Tmall elves, small homes, millet AI speakers, Siri, Alexa… not only can be used for squatting, but also can be used to control furniture, and even if they are ordered, they can help people buy and buy online.
Alexa | Source: Digital Trends
So, the voice assistant is just like a personal butler, and only the owner can give it a command. No one will want others to sneak up with their voice assistant.
From identifying the owner’s voice to preventing the invasion of malicious programs, developers have tried to prevent the voice assistant from being stolen. However, it is hard to prevent, and studies have shown that voice assistants can also be controlled by laser.
If the voice assistant is placed next to the window, the person on the opposite floor will be able to open your door and send your money. Surprising, not unexpected.
In early November, researchers at the Japan University of Telecommunications and the University of Michigan said they found that the voice assistant’s input microphone can be controlled with a laser pointer or even a flashlight [1].
They opened the garage door in this way and started the car in the garage. The experimental distance of this experiment is 70 meters, and with a telephoto lens, the operating distance can be extended to 106 meters.
How does light pass information to the microphone?
Light, like sound, is a wave that travels in space. As long as it is a wave, it can deliver energy and information. For example, when watching an outdoor concert, the sun shines on the face, it will feel warm, and the subwoofer will be on the face, and it will feel embarrassing.
There is a membrane inside the microphone that vibrates when the sound waves come, and then converts the vibration into an electrical signal. This film also converts light waves into electrical signals.
Light can convey information, this is not difficult to understand. To give a very simple example, in the Korean movie “Parasite”,The father who killed the man hid in the basement, and the basement had a switch that could control the lights on the first floor.
He is looking for his son on the ground to read his information according to the Morse code switch light. The situation of inputting information to the microphone is slightly complicated, and it is necessary to simulate the light wave according to parameters such as the waveform and frequency of the sound wave input by the preset.
The researchers also said that all microphones involved in a secure speech recognition system need to be redesigned. Adding a layer of shading can not solve the problem, because the microphone of the experimental voice assistant can be light-controlled even if it has a dust cloth.
Among them, Kevin Fu, an associate professor of electrical engineering and computer science at the University of Michigan, said that the light-controlled microphone itself is too universal, and this discovery is equivalent to a big blow to the security of voice applications.
They have already told the big companies about this discovery, including Tesla, Ford, Amazon, Apple, Google. These companies said they are actively working to solve the problem.
In fact, the light control of the voice assistant is not the most meticulous. There are already many terrible situations in voice control itself.
Malicious instructions can be hidden in other voices
In 2016, researchers at the University of California at Berkeley and Georgetown University said they could hide instructions in white noise, allowing voice helpers to automatically turn into airplane mode or open a web page [2].
Last year, this technology was upgraded. The original team said that not only the white noise can hide the instructions, but also the normal music or voice can hide the instructions.
In other words, when the user thinks that he is listening to music, someone can manipulate the voice assistant without knowing it by hiding malicious instructions in the music. Among them, Nicholas Carlini, a Ph.D. student at the University of California at Berkeley, said the purpose of the experiment was to test how hidden such operations can become.
Malicious instructions can be hidden in frequencies outside the human hearing range.
In 2017, Princeton University and Zhejiang University jointly developed a voice recognition system using sounds outside the range of human hearing, and named this manipulation DolphinAttack [3].
In voice assistIn the application of the hand, this kind of operation will mute the voice assistant in advance, so that the user can not hear the confirmation or response from the voice assistant. Subsequently, researchers at the University of Illinois at Urbana-Champaign also conducted a demonstration experiment to control the speech recognition system with ultrasound at 7 meters. Although the dolphin sound attack can’t pass through the wall, it is not necessary to go through the window.
Malicious instructions can also be disguised as adults can’t understand the sound
In 2015, Georgetown University researchers published a paper that specifically described the differences between people and machines in understanding speech, and wrote how they used this difference to create machines that are understandable and incomprehensible. Voice commands, while suggesting that this difference is easily exploited by malicious people [4].
The title of their paper is “Cocaine Noodles: Using the Difference Between Human and Machine Speech Recognition”, “Cocaine Noodles” is a classic example of this difference, Google’s Assistant Software Google Now you can hear “Cocaine Noodles” as “OK, Google”.
In fact, in the face of something called randomness, the human malicious command itself is not so terrible.
DeepSpeech is one of the most comprehensive undergraduate systems recognized by the academic community, and its function is to convert voice into text.
In January 2018, researchers at the University of California at Berkeley, Nicholas Carlini and David Wagner, said they could deceive DeepSpeech 100% [5]. By making a small change to the original audio, as small as 0.1%, they can make DeepSpeech unable to recognize the original audio, or turn out some random text.
As for random words, I can only say that everything is possible, and the human brain can’t think of it any more. After all, in order to cope with the drama of life itself, people have already consumed a little of their imagination.
God laughs as soon as humans think. People always love to create things that they think they are clever, and God always makes people realize that they are small.
However, please don’t panic, the above experiments for controlling voice assistants are basically carried out under the most favorable conditions for malicious manipulation. In reality, such advantages are beneficial.Conditions are rare, and there are currently no signs that bad guys are doing bad things through these manipulations.
The major tech giants have long recognized this problem.
Amazon said that although it will not disclose specific security measures, it has been upgrading Echo to make it the safest. It’s good to not disclose measures, so hackers won’t be able to start. Google also said that Google Assistant will filter out instructions that are unrecognizable to the human ear.
Amazon Echo and Google Assistant only follow the owner’s orders, which means that Amazon Echo and Google Assistant will recognize the voice of the non-owner and refuse to follow the instructions.
Although the existing technology still has loopholes in identifying voices, such as YouTube, which has a bunch of people with similar voices to tease the voice assistant’s video, the existing technology can avoid most risks.
Apple also said that Apple’s smart speaker HomePod will refuse to execute some instructions, such as opening the door. Some privacy-related instructions, such as opening an album, opening certain apps, or websites, need to be executed if the iPhone or iPad is unlocked.
So, people still think about how to avoid being used by others. It is still very early or too long for people to be ruled by future robots.
A closer look at the concerns about people being ruled by the future robots, I found that these claims are basically made by the West. why? The reason is definitely not that robots originated from Western technology.
The concept of “master” in Judaism and Christian faith implies a hierarchy, claiming that only people have souls. Taoism and Buddhism advocate “harmony” and hope that all things in the world will coexist peacefully, and that all things in the world have an aura.
There are many movies and TV series in the Western countries, such as X-Men, Pacific Rim, Love, Death, and Robotics, and Super-Charging, which are all transmitting information, robots are better than humans, and more Suitable for survival in the future world.
In the Eastern countries, “Iron Tail Trainer”, “Astro Boy”, “Doraemon” and even “Altman” represented by Japan are all about the equal life of humans and robots. Share the story of emotions and sorrows (even if you don’t want to admit it, Japan is indeed the leading modern cultural exporter in Asian countries).