[salt-thrower]

I for one wouldn't assume that "Alexa, give me a challenge to do" would be pulling from the open, uncensored internet. I would assume that Amazon had some kind of curation method that would remove the possibility of Alexa telling you to do something lethally stupid.

Further, what Alexa did is actually even worse than just reading from the open internet, because it only took the "challenge" part and stripped away the context about how dangerous it is.

This is akin to something like, "Alexa, tell me about civil rights" and it quoting Stormfront or some other hate site's take on race, without any other context about what it's reading from.

[annoyingnoob]

When I ask for something really general, I assume that Alexa is just a voice controlled web browser. Since Alexa is really about advertising and selling you things, I would assume that any curation is just to drive you to Amazon products. Once you get outside of products that Amazon can sell you then Alexa breaks down to a basic web search - as evidenced here.

We are calling out Alexa here but I think the same is true for other voice assistants too. Calling these products 'smart' is so wrong - there is nothing 'smart' about them. That 'smart' marketing has you and nearly everyone else assuming that they are a lot more capable and functional than they really are. Its just voice commands to drive things you would otherwise touch in some way - it has no where near human abilities.

[hakfoo]

The assistant platforms fall back on web search because they want to pretend to be omniscent. If it said in the documentation "here are the 1,602 commands explicitly supported, the data sources it pulls from, and anything else goes to an error message", it wouldn't be magical anymore.

Less magical might be a far better bargain for the platform brands, though. It's literally the face (err, voice) of your brand, and you're fighting a completely adversarial war here. At best, you've got unrelated external data sources that could make it say something stupid, but there's an entire spectrum of bad actors with incentivization to get Alexa/Siri/Google Assistant to say something offensive, scandalous, or hazardous.