{"id":27817,"date":"2019-08-02T05:17:06","date_gmt":"2019-08-02T09:17:06","guid":{"rendered":"https:\/\/www.kaspersky.com\/blog\/?p=27817"},"modified":"2019-11-15T06:24:03","modified_gmt":"2019-11-15T11:24:03","slug":"smartphones-eavesdropping","status":"publish","type":"post","link":"https:\/\/www.kaspersky.com\/blog\/smartphones-eavesdropping\/27817\/","title":{"rendered":"Eavesdropping smartphones: Fact or fiction?"},"content":{"rendered":"

It\u2019s an oft-repeated tale: Someone talks with a friend about a certain thing, and then, bang, an ad for it appears on the smartphone screen.<\/p>\n

Some cases are easy to explain. A colleague of mine who was about to get married claims that she never googled wedding dresses, yet the pesky search engine started suggesting them all the same. And then, three months after the wedding, products for newborns began appearing in her search results, although she was not even planning to have children. There\u2019s no great mystery here, we can dispense with the conspiracy theories.<\/p>\n

But some cases arouse more suspicion. For example, you\u2019ve never had a barbecue before, one day you voice the idea, and, all of a sudden, you see an ad for barbecue equipment. Coincidence?<\/p>\n

After hearing such stories, or experiencing them personally, many people become convinced that the Internet giants are all listening to us through smartphone microphones. Fact or fiction? We\u2019ll discuss that a bit later, but first we suggest that you conduct your own high-tech experiment.<\/p>\n

Experiment: Tell your smartphone something it wouldn\u2019t expect<\/h2>\n

Here\u2019s the setup: Get together with some friends, place your phones on the table, and have a lively conversation about something that none of you have ever thought about before. Say: \u201cA fl00d c0ntr0l system? I\u2019ve always dreamed of buying a fl00d c0ntr0l system. Wouldn\u2019t it be great to find a cheap fl00d c0ntr0l system and buy it!\u201d The key words here are distorted on purpose, so that search engines don\u2019t think that you want to buy a certain kind of system only because you are reading this article (which, of course, mentions them).<\/p>\n

Don\u2019t get your smartphone hung up on a product that you might actually want. Be creative and come up with something you\u2019ll never want to buy in a million years. Maybe a w1gwam. Or a cap1barra.<\/p>\n

But there\u2019s one more condition: Do not under any circumstances search for the words you use, or the experiment will flop. That includes using voice assistants such as Siri, Alexa, Cortana, or Google Assistant. Speaking of Google, you can use this link<\/a> to see what it\u2019s harvested on you just today.<\/p>\n

After you and your friends have had a chinwag, keep mentioning the chosen word in passing for the next week. For example: \u201cHey, I was just thinking about getting a fl00d c0ntr0l system for the new w1gwam I want to buy\u2026\u201d<\/p>\n

My colleagues and I also conducted this experiment (see the results at the end of the post).<\/p>\n

Smartphone ads\u00a0\u2014 a magical coincidence?<\/h2>\n

Now for the fact-or-fiction question. There are several explanations for how the search giants sometimes hit a bull\u2019s-eye, that doesn\u2019t involve conspiracy theories about eavesdropping through the microphone.<\/p>\n

Explanation 1: Internet services build accurate models<\/h3>\n

In the article \u201cYour phone isn\u2019t really spying on your conversations\u00a0\u2014 the truth might be even creepier<\/a>,\u201d a former Google employee says that Google and Facebook essentially have a digital avatar of you that tries to replicate your behavior with the aid of machine-learning methods. At some point, your digital stand-in becomes so much like you that it begins to predict what you want.<\/p>\n

<a href=\"https:\/\/vas3k.com\/blog\/machine_learning\/\" target=\"_blank\" rel=\"noopener\">Source<\/a>. Warning: After you read this article, Google may start recommending meatbags to you<\/a>

Source<\/a>. Warning: After you read this article, Google may start recommending meatbags to you<\/p><\/div>\n

About Facebook, I heard it can determine whether a woman is pregnant even before she herself knows\u00a0\u2014 by the rate she scrolls through the social network\u2019s feed.<\/p>\n

Personally, I\u2019m disinclined to believe in such telepathic abilities, but there is definitely some truth to this one. Machine learning is really hitting new heights every single year. By the way, our behavioral model<\/a>, one of our threat detection methods, operates along similar lines. The basic idea is that if it quacks like a duck, it probably is a duck \u2014 that is, if a suspicious file behaves like a known piece of malware, then it\u2019s probably malicious.<\/p>\n

Explanation 2: Accidental activation of the voice assistant<\/h3>\n

Another explanation for the search giants\u2019 omniscience is the random activation of voice assistants. Sometimes, the phone thinks you said \u201cAlexa,\u201d \u201cOk Google,\u201d \u201cHey Siri,\u201d or \u201cCortana,\u201d when in fact you didn\u2019t.<\/p>\n

It\u2019s not necessary to pronounce these trigger words properly.\u00a0Something similar is likely to produce the same effect and turn on the assistant \u2014 after which the phone really does listen in on everything, and then starts making relevant suggestions.<\/p>\n

\"Eavesdropping<\/a>

Eavesdropping smart speakers<\/p><\/div>\n

In standby mode, even stand-alone voice assistants (such as those in smart speakers) do not constantly capture your speech\u00a0\u2014 they wait to be addressed. So as to catch the switch-on command, the device uses a small buffer (a few seconds\u2019 worth of audio storage), a dedicated (and relatively weak) processor, and a speech recognition algorithm tuned to a specific word.<\/p>\n

That\u2019s all running all of the time, but it consumes little power and no Internet traffic at all. Only when it detects the trigger word does the device wake up fully, connect to the server, and transfer recorded data for recognition.<\/p>\n

\"By<\/a>

By the way, the assistant in your smartphone sees what\u2019s on the screen. You don\u2019t have to talk about it, just let it read<\/p><\/div>\n

Explanation 3: User domains<\/h3>\n

Let\u2019s say you talked with your friend or spouse about deodorant or the need for some (hopefully not), after which you did not<\/em> start searching for it. But they did. If you see an ad for deodorant after the conversation, don\u2019t be too surprised.<\/p>\n

The thing is that the Internet service already suspects that your two accounts are somehow linked if you often hang out in the same place, perhaps on the same Wi-Fi network, maybe even take turns logging in to the same device. Such users may get put into \u201cdomains\u201d by search engines, whereupon they\u2019re shown some of the same products because they may make purchasing decisions together.<\/p>\n

It\u2019s impossible to say for sure that this happens, but such actions on the part of Internet services would be logical.<\/p>\n

Explanation 4: Lucky guesses<\/h3>\n

I often see advertising that doesn\u2019t match my interests: sauna equipment, pregnancy tests, trips to wherever \u2014 you name it. You\u2019ve probably seen similar stuff.<\/p>\n

But many people use search engines, so it\u2019s possible that a random ad blitz might one day catch you after you\u2019ve been talking to a friend about the benefits of saunas. And then you post online that your phone is spying on you. Meanwhile, folks who saw the same ad, but didn\u2019t discuss saunas, write nothing. One story is seen, and the other one is never even told.<\/p>\n

Such coincidences are nowhere near as surprising as they might seem. Here\u2019s an example. Eight teams get to the quarterfinals of some championship. There are 4 + 2 + 1 = 7 matches left to play in the tournament. Each match can have two outcomes \u2014 either the first team or the second wins (there are no draws). The total number of possible scenarios is two to the power of seven\u00a0\u2014 2^7 \u2014 128.<\/p>\n

Suppose there is a residential building containing 128 apartments. If we place one prediction for each possible outcome in each apartment\u2019s mailbox, it\u2019s a fact that we will supply the owner of one of the apartments with a 100% accurate prediction. He will be astounded, but in reality, it\u2019s just that our sample was sufficiently large.<\/p>\n

How not to do the experiment<\/h2>\n

One vlogger conducted a similar, yet very different experiment. Broadcasting live on YouTube, he deliberately started talking about dog toys. He then demonstrated that Google ads had adapted in a matter of seconds.<\/p>\n