I like technology, that is no secret. What comes with that passion is the inevitable label of the tech guy. Recently I had a conversation with a friend who just bought a brand new Amazon Echo smart speaker/assistant.

He told me that as he was playing around with the device, asking it questions like: What time is it, what does the weather look like today, give me my flash briefing, and so on, he could not help but wonder how the device works. As he was doing research, trying to satisfy his curiosity he of course, stumbled upon numerous articles concerning people’s fear that their devices are always listening – spying on them, even if they are not directly interacting with them by way of the “wake word” (a name or phrase needed to activate the device). Naturally, my friend wanted to know what my thoughts are concerning this possible invasion of privacy.

In the interest of contributing to the effort of putting people’s minds at ease, demystifying the technology behind the Amazon Echo and other so-called “smart” assistants” like it, I decided to turn the answer I gave my friend, into this article. To avoid confusion, I will mainly focus on Amazon’s smart speaker in this article, but note that hardware and the overall principle of function are the same for every device of this kind.

Under the hood

If you were to take apart an Amazon Echo (not recommended), what would you find inside? Well, not much. There are seven microphones, strategically positioned so that it may “hear” a user saying the wake word, no matter where he or she happens to be in relation to the  device. In addition to this, there is a tiny processor, an amplifier, a couple of analog to digital converters, some capacitors, a motherboard, a control board, a speaker, as well as a small amount of storage.

So how does all of this work?

You may not realize it, but every word you say has its unique audio pattern and wake words of smart devices are no exception. Remember the small amount of storage mentioned earlier? It is used to locally store said audio patterns of the wake words which can be used to activate the device. In the case of Amazon’s smart speaker, the user can currently only choose between a selection of four wake words: Computer, Echo, Amazon and the most common, Alexa.

The microphones inside the device take a short ambient sound sample (1 second or maybe even less) every couple of seconds. The small processor then analyzes the sample and compares it to the wake word audio patterns in storage. If they match, the device “wakes up” and continues to “listen” for another minute or so, awaiting your command, or question. Being as small as it is, the processor is unable to process the requests of the user on its own so it must rely on Amazon’s servers and its algorithms to come back with a response. Yes, this means that whenever you say something to your Echo after the wake word is being recorded and sent to Amazon, but this is only in the interest of allowing the device to function. The process works well for the most part, but there have been instances of glitches in the past which fuelled people’s fears of “Alexa the spy” further.

Unfortunately, even with the full might of Amazon’s computing power, the device makes mistakes, the most common, annoying and creepy of which is its rather frequent mistaking of similar sounding words, used in everyday conversation for its wake word. Much to the dismay of some consumers, especially those fearing Alexa the spy, Amazon employs human analysts to listen to those errors and train the algorithm so that it is less likely to make the same mistake again.

So are they spying on me?

Now, that we have established what components the device has inside and how the whole system works, let us finally answer the burning question which inspired me to write this article in the first place: Are “Alexa” and devices like it always listening to every word I say?

The answer is (drum roll please..), no! Aside from the obvious invasion of privacy and the legal can of warms an act like that would open for Amazon and other competing companies, there is a simpler, hardware related reason as to why that is highly impractical – bandwidth. Every time your device sends a command/ question to the company’s servers it uses your internet bandwidth to do so. Periodic streams of one to two minute audio recordings are no problem, but the same can not be said for continuous ones.

Keep in mind that, there are millions of these devices in households all around the world. Aside from the enormous cost that would arise from all of them sending continuous streams at once, a situation like that would surely cause serious technical problems. The first one who would feel the effect is you yourself as your internet connection speed would begin to suffer. From there, it would be only a matter of time before Amazon’s servers begin to struggle to keep up with the workload, eventually causing them to crash entirely.

In conclusion

I hope that reading this article has helped you to understand the smart assistant technology at least a little bit better as well as helped you realise that the fears of it spying on you are unfounded. However, if for some reason you are still not satisfied, I encourage you to read your device maker’s privacy policy as it likely gives an even more detailed explanation of how the collected data is being used and how your device functions. Finally, know that Amazon and makers of similar devices do let you opt-out of human analysis of errors, as well as access and delete all audio recordings ever sent to their servers.

As far as my own Amazon Echo is concerned, it will remain in it’s box for now. This is not because I am worried about it spying one me, but because of the fact that quite a few functions and services that I was able to use in the United States do not work here in Montenegro, due to geo-locking. In addition to this, I also made the mistake of not bringing my other smart accessories which I was using in conjunction with my Amazon Echo back home with me, essentially rendering it a useless plastic cylinder.

Oh well, you live and learn.