In a series of experiments, a study confirmed that pigeons can learn a variety of category structures – some devised to foil the use of advanced cognitive processes. We then contrive a simple associative learning model to see how effectively the model learns the same tasks given to pigeons. The study finds that the problem-solving skills of pigeons are very similar to that of an AI.

What explains its surprising success? Does it possess elaborate executive functions akin to those deployed by humans? Or does it effectively deploy an unheralded but powerful associative learning mechanism? 

The close fit of the associative model to pigeons' categorization behavior provides unprecedented support for associative learning as a viable mechanism for mastering complex category structures and for the pigeons using this mechanism to adapt to a richly visual world. This model will help guide future neuroscientific research into the biological substrates of visual cognition.

The test

In each trial, the researchers assume that the pigeon encodes the displayed stimulus and computes the association between the stimulus percept and a categorization response based on an associative network. To assess the power of associative learning, they developed a computational model that uses only the principles of stimulus-response association through reinforcement learning and stimulus generalization. Because the model involves just two mechanisms, if the model can successfully capture the pigeon's behavior in a wide variety of categorization tasks, then they can interpret the model's success as evidence that the pigeon deploys only these two mechanisms. In other words, if the machine-like model behaves as does a pigeon, it serves as an appropriate analogue of the principles underlying the pigeon's category learning, which will help guide future neuroscientific research into the biological substrates of visual cognition.

They used four different categorization tasks involving unique structures that afforded a stringent test of the pigeon-as-machine metaphor:

•  (1) A task involving rule-based or non-rule-based conditions with an abrupt change in the category structure once pigeons attained the initial learning criterion, 
• (2) A task involving concurrently trained rule-based and non-rule-based conditions involving abutting but non-overlapping stimulus distributions, 
• 3) A task consisting of stimulus distributions that entailed two concentric rings and 
• (4) A task consisting of stimulus distributions that involved the same two rings but now cut into eight separate sections.

Like an AI

The researchers found that the mechanism pigeons use to make correct choices is similar to the method that AI models use to make the right predictions.

According to Edward Wasserman, study co-author and professor of experimental psychology at the University of Iowa, pigeon behavior suggests that nature has created an algorithm that is highly effective in learning very challenging tasks, not necessarily with the greatest speed, but with great consistency.

In an AI model, the main goal is to recognize patterns and make decisions. Pigeons, as research shows, can do the same. Learning from consequences, when not given a food pellet, pigeons have a remarkable ability to correct their errors. Similarity function is also at play for pigeons by using their ability to find the resemblance between two objects.

According to Brandon Turner, lead author of the study and professor of psychology at Ohio State University, with just those two mechanisms alone, you can define a neural network or an artificially intelligent machine to solve these categorization problems. It stands to reason that the mechanisms that are present in the AI are also present in the pigeon.