Why Alexa isn't worth a $10B loss

Putting a value on things is always tricky, but sometimes things just click – or, in this case, they don’t.

I have an Alexa Echo Listen (the one without speakers) that I got during one of my trips to Seattle years ago.

It sits on my desk, hooked up to a cute passive mixer that takes all the audio inputs from everything I have on the desk and pipes them to a set of moderately decent speakers.

They have a good enough sub-woofer for me to zone out on music from server while I’m working, and I quite like this setup because all my devices benefit from it, and, most importantly, because I spent many years building up my music collection.

Living With The Echo echo echo

Muting the Echo during calls is already second nature, because it annoys the heck out of me to have it perk up occasionally, and it runs on a completely separate Amazon account from any of the family’s, because, well, I don’t trust it.

I’ve been meaning to replace it with a fully local solution for years, but I need it because it controls my lights, heater, and music better than Siri (or anything else) ever could and Open Source solutions are hopeless at doing a lot of stuff it takes for granted.

I also don’t have a spare iPhone with a dead battery I could use instead (which is a likely future scenario) – since, before you ask, Apple doesn’t sell HomePods in Portugal, so I’m kind of stuck with it.

Usefulness Is The Greater Part of Value

After all the noise about the Alexa division being on track to lose $10B this year, I’m not sure we’ll ever get any real competition on voice assistants – but today I had another reminder of their shortcomings:

“Alexa, ask Plex to shuffle the Work playlist.” (this works 99% of the time. However, today…)

“How about something new? Would you like to listen to something by “Various Artists”?”

”… No.”

“How about something new? Would you like to listen to “Time In Motion” by “Lo Fi Records”?”

“No. Alexa, ask Plex to shuffle the ‘Work’ playlist.”

“I’m sorry, I didn’t catch what you wanted to play. Try asking again.”

“No. Alexa, ask Plex to shuffle ‘Work’.”

“You seem to really like music released by Hollywood Records. Would you like to listen to “Guardians of the Galaxy OST/Awesome Mix by this label?”

“No, goddammit.”

“You seem to really like the Cowboy Bebop Original Motion Picture Soundtrack by The Seatbelts/Yoko Kanno. Would you like to hear it again?”

”… maybe?” (by this time, I was willing to compromise)

“You seem to really like Cowboy Bebop OST One by The Seatbelts/Yoko Kanno. Would you like to hear it again?”

“No.” (Yes, I tracked down the TV soundtrack too.)

“You were right in the middle of an episode of “Allo Allo”. Would you like me to resume it?”

”…” (that was three weeks ago, and it’s not music, it’s my ripped DVD collection)

(Boop) starts playing Cowboy Bebop, unprompted.

“Alexa, ask Plex to shuffle ‘Work’.”

“Shuffling the playlist ‘Work’.”

Daft Punk intensifies

The weird thing about this is that it was actually pretty straightforward:

  • It didn’t try to take me to Amazon Music (which is its default)
  • It didn’t just go and play Spotify (I have a free account attached to it, for when I want to check out new music by name)

No, it just utterly failed at doing the simplest possible thing while staying in context, which is nothing short of remarkable.

Now, this kind of thing isn’t rocket science – I have an itch to go, scrape my library and at least try to paste together something with, say, Rhasspy, but I already know it won’t scale1.

And yet, $10B apparently wasn’t enough to get it to work properly.

  1. I could also go and use Azure services, but that would be another cloud solution (albeit one I would have full control over), and I’m not sure I want to do it in the first place (although as far as intent detection goes, it’s pretty easy). ↩︎

This page is referenced in: