It all added up. The microhabitat: arid “PJ” woods in the “redrock” (sandstone) country of the Four Corners region; the fast and sweet song, proclaimed exuberantly even at 1pm; the plain gray bird with a long tail; and the spectrogram.
Come again? The spectrogram?
As recently as a decade ago, only a very small number of us were identifying birds by their spectrograms. Go back a decade earlier, and the practice was essentially unheard of. Today we’re all doing it. Well, a great many of us. Indeed, the newest recruits to birding are the ones who are going all out with ID by spectrography. I’m talking about the explosion in users of the Merlin Bird ID app.
Before we go any further, I want to be clear on something: What follows is not a diss on Merlin, not a diss on the people who use Merlin, and not a diss on the ornithologists and programmers who created Merlin. If you’re looking for the preceding, you may as well stop reading right now. At the same time, I hope to offer some guidance in the matter. I was an “early adopter.” I was analyzing spectrograms before Merlin was even a glint in its programmers’ eyes; I was poring over computer printouts of birdsong before some of the most enthusiastic users of Merlin were glints in their biological parents’ eyes. Alrighty, with that out of the way…
A couple of days before Erik and I were vireoing in that little canyon along County Road G, I was birding with other friends in another canyon nearby. On that occasion, we had gotten out of the “PJ” and were now in a stand of ponderosa pine with even some aspen mixed in. A vireo was singing in one of the aspens. The bird looked like this:
I’ve written about this before, so here’s just the barest of bare-bones treatments: The world around us is full of “signals,” some of which our human brains are interested in: for example, signals that reach our “receptors,” our eyes and ears, in the form of electronvolts and millipascals, respectively. Our brains then map variation in those signals onto our visual and auditory cortexes, respectively, and we experience the results of those mappings as seeing and hearing. That’s exactly how software like the Merlin Bird ID app and Seek by iNat work—minus the seeing and hearing part.
Don’t get tripped up on the supposed differences between the subjective experiences of hearing and seeing. We make sense of sound, largely without knowing we’re doing it, through the non-conscious process of something strikingly similar to spectrographic analysis. The spectrogram of that vireo’s song is absolutely a picture of the bird. It depicts a plumbeous vireo, not a gray vireo, because of, among other things: how often the phrases are delivered (faster in the gray vireo); how “buzzy,” or modulated, they are (plumbeous is the buzzy one); and the mean “frequency sweep” in the song elements (again, plumbeous for the win in that category).
Merlin assesses all of that, and then some, and tells you what you’re listening to. I ran Merlin on the vireo, and the app got it:
Shown here is output from the author’s phone–a screen-capture made in the field while the vireo was singing. The Merlin Bird ID app and the author agreed on all four bird species vocalizing at 10:43am on May 19, 2023.
The other birds—yellow warbler, yellow-breasted chat, and bushtit—were all out there, too. Indeed, you can see their spectrographic traces in that snippet of output. Merlin was batting 1.000. Then I did something sort of unfair to poor Merlin. I kept running the app. I gave Merlin the opportunity to overthink things. And Merlin took the bait:
The author kept the app running, and Merlin started to overthink the vireo. Could it have been a Cassin’s vireo instead? Maybe even a yellow-throated vireo. Although those are the wrong answers, the app’s mistakes are instructive for learning birdsong.
Those other birds, Cassin’s vireo and yellow-throated vireo, weren’t there. “First thought, best thought,” advised the American poet Allen Ginsberg (1926–1997). That’s good advice, much of the time, for humans, and that’s perhaps not bad advice for the machines, either.
It’s easy to get a chuckle at Merlin’s expense for suggesting Cassin’s and yellow-throated vireos as IDs for that plumbeous vireo. Because location. Because time of year. Because the proper “confusion species” there is the gray vireo. Poor Merlin is applying brute-force spectrographic analysis and missing all the context: landscape, microhabitat, our own human experiences in that place, and more. But hold on a second. Because that accusation of unthinking brute-force computing sounds awfully similar to many people’s take on computer chess a generation ago and, more recently, computer go.
Programs like Deep Blue in the 1990s and AlphaGo in the 2010s routinely made bafflingly weird moves, akin to calling a plumbeous vireo in a canyon in the Four Corners region a yellow-throated vireo. At the same time, those programs were making some shockingly brilliant moves. They soon were vanquishing the greatest human grandmasters. Even more eerily, the machines were discovering new tactics and strategies, especially in the case of go, that have significantly impacted how humans play against one another.
Maybe yellow-throated vireo wasn’t such a bad mistake after all. Spectrographically, that species and, even more so, Cassin’s vireo, are the closest matches to plumbeous vireo. Not gray vireo. Perhaps Merlin is already so good that it immediately dismissed gray vireo and went straight to the two other soundalikes. The chess and go experts came around, in some cases rather grudgingly so, to giving the machine output a second look. Doing so has made the human game richer.
Consider this take on Merlin’s “mistake” in the canyon: Initial determination, plumbeous vireo. Bingo. First thought, best thought. Second iteration, Cassin’s vireo, which sounds exceedingly similar to plumbeous vireo, and in some to many cases cannot be distinguished from plumbeous vireo. Third try, yellow-throated vireo, a burry, slow-singing vireo a lot more similar to plumbeous vireo than I suspect many of us realize. Years ago, when a yellow-throated vireo showed up in a canyon near my home in another part of Colorado, the bird paired with—and may have produced young with—a soundalike plumbeous vireo.
Put all of the preceding together, and I think we’re arriving at a more advanced take on vireosong than ever before. Like back in the days when Deep Blue and AlphaGo were making boneheaded blunders—while merrily on their way to trouncing the greatest grandmasters of all time. My suspicion is that Merlin is on its way to similar attainments. That’s cool. That’s intriguing. That’s disturbing for some—and thrilling for others. But none of that is the main point, as I see it.
The main point is that Merlin is a wonderful tool for learning about bird ID right now. Merlin can guide us to better birding, to a better “game,” if you will, even while the app is figuring out, increasingly on its own, how to surpass the birderly equivalents of Garry Kasparov and Lee Sedol. What’s so cool about Deep Blue (which defeated Kasparov) and AlphaGo (which bested Lee) is that they have reinvigorated the human game.
Sure, there are the birders who “just ask Merlin.” In the same way, there have always been birders who just consult the field trip leader or tour guide. Or who just match the bird in life to a picture in the book. But not most of us. We integrate those inputs, those authorities, those humans and books and apps, and arrive at a satisfying, and perhaps even definitive, view of the world around us.
I knew that vireo in the hot canyon was a gray vireo because Erik said so; and because other birders have talked to me about gray vireos through the years; because I’d read about the species in bird books and online; because I’d studied drawings and photos; and because I’ve inspected spectrograms and, yes, output from the Merlin Bird ID app; and, of course, because of my own experiences in the field with gray vireos for well over 30 years now.
Maybe Merlin will one day surpass every other authority: the field guides and websites, the trip leaders and tour companies, the spectrograms and digital photos, and more. Maybe that day will arrive sooner than any of us imagine. But I don’t think that’s going to kill birding and nature study—no more so than books and art, and teachers and tour guides, and cars and cameras have killed birding and nature study.
No question about it, Merlin is the shiny new object for birding in 2023. But, if I may mix my metaphors: This too shall pass. The post-Merlin era will arrive before too long. And you know what?—We’ll still be birding. Merlin will have a lasting impact. The “game” will change, much as chess and go have been reshaped by Deep Blue and AlphaGo. But, again, we’ll still be birding. Perhaps birding will become “better,” whatever that means. The bigger point is that birding will become different, and that’s the best thing of all.
Birders are so cool. Moments after Ted and Erik had documented a gray vireo in Montezuma Co., Colo., Erik presented Ted with a “Montezuma County 200” patch. Only problem is—Ted’s Montezuma County list is nowhere near 200. Which means additional visits to Montezuma County! Birding is changing rapidly at the present time, and artificial intelligence, including the Merlin Bird ID app, is central to all that change. But what we do is still birding, and it’s still as satisfying and as stimulating as ever. Selfie by Ted Floyd.
Ted Floyd is the longtime Editor of Birding magazine, and he is broadly involved in other programs and initiatives with the ABA. Ted has written 200+ magazine articles and 5 books, including How to Know the Birds (National Geographic, 2019). He is a frequent speaker at birding festivals and has served on several nonprofit boards. Join Ted here for his semimonthly spot, “How to Know the Birds,” celebrating common birds and the uncommonly interesting things they do.
Birding is a force for good in our society. Learning and sharing about birds translates into concern for birds and the environment, and the American Birding Association provides resources and community for all people interested in birds!