Who you callin' bird-brained? Pigeons learn the same way AI models do
Despite many studies showing pigeons are surprisingly smart, from being as good at counting as primates, to being able to identify breast cancer in X-rays, scientists are fighting a losing battle to dispute their widely held reputation as being a bit “dim-witted.”
A new study has pitted the pigeon up against an artificial-intelligence model and found that both bird and computer follow a similar process in order to work out the problem they’re presented with.
"We found really strong evidence that the mechanisms guiding pigeon learning are remarkably similar to the same principles that guide modern machine learning and AI techniques," said Brandon Turner, lead author of the study and professor of psychology at Ohio State University. "Our findings suggest that in the pigeon, nature may have found a way to make an incredibly efficient learner that has no ability to generalize or extrapolate like humans would."
There are around 300 species of pigeons and doves, which belong to the Columbidae family of birds. One species, the Nicobar (Caloenas nicobarica) is the closest living relative to the dodo.
For this experiment, Turner along with Edward Wasserman, a University of Iowa professor, studied 24 of the type of pigeon you’d see in the wild, if the wild was the local parking lot at a strip mall. (Perhaps their love of hanging out around places where they’re prone to being run over might have something to do with their less-than-intelligent street smarts reputation.)
The birds were shown a stimulus – different shapes, and concentric and sectioned rings – and then had to peck a button that aligned with the category the stimulus belonged to. If correct, they received the reward of a food pellet. And there were no prizes for wrong answers.
Four tasks of varying skill levels showed that through trial and error, pigeons were quickly able to correct themselves to find the right answer, making associations with the stimulus and categories as they went along.
In the easiest experiments, the pigeons improved from an average 55% to almost all correct – 95% – while the harder tests were a little slower-learning, going from 55% to 68%.
But it was less about what they got right and more about the processes in which they were able to learn. The researchers believe the birds used associative learning, or simply linking two things together.
"Associative learning is frequently presumed to be far too primitive and rigid to explain complex visual categorization like what we saw the pigeons do," Turner said. "Pigeons don't try to make rules. They just use this brute force way of trial and error and associative learning and in some specific types of tasks that helps them perform better than humans."
Yep, for these tasks, the pigeon’s way of learning is more efficient than that of the human process, which tends to complicate a task with rules to try to make the job easier.
"But in this case, there were no rules that could help make this any easier,” said Turner. “That really frustrates humans and they often give up on tasks like this.”
The AI model used involved two simple mechanisms that the researchers presumed the pigeons to be using: associative learning and error correction. And much like the way the birds corrected themselves to get more and more answers correct with repeat experiments, AI made increasingly correct choices as it went along, too.
While a simple model, it’s nonetheless the foundation that AI is built on; these systems are out to find patterns and associate those to provide problem-solving linkages.
"The learning principles that guide the behaviors of these AI machines are pretty similar to what pigeons use,” Turner added.
While the experimental design had its limitations – such as just one type of associative learning model being measured – the researchers are buoyed by the results and will now collaborate with scientists who study the neurobiology of pigeons.
The study was published in the journey iScience.
Source: Ohio State University