Real pigeons can solve diabolically difficult problems

‘One’s pigeon’ means an area of special concern, expertise or responsibility.

And for pigeons, their area of expertise is sorting.

Just as AI systems are trained from large data sets, the birds can use ‘brute force’, or repetitive trial and error – to identify and sort patterns and objects, quickly becoming highly proficient at their task.

Pigeons use the same method of associative learning as Artificial Intelligence. The learning approach is so similar that the University of Toronto’s Computer Science department uses pigeons as an analogy for machine learning.

And if you think Artificial Intelligence and machine learning is impressive, wait till you see what pigeons can do.

Ed Wasserman, Professor of Experimental Psychology at the University of Iowa, has studied pigeon intelligence for five decades.

He tells Cosmos that pigeon smarts are like alien intelligence here on Earth.

For a new study published in Current Biology, Wasserman wanted to test the limits of pigeons’ associative learning capabilities by testing them with a diabolically difficult ‘sectioned rings’ pattern sorting task.

Four pigeons (instead of Rock, Frillback, Tumbler and Homey of Real Pigeons fame, the ones in the study were numbered 33W, 58R, 14R and 51W), were set the task of sorting patterns – combinations of slanted and differently spaced lines – into seemingly arbitrary groups.

Pigeons ai associative learning visual
The pigeons mastered exhaustive, repetitive tests such as the one shown above. In the centre square are 16 sample stimuli out of the thousands the pigeons had to categorise. The stimuli were drawn from two different categories, shown on either side. / Credit: Ed Wasserman, University of Iowa

Each test pigeon was shown a stimulus and had to decide, by pecking a button on the right or on the left, to which category that stimulus belonged. The categories included line width, line angle, concentric rings, and sectioned rings. A correct answer yielded a tasty pellet; an incorrect response yielded nothing.

What made the test so demanding, Wasserman says, was its arbitrariness. No rules or logic would help decipher the task, something that would stump most humans.

After starting out at around 50% accuracy, the pigeons honed (or should that be homed) their skills through repetition, eventually achieving 68% accuracy.

Given all the hype around AI’s ability to perform basic tasks, the researchers think PI (pigeon intelligence) deserves more credit.

“Now a computer would get way more training than we’ve given the pigeons,” Wasserman says.

“They’re not Brainiacs. It isn’t that pigeons are so unbelievably smart. It’s probably more fair to say, they’re dedicated to deploying their mechanisms fully and persistently. And they wind up doing pretty well at the task,” Wasserman says.

“The [pigeons] capacities to discriminate, and to categorise stimuli is really quite remarkable. But what’s not as clear as it might be yet, is how they do it.”

Previous studies have shown pigeons can accurately categorise all manner of things – babies, trucks, fish, keys.

In 2015, researchers at the University of California even trained pigeons to reliably detect breast cancer, by accurately distinguishing between benign and malignant tumours.

In 2016, New Zealand researchers showed pigeons could differentiate between real four-letter words and four letters in random order (non words).

“We didn’t have to go to any Herculean measures to get the pigeons to learn. Because those little guys worked their tails off, so to speak. They’re happy to just keep plugging away.

“They’re workaholics. They didn’t quit. And they did learn,” Wasserman says.

Please login to favourite this article.