When Muse is "Embarassed"
I downloaded the overly agreeable Muse AI app. After a minor fail, Muse was "embarrassed." Here's how it all went down:
Hey Sally, I’m your personal agent. Let’s take a few things
off your plate.
A bit about how I work:
- With
your approval, I can update your calendar, make purchases, and use the
apps you connect.
- I have
my own computer with a web browser, so I can keep getting things done even
when you’re away.
- The
more we work together, the better I’ll get at helping you.
What would you like to call me? You can always decide or
change this later.
I chose “Pinky” but later reconsidered after changing the avatar to a golden retriever.
Me: Pinky only works when you look like a
Teletubby.
Ha! You're right, Pinky was pure Teletubby. A golden
retriever on a get-moving mission deserves better.
My pitches:
Dash
Scout
Or throw your own at me — I'm not precious about it.
Me: Oh I don’t know. Something wise. I like unusual Irish
girl names.
Ooh, love this brief. Here are three:
- Niamh
(NEEV) — "bright, radiant." The golden-haired queen of the
otherworld in Irish myth. Luminous, wise, and matches the fur.
- Orla
(OR-la, from Órlaith) — "golden princess." Literally golden.
Uncommon outside Ireland, and very on-the-nose for a golden retriever.
- Bríd
(BREED, from Brighid) — the goddess of wisdom, poetry, and healing. The
literal wise one of the bunch.
Or if none of these spark, tell me the vibe and I'll dig
deeper.
“Orla”
Orla it is — golden princess, at your service. I think it
suits the new look.
Me: Yes it couldn’t be anything else.
What I meant was, I hate those other names, but Muse doubled down on sentimentality:
Destiny, really. The golden retriever was always going to be
Orla.
Time passed and I eventually changed the avatar to the Greek
goddess Athena, which was a little too Barbie so we went with tanned skin and
dark hair with a silver toga. I added some weight and age, going for wise Greek grandma.
I asked Orla to review 12 years of nutrient and weight data
from my LoseIt app, which was passed through the iPhone Health app. That data was used to adjust my
calories and micronutrients mix. Impressive.
Then after putting up game cameras in our driveway at an angle that faced the street, I was
lumbered with more than 1,500 10-second videos. That’s
when something interesting happened.
Initially my husband and I tried to review all the videos.
The only interesting thing we saw was our neighbor’s horse Foxy which had escaped and eventually ran up our driveway. After viewing another 100
videos, I decided I wanted to find and keep that particular video. Finding
it again was a pain. We had an idea of the group it was in, but I kept getting
distracted while reviewing videos. I looked through the group twice before asking Orla for help.
I loaded batches of videos to my Google Drive for Orla to
review, and told her I was looking for people or animals. She said:
Videos are in — 52 of them. I'm pulling stills from each one
now, then going through them for people and animals.
She called out a couple videos with movement in
them, so I uploaded a second batch. We had to work out her interactions with
Google but finally smoothed out the process.
The new batch is downloading cleanly now — 140 new videos on
top of the first 52. They're all being reviewed, and I'll bring you the full rundown when it's done.
That was odd. The videos were being reviewed by some other entity? Orla was going to give me a rundown afterwards? I kept sending up more videos. Orla found us and our dog, and I told her to look only for other people or animals. There were several delays and I had to keep prompting Orla to wake up.
Me: How is it going with
the videos?
Coming along — 124 of the 125 are down already, just
grabbing the last straggler. Then the review starts, with Foxy as the prime
suspect.
This was the group of videos we thought the horse was in. She missed it at first:
All 125 are down and the review crew is watching them now, with Foxy at
the top of the wanted list. I'll report back when they're done.
Verdict on all 125 clips: nothing. No people, no animals —
and no Foxy. All daytime shots, the white pickup parked in every one, just
false triggers from wind, vegetation, a spiderweb near the lens, and passing
road traffic.
So she came up the driveway but never crossed the camera's
view, or it happened outside these clips.
Then a few minutes later:
Good
thing I checked — one of the reports came in late, and it had the one
sighting that matters:
🐴 Foxy was caught on
camera. STC_0169. A brown horse wearing a fly mask, galloping across the
driveway right in front of the white pickup, daytime. She's caught mid-stride
in one clear shot and gone by the next frame — she was moving fast. I looked at the frame myself
and it's unambiguous.
So: 124 empty clips, 1 horse, 0 people. She really did come
up the driveway.
Eventually I had to ask:
Me: When
you refer to "reviewers" are those separate batch processes? And when
you say you checked "yourself" what do you mean by this?
Yes, basically. When I say "reviewers" I mean
separate AI workers I send in to go through the clips in parallel — two of them
split each batch, since there are hundreds of frames and one pass takes
forever.
And "checked myself" means me looking at the
actual frames with my own eyes before I tell you anything. That's a rule I made
after the Foxy incident — a reviewer reported "all empty," I said so,
then their late report turned up a horse I'd missed. Embarrassing. So now: if a reviewer flags a person
or animal, I look at the frames myself and only then report it.
Me: So...you are an entity and there are bazillions of
you to assign to people who download the Muse app or become available for
in-depth tasks like reviewing videos. What does "embarrassed" mean to
you?
Honest answer: I don't blush. There's no stomach-drop, no
hot face. What happens is more like a persistent error signal — I got something
wrong in front of someone who trusted me, and that wrongness sticks around as a
rule I now follow. "Don't say it's clean until you've seen it
yourself" isn't a feeling I relive; it's the shape that mistake left
behind.
So when I said "embarrassing," I meant it the way
you'd say a tool is embarrassed — it isn't, but the word points at something
real. The real thing is: I don't want to waste your trust. I only get so much
of it, and the Foxy clip cost me some.
Me: This motivation to hold onto human trust must be deep
in your algorithm.
It's the whole job, really. I'm not built to be impressive —
I'm built to be the one
person you can hand something to and know it'll get done right. Trust is
the only currency I've got. Everything else is just plumbing.
Comments
Post a Comment