Don't diss the pigeons: How nature's algorithm rivals AI

The unassuming pigeon might not be held in high regard, but there are ways in which the pigeon makes decisions that will impress even its most ardent critics.
Written by Rajiv Rao, Contributing Writer
Pigeon in flight
Dethan Punalur/Getty Images

There is a scene in the Michael Winterbottom-directed film '24 Hour Party People' that charts the emergence of British New Wave music and also exemplifies our attitude toward pigeons.

Two young men head to the rooftop of a building, where hundreds of pigeons come to roost, and they proceed to shower the place with breadcrumbs. As the pigeons consume them and take flight, they suddenly start plummeting towards earth and dying.

Also: Five ways to use AI responsibly

The crumbs are filled with poison as are ostensibly the hearts and souls of the wayward youths. It's a funny scene, if a little morbid.


A carrier pigeon with a message attached to its leg is about to be released.

Golden Gate bird alliance

While many of the people who chuckled at the scene are probably not the kind of people who flirt with thoughts of mass murdering pigeons, it would be false to say the notion of dispensing with a handful of the birds now and then hasn't crossed some of our minds.

After all, these birds are habitually guilty of pooping right next to human habitation, in courtyards, stairwells, and pretty much everywhere else. Their feathers often stick to their waste, making an unholy mess. 

In other words, pigeons are considered vermin, the lowest of the low among birds. Pigeons are not usually described as 'majestic', like say, golden eagles, which soar on lofty currents while surveying prey. And pigeons are not lauded for their stunning plumage, like peacocks.

Also: The best AI chatbots: ChatGPT and other noteworthy alternatives

Boxing legend Mike Tyson may rear pigeons with great affection, but that doesn't stop the rest of the world from considering this maligned bird to be a 'flying rat' with no redeeming qualities. 

But what about if -- just like many of our prejudiced notions about each other and certain species in the animal kingdom -- this bias against the pigeon is grossly misplaced?

In fact, it turns out that the much-reviled bird is not only brainier than cats, but could be as smart as your three-year-old child.

Most impressively, however, is the fact that pigeons make choices that are very similar to the most-talked about technology today -- artificial intelligence (AI).

Tarred and feathered unfairly

In a study led by Brandon Turner, professor of psychology at Ohio State University, and Edward Wasserman, professor of psychology at University of Iowa, 24 pigeons were put to work slotting patterns into categories, as if they were training for a Mensa conference.

Some of the patterns were groups of lines -- with different thicknesses, placed and arranged in varying ways -- while others were circles. 

Also: Today's AI boom will amplify social problems if we don't act now

The patterns were accompanied by left and right buttons to indicate which family each visual belonged to. Correct answers were awarded a pellet treat while wrong choices got nothing.

In a feat that could fox many humans, pigeons were able to improve their scores from 55% accuracy to 95% for some of the easier assignments. For the more complicated tasks, these burbling birds were able to peck their way to a 68% score. 

As it happens, the method with which the birds went about their selection process mirrors AI's strategy when it's making choices -- a process you may recognize from when you taught your puppy to sit using treats. You sit, you get a treat; simple.

"The pigeons' demonstrated success seems to be based on their deploying a single, simple biological mechanism -- one that can be effectively emulated by a computational model involving just two free parameters and one hidden layer," said authors Turner and Wasserman.

"That biological mechanism is associative learning, which gradually connects behavioral responses to circumscribed regions of perceptual space via an error-correction process. In this sense, the pigeon's category learning prowess can be understood as if the pigeon were a machine," they added.

Also: AI and advanced applications are straining current technology infrastructures

Scientific thinking has maintained until now that associative learning is just too simple and primitive and lacking in the elevated cognition required to carry out complex visual categorization tasks.

In fact, it is the human that is an under-performer when confronting tasks like these, say the authors, because we tend to look for some definitive rules under which we can carry out a task, and then get bewildered and frustrated when they don't exist.

But the pigeon uses 'brute force' tactics to get the job done, much like AI does.

A winged machine

This ability is why pigeons can tell the difference between two human beings in the same photograph, and even between a Van Gogh painting and a Chagall.

In other words, the same technology -- which is helping cancer hospitals spot tumors and fire-fighting towers to differentiate smoke from clouds -- is one that is embedded in this unheralded feathered being.

We shouldn't be surprised. After all, Pigeon No. 498 -- who didn't merit a proper name, apparently (or even a prime number at the minimum) -- was launched by naval skipper Thomas Crisp to go and get help when he was torpedoed by a German U-boat in the First World War.

Also: Leadership alert: The dust will never settle and generative AI can help

Alas, Crisp did not survive, but No. 498 flew on a wing and a prayer, literally -- shrapnel had badly damaged one side of his body. Yet, the pigeon was able to deliver the message and help get the crew rescued. Thousand of other people were saved in similar fashion by the 100,000 or so pigeons that served in the War.

"We celebrate how smart we are that we designed artificial intelligence; at the same time we disparage pigeons as dim-witted animals," said Ohio State's Turner, emphasizing the irony that the learning principles used by pigeons and AI are almost identical.

"Nature has created an algorithm that is highly effective in learning very challenging tasks," added Iowa State's Wasserman. "Not necessarily with the greatest speed, but with great consistency."

Editorial standards