The difference is that the Alexa connects to your wifi, so if you wanted to, you could trivially tell if it's communicating when it shouldn't be. When I worked at Amazon, I was given the impression that the system that handles detecting the wake word was implemented in hardware, and the software system that does the real speech recognition doesn't "wake up" or gain access to the audio channel unless the wake word is detected by that hardware system -- and it's very obvious when you've woken it up (the colored ring lights up, it speaks, etc.)
Echo devices also sit in one room. If you're like most people you take your phone everywhere, which means that if it's spying on you, it could literally have a transcript of every word you spoke the entire day, as well as any people you've been around. To make matters worse, it would be difficult to tell if that was happening. Unless you're an uber-hacker who knows how to root an iPhone, or a radio geek who knows enough to monitor their device's cellular transmissions, good luck figuring out whether Siri is listening to and passing on audio that it shouldn't. The problem is that phones have so many apps and responsibilities -- given that they are essentially full computers -- these days that nonstop data transfer on a wifi network from my phone wouldn't be alarming: it might be backing up pictures to a cloud, or syncing the latest version of apps, etc.
I think the dedicated devices like Echo/Alexa are what you should buy if you're the most privacy-sensitive, since they have zero reason to be uploading to the Internet unless you're actively talking to them, and they have zero reason to be downloading unless they're receiving a software patch, which should be very rare. And because they're on your wifi (not cell) you can monitor their network traffic very easily.
As to the wake word point, I agree. I don't think alexa/siri/etc are currently bad or disrespecting privacy. I actually have a smart home with a voice assistant.
However, my smart home is all local mesh network (zwave and zigbee) based through a FOSS bridge that doesn't talk to the internet. All lights are through smart switches, not bulbs. The end result is such that if the voice assistant service ever pisses me off, I can simply disconnect from it.
If you read my comments in this article, I think I come off as a tin foil hat wearing lunatic, to some degree at least.
But actually, I'm not a super privacy paranoid person. Telemetry, voice recognition datasets, etc... I think those are a reasonable price to pay for free stuff! I just want to have my thumb on the scale and a go-bag packed for when/if these services become "evil" ;)
The final model is just a series of vectors that are typically pretty fast to run on the local machine.
You want to not suffer through political memes? And jump to scanning private messages for dissent, by authoritarian governments being okay!?
What?!
there's a good chance that if we continue to improve surveillance law enforcement agencies and justice departments could begin to focus on rehabilitation and growing a kinder world.
right they are like firemen trying to put out fires after the building has been ruined
if it doesn't work we're doomed anyway, so what's the problem?
the way the laws go good or bad is going to be up to everyone