Thank you for this information. Great advice!
Simpson County
Ed Leathers
On Thu, Apr 14, 2022 at 10:54 PM Mary Yandell <mtwyandell@xxxxxxxxx> wrote:
Graham Gerdeman shared this on the TN listserve and I thought it made some
really good points that others might find interesting.
Mary Yandell
Louisville
---------- Forwarded message ---------
From: Graham <grahamgerdeman@xxxxxxxxx>
Date: Thu, Apr 14, 2022 at 10:15 PM
Subject: [TN-Bird] Re: Observations about Merlin
To: tn-bird <tn-bird@xxxxxxxxxxxxx>
I would like to comment on David Stone's message re: Merlin, (and Kirk
Huffstater's response):
It's funny that this came up in this forum today, as the very topic was
one I was discussing at length this morning. In a nutshell, I agree 100%
with Kirk's comments. But I'd like to provide a little more personal
commentary.
First, with both the photo ID and sound ID, Merlin always provides
*suggestion*s, not answers. This seems to be the biggest source of
confusion with people using it. It absolutely relies on human input. It
provides a suggestion (usually 3-4), and you, the observer, must compare
photos and/or audio clips to what you heard - and determine what's correct,
if you can. It is merely steering you. So this is a fundamental
misunderstanding on the part of many.
Though Merlin is a Cornell / eBird product, it is the absolute* bane *of
every eBird reviewer. With a positive ID, Merlin prompts people to report
it to eBird. PLEASE do not ever write "identified by Merlin" in the
comments for a species which flags as rare. This drives eBird reviewers to
despair, and even if it is accepted now, I guarantee it will be dumped at
some point in the future. If it is an audio ID, go ahead and add the audio
to the checklist - those files are saved locally on your phone and can be
uploaded in a web browser. It's quite easy. I've had checklists in review
recently from people who are brand new birders literally using audio ID to
identify birds calling or singing at a stop light while they are driving -
and they simply accept the ID's and put in an ebird list. They don't know
any better - they have no idea what they are hearing. PLEASE DON'T DO THIS.
(why would you do this??). If you can visually or otherwise confirm the
ID, by all means, report it. Otherwise - just don't.
As it happens, I've been a Merlin beta tester since before it was released
to the public, so I've really put it through the paces, through multiple
versions, on multiple devices. I've seen it improve greatly - getting to
where I agree with Kirk's 50% average for the top suggestion - but
recently, it seems to have gotten much worse. So bad, I'm dumbfounded. Just
this morning, an obvious Eastern Bluebird was posted in a Facebook group,
with the comment that Merlin said it was an Eastern Phoebe. This is the
visual (photo) Merlin ID, not the audio. I had a hard time believing it, so
I took the same photo and ran it through Merlin myself. The first two
suggestions were Cedar Waxwing and Red-shouldered Hawk!! The THIRD
suggestion was Eastern Bluebird - so it did suggest the correct species -
but in what universe are those 3 birds going to be contenders for a
reasonably clear photo identification? My best explanation for why it's
gotten so bad so quickly is that so many users who have never opened a
field guide are using it to ID a bird they have no idea about, and simply
say "That's My Bird!" to the top suggestion (because they are
misunderstanding that "Merlin says it's this," rather than Merlin gave its
best guess) - thus they are training the AI with bad data. I don't know if
that's the reason - but it's the best I can come up with. Maybe the
database is simply overwhelmed with input and it's flaking out.
The audio ID is no more reliable than the photo ID. It's often really,
really bad. I am regularly testing the app in controlled circumstances, and
sometimes it's amazing - correctly identifying a chip note or a very faint
song. But - it also picks up birds that aren't there, misidentifies others,
and doesn't even hear others still. My favorite mis-ID I've experienced
with audio ID is the app thinking that the sound of roofers tearing off a
roof in my neighborhood was the ku-ku-ku- of a Yellow-billed Cuckoo.The
observer must use good judgment and their own skills to confirm the
suggestion.
Anyway, I'm certainly not trying to tell people not to use Merlin at all
(or any other similar app), but there are a lot of things for birders (
*especially* new birders) to keep in mind with this technology. It has
limitations, and relying on it has consequences. It is a massive crutch to
actual learning. I had the realization today that Merlin and other
automatic ID apps are for birding what GPS now is for driving. When you use
GPS to go everywhere, you get worse at actually knowing the way. I know
people who have lived in Nashville for 20 years and don't know how to get
from point A to point B in this city without the help of GPS navigation.
Looking up a bird in an actual field guide and piecing it together is
analogous to looking on a map and actually learning how the roads connect
and interact. More to the point, people take routes which aren't good -
some have famously even driven into lakes, or gone hours out of their way -
because the GPS told them to. It's incumbent on us all to study the map, or
the field guide, and learn to recognize when a suggestion is good, and when
it is a very bad idea. If we want to ever find our way without help, that
is.
Cheers - and good birding!
Graham Gerdeman
Nashville
grahamgerdeman.com
www.nashvillebirder.com
