Pigeons Can Learn in a Similar Way to Artificial Intelligence, Study Finds

A new study has found that pigeons are capable of learning in a similar way to artificial intelligence (AI) and can reach almost 70% accuracy in tests of association learning. The study is published in Current Biology.

High- and low-level learning

The researchers, led by senior author and Stuit Professor of experimental psychology at the University of Iowa Ed Wasserman, aimed to investigate two types of learning – declarative learning and associative learning.

Declarative learning is a “higher-level” skill mostly associated with people and applies reasoning using groups of rules or tactics. On the other hand, associative learning centers around recognizing patterns and forming associations – for example, between the words “sky” and “blue”, or “water” and “wet”.

Many different animal species are capable of the “lower-level” technique of associative learning. However, only a select group of species, like chimpanzees and dolphins, are believed to be able to use declarative learning.

Computers employ the same basic methodology, the researchers contend, being “taught” how to identify patterns and objects easily recognized by humans.

The researchers suggest that a methodology similar to associative learning is also used by computers and AI, which can be instructed on how to identify patterns and objects. In an era in which AI is becoming increasingly commonplace, Wasserman and colleagues developed a test to investigate if we have “shortchanged” associative learning in the animal kingdom compared to AI.

“You hear all the time about the wonders of AI, all the amazing things that it can do,” said Wasserman, “It can beat the pants off people playing chess, or at any video game, for that matter. It can beat us at all kinds of things. How does it do it? Is it smart? No, it’s using the same system or an equivalent system to what the pigeon is using here.”

Can pigeons match wits with AI?

The investigators used a small group of four test pigeons for the study, drawing upon Wasserman’s five decades of experience in pigeon intelligence research.

Each bird was shown a series of visual stimuli that belonged to either one of two categories, requiring them to peck a button on either their right or left side to indicate which category each stimulus belonged to. If they answered correctly, they were given a food reward, while there was no reward for an incorrect response.

A sample of stimuli out of many thousands that the pigeons had to categorize (center). These were drawn from two different categories (left and right). Credit: Ed Wasserman, University of Iowa.

Wasserman describes the test as “diabolically difficult”, as no amount of rules or logic would help to sort the stimuli into categories, which differed by characteristics including line width and angle and concentric or sectioned rings.

“These stimuli are special. They don’t look like one another, and they’re never repeated,” Wasserman explained. “You have to memorize the individual stimuli or regions from where the stimuli occur in order to do the task.”

Initially, the test pigeons were correct around half of the time, though after many hundreds of tests and the incentive of a tasty reward, they eventually reached an average of 68% accuracy.

The “associative learning paradox”

“The pigeons are like AI masters,” Wasserman said, explaining that both pigeons and AI are capable of associative learning. “They’re using a biological algorithm, the one that nature has given them, whereas the computer is using an artificial algorithm that humans gave them.”

“People are wowed by AI doing amazing things using a learning algorithm much like the pigeon, yet when people talk about associative learning in humans and animals, it is discounted as rigid and unsophisticated,” he continued.

Wasserman claims that if this test were applied to people, they would not score highly and would likely quit, as it is the pigeons’ skill in “low-level” associative learning that enabled their success, ultimately demonstrating the paradox in how we see associative learning.

Reference: Wasserman EA, Kain AG, O’Donoghue EM. Resolving the associative learning paradox by category learning in pigeons. Current Biology. 2023;0(0). doi: 10.1016/j.cub.2023.01.024

This article is a rework of a press release issued by the University of Iowa. Material has been edited for length and content.