Sometimes an idea is so delightfully stupid that you really, really want it to work. That was very much the case with Crowbar, a project from the Dutch startup Crowded Cities. What would Crowbar involve? you ask. Oh, nothing much. Just specially designed machines that would train urban crows to pick up cigarette butts in exchange for peanuts.
Much of the initial coverage of Crowbar focused on the smarts of crows themselves. The Next Web, which brought wider attention to the story in early October, noted, “Crows are currently ranked among the most intelligent species on the planet, with an encephalization quotient (fancy word for smarts) equal to that of chimpanzees.”* Among other things, we know that they can recognize and form gift-giving relationships with humans. And some crows, especially those raised in captivity, can show strong aptitudes for human-made technology.
In that light, it certainly seemed possible that the Skinnerian conditioning method described by Crowded Cities would work. Maybe crows really could learn that depositing discarded butts would yield a reward. And maybe—just maybe—the crows who figured it out really would somehow teach the trick to others. What a wonder that would be!
In support of all of this, many articles about Crowbar noted that it was based on the work of Joshua Klein—a man sometimes described as a “technologist.” Klein is arguably most famous for a 2008 TED talk in which he described a “vending machine” for crows. It was a charming idea, one you can try for yourself, if you’re curious enough: Today Klein advertises open source Crow Box kits online. The only trouble is that there’s little evidence it ever worked as promised, and even less that it would function in the way Crowded Cities claims it will.
“The whole premise is bogus,” Kevin McGowan, a crow researcher who teaches at the Cornell Ornithology Lab, told me over the phone. “My collaborator here, who works with me in Ithaca, she worked with [Klein] for a couple of weeks with the friendliest family of crows that existed on earth. They barely would take food off the box. It just doesn’t work. … Basically, to me, it’s the same as saying it would be a cool idea to graft wings onto sheep dogs so they could do better with rounding up sheep.”
(I reached out to Klein by email and he was unavailable to provide comment this week, but he suggested announcements would be coming on his website “in the next month or so.” In 2008, the New York Times Magazine published an article about Klein’s work that suggested his system had worked, but in April 2009, it issued a lengthy editor's note effectively retracting much of the piece.)
McGowan acknowledged that the system might be theoretically possible. “If you have [a crow] in the lab, they’re eminently trainable,” he said. And there is at least one video of crows seeming to use a similar kit. Nevertheless, McGowan still laid out a series of reasons why it would likely be impracticable with wild birds in an unregulated urban setting.
“One is that crows tend to be somewhat phobic of new things,” he says. “They really don’t like to go on to stuff that they aren’t familiar with. The other is that they have no interest in shiny objects or things like that. The motivation isn’t there.”
Motivation seems to be a sticking point, even for those who are less skeptical than McGowan. John Marzluff, a wildlife science professor at the University of Washington who has conducted important research on corvids, pointed out that if you want to convince crows to work for you, you might have to offer them something better than peanuts.
“The challenge is making sure the food rewards are ... always there and they’re very high quality,” Marzluff said. Crows “can find lots of other things in their environment. And they routinely shift between things.”
There is, after all, a reason that crows live among humans: Our cities are full of things that they like to eat, from fallen Cheetos outside the corner store to discarded leftovers in the trash. Any Skinnerian training system would have to regularly offer them a reward that is better and more readily available than the options they can easily locate on their own. As Marzluff puts it, “The reward of a peanut is not very high. If they can go to a dumpster and get a pork chop bone with some fat or meat on it, they’re going to take the pork chop over the peanuts.” McGowan takes a similar approach: “It’s just not worth it to go around looking for cigarette butts to find a peanut.”
Marzluff also suggests that the basic premise of the vending machines, where whole flocks of crows would scour the streets clean, might not line up with reality. “You’re likely to only have a pair of birds working a given area,” he said. “They’d keep others away, especially if it’s a food resource that can be defended like that vending machine.” Even if they didn’t, the Crowbar process (dropping a cigarette off in exchange for a reward) is complicated enough that other birds might not pick it up through observation. That kind of skill transference has, Marzluff told me, proven rare, even in strictly regulated experimental settings.
But if they were to figure it out, their vaunted intelligence might lead to other problems. Even if they somehow did take to the Crowbar system, what’s to stop them from finding ways to hack it? Crowded Cities proposes that the machines would include image recognition software, which would allow them to confirm that a cigarette butt has, in fact, been inserted. To ensure consistent rewards, however, it would presumably have to allow for a sizable margin of error. Clever as they are, some crows would, no doubt, learn to exploit the system, perhaps inserting twigs or rocks instead.
In one way or another, this is all to say that crows may be too smart—or at least too peculiar in their smarts—to embrace Crowded Cities’ project. As is the case with cephalopod intelligence, we’d do well to learn more about their intellectual otherness before we try to make them do our dirty work for us. Maybe if we did, they’d help us dream up a startup pitch that’s actually worth swinging at.
*Correction, Nov. 6, 2017: This sentence originally misidentified the source of the article as the Verge rather than the Next Web.