Date: 5/3/24 2:03 pm From: Dennis Paulson via Tweeters <tweeters...> Subject: Re: [Tweeters] Use of Merlin app
Stephen, thanks much for your great discussion of this topic. That’s pretty much how i see Merlin Sound ID. I think it can be very helpful in learning songs, and I hope few if any people are using it as a substitute for that.
But that makes me think about the new binoculars that are claimed to identify every bird for you—not that most of us can afford them. Will no one who has such binocs ever look in a field guide again, searching for that unknown bird and learning a lot as they search? AI is such a mixed blessing.
Dennis Paulson
Seattle
> On May 3, 2024, at 1:32 PM, Stephen Chase via Tweeters <tweeters...> wrote:
>
> Hi Tweeters,
> I'd like to offer a bit of a different perspective with Merlin. It's a bit long, so if you're interested in the topic, I ask that you bear with me a bit. It's specific to Merlin Sound ID.
>
> I would have no concern whatsoever with seeing an empid, having Merlin ping "Hammond's," and then reporting to eBird as Hammond's. We're quick to think of Merlin in human terms, like a mistake-prone birder, which it is not. As humans, we occasionally make mistakes with similar songs - likely even the same ones that Merlin mistakes. I'll be the first to admit that I've mistaken a flat Orange-crowned Warbler song, and probably a Chipping Sparrow song, for Dark-eyed Junco. That said, once I see the bird, I have no doubt. Orange-crowned Warblers, Chipping Sparrows, and Dark-eyed Juncos look nothing alike. I'd argue the reverse is true as well: Kenneth, in your case you clearly identified the bird as an empid. Hammond's and Dusky Flycatchers, while very similar by sight, have unique songs that are considered diagnostic for identification. In fact, many guidebooks argue that they are the only reliable way to differentiate between the two. Merlin did not even see an empid, and yet it, listening to the wide world of sound, immediately flagged hearing a Hammond's Flycatcher. I can't think of a better way to confirm Hammond's, especially if you did not (cannot) hear the song yourself.
>
> On here and on eBird, I see birders are adjusting to Merlin. Quite often on eBird, birders like to comment about mistakes Merlin made. I do the same thing. You know what that tells me? It tells me that both new and experienced birders are using Merlin - often! So am I. Since I finally got around to buying a spotting scope, I think no tool has had a more significant impact on my birding than Merlin - and undoubtedly for the better. Last week in Newhalem, a fellow birder and I were trying to figure out a quiet flycatcher that was behaving like a Dusky. Then it began calling. I know that my fellow birder knows Dusky's dry whit, and I like to pretend that I do too, and yet we both simultaneously whipped out our phones, turned Merlin Sound ID on, and yes - it immediately pinged Dusky for both of us. Woohoo!
>
> I often turn Merlin on while I'm walking along quiet trails or country roads, and virtually always when I'm parked and scoping a pond or field. I love to have Merlin running on the side. It provides a list of the birds I am hearing, allowing me to concentrate on scoping. It doesn't do well with road noise, and I find it tends to miss lower notes, especially American Bittern. But on the other hand, I find it remarkably good at picking up some species, particularly American Robin from even very long distances. Then as I'm birding, I can refer to the list Merlin pulls up when I'm inputting information into eBird, which I do at regular intervals while I'm scoping. If Merlin hears something I don't, I'll make a point of listening for that bird. I usually let Merlin run for no more than five minutes at a time before turning it off to input seen and heard data into my eBird checklist. If I let Merlin run too long, it can take a while to save the file, and it becomes a hassle to use that file later on if I need it to document a species. The biggest downside with Merlin is that it drains my phone's battery very quickly - and probably isn't good for the long-term health of my battery. Oh well - birding is good for my long-term health and Merlin makes birding even more fun!
>
> A challenge I like to play with Merlin is comparing what I hear with what Merlin hears. I'm blessed at this stage in my life to have better ears than Merlin does - I'm sure that will change as I get older. I usually hear and identify a bird well before Merlin does, especially when I'm walking, which causes a lot of white noise that affects Merlin's ability to keep up with me. Sometimes it beats me to a bird, and when it does, then I've got a target to try to find! I have found Merlin very reliable at differentiating between sounds I find challenging, in particular Townsend's and Black-throated Gray Warblers - certainly significantly more reliable than I am. When it pings one or the other, I'll make a point of looking closely for that bird. If I see it, great! If not, of course I won't add it to my checklist, except maybe as Townsend's/Black-throated Gray Warbler. As I bird, Merlin gives me frequent targets to find and focus on along the way, as it picks up species that I would have otherwise missed. In this way, it improves the accuracy of my checklists, reducing human error and helping me develop more thorough checklists that paint a more accurate picture of what really is in the area. Does that not significantly improve the quality of the eBird database?
>
> This morning <https://ebird.org/checklist/S171698023> while birding, Merlin beat me twice, as it picked up Mourning Dove and Solitary Sandpiper before I did. The Solitary ping was a flyby Solitary Sandpiper that I might have missed altogether. I did not catch the initial call myself, but Merlin did, and as soon as Merlin pinged it, I got excited, listened closely, and heard it myself. At first, I was a bit torn about whether or not to include it on my checklist. But I felt confident in doing so, even though it felt a bit lame to not get a nice look at the bird. Because Merlin allows me to save my recordings, I even had a good audio clip to back up the observation should anyone want confirmation. Thanks, Merlin! A few minutes later, I saw a (the?) Solitary Sandpiper in the wet spot I was walking towards: taller than a Spotted with greenish legs, a slight body bob once in a while - yep, that's a Solitary!
>
> To be clear, if Merlin had pinged the bird, and I did not see or hear it, I would not report it. As many have said, Merlin's IDs are suggestions. A few weeks ago, Preston Mui previously shared a link from Cornell that explains the science of Merlin. Merlin is AI - artificial intelligence. The link is here <https://www.macaulaylibrary.org/2021/06/22/behind-the-scenes-of-sound-id-in-merlin/?<doing_wp_cron...> if you missed it. I'm a high school teacher, and AI is very much at the forefront of my attention in my profession. The debate in school settings is what to do about AI: ban it or encourage it? In a school setting, we're talking about AI in the context of Large Language Models (LLMs) such as ChatGPT or Bing AI, but I think the principle still applies to Merlin. I absolutely recognize the concerns about AI - in a school setting cheating/plagiarism is the major concern. In birding, the concern is users inputting data based on what Merlin hears and IDs, not on what they hear and ID. On the other hand, the potential of AI in my profession is enormous. It's fantastic for improving differentiation for my students, and does a wonderful job of simplifying mundane teacher tasks like making schedules or turning lecture notes into a test. If used correctly, I think Merlin has incredible potential as well. Two examples in particular: 1) identifying unfamiliar sounds to help new and experienced birders to get onto birds, and 2) databasing audio clips of recorded birds for improved documentation.
>
> One buzz word in the AI world is "hallucination." Every once in a while ChatGPT will hallucinate - produce content that is clearly false or in some cases makes no sense whatsoever. When Merlin IDs Dark-eyed Juncos as Chipping Sparrows, that's an example of a hallucination. As you get to know AI products, you'll quickly see how hallucinations tend to fall into patterns. Orange-crowned Warblers, Chipping Sparrows, Dark-eyed Juncos all fit within those regular hallucination patterns that Merlin makes. I'm sure there are many other examples as well - Swainson's Thrush seems to be a regularly-hallucinated ID on Merlin too. Going back to Kenneth's dilemma, I'd argue that it's extremely unlikely that Merlin would hallucinate a Hammond's for a Dusky. Besides, if you still do feel uncomfortable about IDing the empid as Hammond's just because Merlin did, remember that if you saved the audio file, you can upload that clip to your Hammond's observation on your eBird checklist, documenting your sighting with an audio clip that others can review.
>
> If you made it this far, thanks for reading!
>
> In Everson,
> Stephen Chase
