior of trials and errors informed by recent information.
However, traditional frameworks for understanding these behaviors assume that best decisions are made after considering past experience, and thus may overlook the specific reality of decision making.
The newly released research by a team of scientists deploys AI in innovative ways to better understand this process.
By using small artificial neural networks, the researcher's work illuminates in detail what drives an individual's actual choice, whether these choices are best or not.
“Instead of assuming the brain Should Learning in optimizing our decisions, we have developed an alternative approach to how to discover individual brains. actually Marcelo Matter, an assistant professor at the Department of Psychology at New York University and one of the authors of the papers published in the journal, explains: Nature.
“This approach works like a detective and reveals how decisions are actually made by animals and humans. Using small neural networks, they discover decision-making strategies that scientists have overlooked for decades.”
The authors of this study point out that small neural networks, sampled versions of neural networks commonly used in commercial AI applications, can predict animal selections far better than classic cognitive models that assume optimal behavior due to their ability to illuminate optimal behavioral patterns.
In laboratory tasks, these predictions are as good as predictions produced by larger neural networks, such as those that power commercial AI applications.
“The advantage of using very small networks is that you can deploy mathematical tools to easily interpret the reasons or mechanisms behind an individual's choices. This is more difficult when using large neural networks, such as those used in most AI applications.
“The large neural networks used in AI are very good at predicting things,” says author Marcus Benna, an assistant professor of neurobiology at UC San Diego's Faculty of Biological Sciences.
“For example, you can predict which films you want to see next. But it's very difficult to briefly explain the strategies these complex machine learning models use to make predictions. why I think they'll like some movies more than some movies.
“Training the simplest versions of these AI models to predict animal choices and analyse dynamics using physics methods can shed light on internal mechanisms in easier to understand terms.”
Understanding how animals and humans learn from experience and making decisions is not only a major goal of science, but it is also broader and useful in the business, government and technology fields.
However, because existing models of this process are intended to depict Best Decision making often fails to capture realistic behavior.
Overall, it is explained in the new model Nature The study coincided with the decision-making process of humans, non-human primates and experimental rats.
In particular, the model predicts suboptimal decisions, thus better reflecting the “real world” nature of decisions and contrasts with the assumptions of traditional models that focus on explaining optimal decisions.
Furthermore, models of scientists from NYU and UC San Diego were able to predict decision-making at the individual level, revealing how each participant deploys different strategies as they arrive at decision-making.
“Just studying individual differences in physical characteristics has revolutionized medicine, understanding individual differences in decision-making strategies can change our approach to mental health and cognitive function,” concludes Mattar.
Funding: This study was supported by grants from the National Science Foundation (CNS-1730158, ACI-1540112, ACI-1541349, ACI-1541349, OAC-1826967, OAC-2112167, CNS-2100237, CNS-2120019). Communications and Information Technology/Qualcomm Institute.
About this AI and decision-making research news
author: James David
sauce: NYU
contact: James David – NYU
image: This image is credited to Neuroscience News
Original research: Open access.
Marcelo Mattar et al. Nature
Abstract
Discover cognitive strategies with small recurrent neural networks
Understanding how animals and humans learn from experience and making adaptive decisions are fundamental goals of neuroscience and psychology.
Normative modeling frameworks such as Bayesian inference and reinforcement learning provide valuable insight into the principles that manage adaptive behavior.
However, the simplicity of these frameworks often limits their ability to capture realistic biological behaviors, leading to a cycle of handmade coordination that is prone to researchers' subjectivity.
Here we present a novel modeling approach that utilizes recurrent neural networks to discover cognitive algorithms that manage biological decision-making.
Neural networks of only 1-4 units often outperform classic cognitive models and show that they match larger neural networks in predicting individual animal and human choices across six well-studied reward learning tasks.
Importantly, it uses the concept of dynamic systems to interpret trained networks, allowing for a unified comparison of cognitive models, and reveals the detailed mechanisms underlying selection behavior.
Our approach also estimates the dimensions of behavior and provides insight into the algorithms learned by Meta's extended learning artificial intelligence agents.
Overall, we present a systematic approach to discover interpretable cognitive strategies in decision-making, providing insight into neural mechanisms and a foundation for studying healthy, dysfunctional cognition.
