Date: 4/21/26 11:10 am From: Robert O'Brien via Tweeters <tweeters...> Subject: Re: [Tweeters] The Merlin discussion
An interesting Merlin discussion. As it happens I have extensive
Merlin experience
in the last few months and some detailed discussions with others on OBOL.
I've been using it regularly around my rural property and, whatever its
limitations, it's great. Likely 10x more sensitive than my ageing ears,
even with hearing aids. An birding around here for 50 years, I know what
birds are around.
As to Bill's comments and more or less the same in part as Chug's:
*Merlin has undoubtedly helped increase the number of new or novice
birders, but is there any evidence that it has changed the ratio of
novice:skilled birders? I’ve not seen such evidence, and my subjective
assessment is that the wave of new birders has been simultaneous with a
significant increase in the number of proficient birders. If the ratio
has been relatively constant over the span of eBird years, then Merlin has
not increased the magnitude of error in the database.*
*I*n fact, Errors in Merlin submissions have recently led to a significant
change in the CURRENT version. There is a continuation of improvements,
some seem to download automatically, others not. But the current version
has this major changel Merlin has learned the 'expected' birds for many
areas, mine for instance. It has compiled such a list (I think I have at
my place SE of Portland, 109 allowed species) If it encounters a bird at
my place that is not on this list, it simply refuses to identify it
altogether and just continues recording. If I turn off the location
option in Merlin, it refuses to identify any sound. It will still record,
and you can see bird sounds go by on the screen, but none will be
identified.
As to no hint of the certainty of an ID, that is incorrect, as well. It
now has a system of varying colored 'dots' accompanying a bird ID on my
list, but unlikely to be present will receive a red dot. Migrants out of
migration times, for instane. A bird normally expected will receive a
different colored dot to indicate that. Etc. I was able to learn the
meaning of these 'dots' simply by Googling them.
One further point, as I stated above, I'm referring here to the *Current
Version*, which I recently downloaded. Merlin is not so much a 'tool' but
a work in progress. Cornell is working very hard to keep the improvements
coming and I personally believe that many of its problems of today will be
gone tomorrow, or pretty soon. But as a tool, it is quite amazing as it
stands, and will only get better. Although there are bumps on the road, I
think Cornell is doing an outstanding job with it.
As to the statement that it is created by humans and therefore has errors
as do humans. That was once undoubtedly true, and is still true. (Take
the refusal to identify birds not on the locality list, Seems better to
identify them (for the user) but simply not to submit them to eBird).
But, in the current world of AI, that general statement of human error, for
better or worse, is no longer true. Humans are involved, but AI will be
doing all the 'work' if it isn't already doing so. Sure, AI makes mistakes
as well, it is also a 'tool' of continuing improvement.
I could continue, but, as they say, *Nuf said, *at least for now.
Bob OBrien Portland
On Mon, Apr 20, 2026 at 9:40 PM Chuq Von Rospach via Tweeters <
<tweeters...> wrote:
>
>
> On Apr 20, 2026 at 13:33:08, Bill Tweit via Tweeters <
> <tweeters...> wrote:
>
>
>> Everybody seems to agree that Merlin is simply a tool, but then
>> discussions range well beyond that in terms of expectations and
>> limitations. Merlin’s inability to accurately discriminate between CHSP
>> and DEJU is a great opportunity to examine some of our expectations about
>> Merlin, and determine whether they are realistic. First, why do we expect
>> Merlin to be any better than a reasonably skilled human observer?
>>
> Bottom line, please do not expect Merlin to be perfect, even in North
>> America where it is quite good.
>>
>
> I don’t, but this is my biggest issue with Merlin. If I’m out birding with
> an expert ear birder, they will head and ID many birds I’ll never catch —
> but they can also explain what’s going on and note when they aren’t sure of
> an ID.
>
> Merlin issues every ID as if it’s perfect. “I hear a bird!” — and it’s
> this. I expect all of us on this list know to take that with a grain or
> three of salt, but we are not typical Merlin users. I looked at Merlin this
> afternoon at a random time and it told me there were 74,000 current users.
> I just checked again (at 9PM Washington time) and there are 18,000 users.
>
> Most of those users are newer/casual users. Statistically speaking, none
> of them are on this list, or a member of an audbon/alliance chapter, or go
> out on group outings with trained leaders that can help guide them. Merlin
> is their guide.
>
> And Merlin presents its results as correct and infallible (and yes, I’m
> reflecting back to my comments a few days ago on the need for a confidence
> factor on these IDs. If Merlin did that, it’d resolve pretty much all
> discussion on its accuracy and reliability).
>
>
> Second, why do we think that use of Merlin has increased the level of
>> error in the eBird data?
>>
>
> My take is that it hasn’t. Or, with a bit more nuance, I think tools like
> Merlin have actually reduced the percentage of errors recorded, but has
> also encouraged many more people to submit records, so I expect the total
> number of errors is up but the percentage of records that are incorrect has
> gone down. I have no data on this, but in my time working with and teaching
> newer birders, I’ve seen the trajectory from enthusiastically and
> energetically wrong to more thoughtful and skeptical many times (and god
> help me, I remember when I was a new baby birder and every freaking bird I
> saw was both rare and wrong. That others were willing to drag me into
> competence is one reason I’ve been willing to do the same in my pay forward
> times)
>
> And, let’s not forget that Merlin was built to assist with several
>> objectives, only one of which is helping birders in the field. It has
>> another, very significant, purpose: to increase awareness of birds among
>> the general population. Most of the Merlin users I encounter now are
>> people who take some joy in finding out what birds are present in their
>> yard, or on the favorite walking routes, or while they are camping, …
>> They aren’t contributing sightings to eBird, they just want to know more
>> about the world around them.
>>
>
> And it’s amazing at that — and anything that helps people find that spark
> that begins the journey into being bird people I’m all for. I just wish
> Merlin was willing to say “That’s a Robin” and “This might be a Chipping
> Sparrow, but I’m not positive because Juncos sound similar. What are you
> seeing?”
>
> My wish here is for Merlin to grow from presenting itself as the
> infallible expert into being that AND a mentor and teacher, by putting in
> some context when appropriate and leading people into being more skeptical
> and looking for more data (like a sighting) to verify the ID. Imagine if
> Merlin admitted it wasn’t sure and added a button “I saw it1” that people
> could click. They become part of the teaching that improved Merlin down the
> road…
>
> But I think it’s important that Merlin add that skepticism into how it
> presents itself, because of those 74,000 concurrent users this afternoon,
> almost all of them have no contact with the people (like those here on this
> list) that can do the mentoring and teaching to help them become more
> skeptical and thoughtful about Merlin’s results. Merlin needs to be
> responsible to set those expectations more realistically for the users that
> don’t yet know not to implicitly trust it.
>
> So, let’s be realistic about Merlin.
>>
> I’d say everyone on this list is, but my worry is about those new/novice
> users who are presented with a tool that gives no indication it might be
> wrong at time, and therefore has no reason to assume it’s sometimes giving
> them bad results. And an easy way to tweak Merlin is to add that confidence
> factor into the IDs, as tools like Haiku already have.
>
> Chuq
>
>
> ---------------------------------------
>
> Chuq Von Rospach (http://www.chuq.me)
> Silverdale, Washington
> Birder, Nature and Wildlife Photographer
>
> Email me at: <chuqvr...>
> Mastodon: @<chuqvr...>
>
> Stay Updated with what I'm doing: https://www.chuq.me/6fps/ > My latest e-book: https://www.chuq.me/ebooks >
> I have opinions
> _______________________________________________
> Tweeters mailing list
> <Tweeters...>
> http://mailman11.u.washington.edu/mailman/listinfo/tweeters >
