What can associative learning do for planning?

Area of Science:

• Comparative Cognition
• Artificial Intelligence
• Animal Behavior

Background:

• The power of associative learning in non-human animals is often underestimated by researchers in animal cognition.
• Artificial intelligence research demonstrates associative learning models can outperform humans in complex tasks like chess.
• Flexible planning in animals is frequently cited as a reason to rule out associative learning, but these studies face methodological critiques.

Purpose of the Study:

• To explore the potential of associative learning to account for flexible planning behaviors in non-human animals.
• To investigate whether a combined Pavlovian and instrumental conditioning model can simulate planning in apes and corvids.

Main Methods:

• A previously published sequence learning model, integrating Pavlovian and instrumental conditioning, was utilized.
• The model was used to simulate two key studies on planning in non-human animals: Mulcahy & Call (2006) and Kabadayi & Osvath (2017).
• Simulations tracked behavioral patterns consistent with planning, focusing on conditioned reinforcement and future state decision-making.

Main Results:

• Simulations demonstrated that behaviors aligning with current definitions of flexible planning can emerge from associative learning.
• The model exhibited planning behavior through conditioned reinforcement, learning to associate actions with future high-value rewards.
• Key patterns observed in the original planning studies were successfully replicated by the associative learning model.

Conclusions:

• It is not possible to rule out associative learning as a complete explanation for the observed flexible planning in apes and corvids.
• The findings suggest that complex planning behaviors may arise from sophisticated associative learning mechanisms.
• Future research on animal planning can benefit from integrating insights from artificial intelligence and animal learning theory.