Although many previous studies have shown that pigeons are amazingly intelligent, with abilities ranging from primate-level counting skills to the ability to detect breast cancer on X-rays, scientists are still struggling to shake off the reputation of these birds as “creatures”. to challenge “slow mind”.
By placing pigeons in front of an artificial intelligence model and comparing their learning and problem solving methods, a new study shows that both the bird and the computer follow a similar process to solve the problems at hand.
Brandon Turnerlead author of the study and professor of psychology at Ohio State University Quoted from New Atlas It says the evidence found shows that the pigeon learning strategy mechanism is remarkably similar to the same principles that drive modern machine learning and artificial intelligence techniques. “Our findings suggest that the pigeon may have a highly efficient learning system that lacks the ability to generalize or infer like human learning,” Turner added.
In nature, there are about 300 species of pigeons and pigeons, all of which belong to the pigeon family.Columbidae) are. One of these species, the Nicobar pigeon (Caloenas nicobarica) in the pigeon family is the closest creature to the dodo bird (a flightless bird that became extinct in 1681).
Turner along Edward Wasserman, the professor of the University of Iowa measured and studied 24 pigeons of the type that you usually see in nature to conduct an experiment. The pigeons that live in nature are the ones that are always seen in the street. Perhaps this creature’s interest in exploring crowded places where it may be trampled is not unrelated to its reputation for mental retardation.
In the experiment, the birds were shown a stimulus consisting of different shapes, concentric and isolated rings. Then, the pigeons had to peck a button that lined up with the category belonging to the stimulus. If the answer was correct, the pigeon received food as a reward, and there was no reward for an incorrect answer.
Each of the pigeons had to solve four exercises with different skill levels; The researchers found that the pigeons quickly corrected their answers through trial and error to find the correct answer, associating stimuli and categories along the way.
Using this method to solve the easiest exercises, the pigeons’ learning and responding improved from 55% “average” to 95% “almost all correct”, while for the more difficult exercises the learning process was slightly slower and the figures from 55% It reached 68 percent.
However, the latter study placed less importance on correct answers and more on the processes through which the pigeons learned. The researchers believe that the birds used “associative learning,” or simply linking two things together, to respond.
According to Turner, the prevailing view is that associative learning is too rudimentary and inflexible to explain complex visual categorization, as seen in pigeons. Pigeons don’t try to make rules. They simply use blind trial and error and associative learning to solve certain types of exercises, and in this case they can perform even better than humans.
To answer such questions, the pigeon learning method is much more efficient than the human process. Instead of making things simple and easy, people tend to make things more complicated by making rules. “There was no law in this case that could have made it easier,” Turner added. “It’s really frustrating for humans, and they often stop doing these exercises.”
The AI model used in the experiment consisted of two simple mechanisms that the researchers believed pigeons use: associative learning and error correction. The pigeons corrected their errors in exactly the same way, so that by repeating the test, they would answer more exercises correctly. The artificial intelligence model also had more correct answers as it progressed.
Despite its simplicity, this model is the underlying foundation of artificial intelligence. These systems models rely on searching and finding patterns and relating them to each other to create problem-solving links. Turner notes that the principles of strategic learning used by AI machines are almost identical to those used by pigeons.
Despite all the limitations of the experiment design (for example, only measuring one type of associative learning model), the researchers are very pleased with the study’s findings and will now collaborate with scientists who study the neurobiology of pigeons.
The findings of the study in iScience Journal It has been published.