I've enjoyed the stimulating conversation about the Merlin Bird ID app;
thanks to Paula Hansley for getting it going in the first place.
My take: By far the most valuable output from Merlin is the scrolling
spectrograms of birdsong; the ID suggestions are of distantly secondary
value. Here are a couple of case studies from earlier today, Sat., June 28:
First, at the Big Bluestem Loop Trail, Rocky Mountain Arsenal National
Wildlife Refuge, Adams Co., my companions with Denver Audubon and I were
enjoying a spectacular show by at least five sage thrashers, a species that
comes up with the dreaded "red dot" on the eBird app. So we needed proof.
Well, we took photos. :-) But I also quickly ran Merlin, and you can
totally see the herky-jerky phrasing in the spectrogram. Yes, "According to
Merlin," but the proof is in the pudding: the actual spectrogram, as
diagnostic as any photo of the species. Here ya go:
[image: real time output.png]
Now let's be honest. Reading a scrolling spectrogram in real time whilst
the bird is under observation isn't for everybody. That's totally fine.
Which brings up the second and, I would say, even more powerful case study
from earlier today. Before all the sage thrashers, as I was pulling up at
the refuge gate shortly before sunrise, I thought I heard a Brewer sparrow,
another "red dot" species. The song was distant, the prairie was still
pretty dark, and I needed an assist from Merlin. Sure enough, Merlin
reported Brewer sparrow. But was it really that species? I'd moved onto
something else, but I still had the audio because Merlin doesn't dump your
recordings; the user has to do that. So I clicked on the file; and then
clicked on the name of the bird; and, then, *voilà*, Merlin took me right
back to the exact point at which the bird started singing. And you can
totally see the diagnostic spectrogram:
[image: reviewed ouput.png]
I ran that diagnostic just a minute or so later, whilst in the field, but I
could have done it back home this evening—or even weeks or months later.
Merlin never forgets. (By the way, we documented an astounding 25 Brewer
sparrows at Big Bluestem; also 7 Cassin sparrows there.)
As an eBird reviewer, and on behalf of all the eBird reviewers in Colorado
and elsewhere, I'm practically begging you: Include the audio from Merlin!
Imagine getting killer photos of, oh I dunno, a yellow grosbeak in Larimer
Co., and simply stating: "Merlin." Same deal with relatively more prosaic
birds like singing Brewer sparrows and sage thrashers at The Arsenal at the
end of the third week of June. Upload the audio, and an eBird reviewer will
do the rest; for particularly tricky outputs, we'll appeal to Nathan
Pieplow to figure it all out... ;-)
A participant in this morning's field trip made an excellent analogy with
reading an x-ray. Most doctors would probably want to, you know, *actually
examine the image*, rather going solely off an AI's best guess that it was,
say, a transverse fracture of the distal radius.
The convo here at COBirds all started with the concern that Merlin had
misidentified a yellow-breasted chat as a lesser goldfinch. Did that really
happen? Perhaps. Both species are exceedingly protean in their utterances.
But we have the spectrograms!—really! we do! So we can go back to the
soundfile, click on the name of the species, get popped right back to the
presumptive lesser goldfinch, and see whether it was that species or
perhaps a yellow-breasted chat.
Two other matters have arisen in the course of this interesting discussion.
1. The accuracy of the Merlin Bird ID app. It is, in a nutshell, fantastic.
My sense is that we have crossed the "Garry Kasparov rubicon" with Merlin;
or, if your inclinations lie elsewhere, the "Lee Sedol rubicon." Merlin has
become that good. Here in Colorado, maybe Christian Nunes is still better,
but, as brilliant as Christian is, Merlin is learning *much faster* than
Christian, a mere human, can ever hope to. It won't even be close in five
years, perhaps not even one or two years. I firmly believe that we humans
will always derive immense satisfaction from recognizing birdsong without
AI assistance. But that's different from being better than AI. We're not.
Computers figured out how to beat the world's best chess players several
decades ago, and the world's best go players in the past decade;
identifying birdsong is, let's be honest, a comparatively trivial
undertaking.
2. Human sensory perception *vis-à-vis* electronic gadgetry. It is true
that our human ears and our smartphones' microphones are, in the broadest
sense, "the same thing": Both are, to be technical about
it, transducers—which detect sound energy, convert it to an electrical
signal, and transmit it to a receiver. But the analogy breaks down when we
compare what happens in the auditory cortexes in our brains with the
motherboards in our devices, with substantially different capacities for
processing the dynamical properties of frequency sensitivity, simultaneous
inputs in acoustically complex environments, and the subjective impressions
of amplitude, intonation, and modulation as they relate to the
interpretation of natural sounds. In plain English, what we notice (what
our brains think is "important") isn't the same as what the computer
notices.
Finally, here's something that the AIs don't yet have on us—and that I have
a hard time envisioning ever will: our deeply moving response to being on
the prairie at sunrise. The "mental" component of it—"that's a lark
bunting, told by its flight call, rising quickly through the 3kHz
band"—isn't really the point, is it? Rather, the entire experience: the
fresh morning air, the sunshine on our flesh, the twinkling spiderworts and
luminous globemallows; and most of all, the human camaraderie, taking it
all in, smiling together at a funny-looking beetle on the path, laughing
out loud at bison rolling in the dust, learning and sharing and wondering
together.
Ted Floyd
Lafayette, Boulder Co.
--
--
You received this message because you are subscribed to the Google Groups "Colorado Birds" group.
To post to this group, send email to <cobirds...>
For more options, visit this group at
http://groups.google.com/group/cobirds * All posts should be signed with the poster's full name and city. Include bird species and location in the subject line when appropriate.
* Join Colorado Field Ornithologists https://cobirds.org/membership/ ---
You received this message because you are subscribed to the Google Groups "Colorado Birds" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cobirds+<unsubscribe...>
To view this discussion visit https://groups.google.com/d/msgid/cobirds/CAGk944fN%3DZ1hPRm-gSh7K5g%<2BmiKsqenMejDyPPbEp3WWYaRSJQ...>
