Zhou Hongyi's World Intelligence Conference Reopens the Golden Sentence: Artificial Intelligence Has Its Security Problem


Zhou Hongyi's World Intelligence Conference Reopens the Golden Sentence: Artificial Intelligence Has Its Security Problem

Entrepreneurs News May 16 news, the second World Smart Conference held in Tianjin today, 360 Group Chairman and CEO Zhou Hongyi was invited to deliver a keynote speech.

Zhou Hongjun pointed out that artificial intelligence itself has security problems and we must be cautious.

Zhou Hongyi's World Intelligence Conference Reopens the Golden Sentence: Artificial Intelligence Has Its Security Problem

Zhonghong Zhou speaks at the 2nd World Smart Conference


Zhou Hongyi believes that the problems of artificial intelligence itself are divided into the following aspects.

The first is the unattended system we mentioned above. Zhou Hongyi believes that they all have the risk of security holes. Once they are hijacked by the network, they will bring serious security problems.

For example, weapons like machine guns in the past were controlled by soldiers, and were judged and pulled by people. Once the artificial intelligence technology is introduced more and more in the military, if the artificial machine intelligence is used to make a smart machine gun in the future, it can find the target and shoot it by itself. If such a system is inadvertently controlled by hackers some day, he can remotely control the shooting of weapons and indiscriminately kill them. The consequences are terrifying.

Second, the sensors, training data, and open source software used by artificial intelligence may have security risks. For example, sensors can be disturbed and the system itself can be deceived or invaded.

The example of Zhou Hongxuan stated that the 360 ​​security team had used ultrasonic interference technology to successfully decipher Tesla in tests, and convinced it that obstacles in front of it did not exist. The 360 ​​security team also entered the Tesla Hall of Fame because of this loophole.

For another example, in recent years there have been dolphin-tone attacks against smart devices. Hackers generate meaningless ultrasonic signals that can be heard by mobile phones, smart homes, and smart cars. After receiving the voice control system of the smart device, the hacker can use the hardware loophole of the microphone to automatically demodulate and restore the original voice command, and can silently turn on the voice assistant of the smart phone, smart home, and smart car, and then randomly make calls and send. SMS, video calls, or switching devices to flight mode, etc.

Third, there is a problem with artificial intelligence, which is that today’s artificial intelligence is still just a desperate way to train with large amounts of data.The probability judgment system that comes out is far from perfect.

Zhou Hongyi said that artificial intelligence may have a 99.99% probability to ensure that the identification is correct, but for security, it will only have one recognition error, it will have serious consequences.

After a while ago, Uber's modified automatic driving test vehicle killed a woman in the United States, fully demonstrating that today's artificial intelligence technology is not a complete system. Uber’s car was in a dimly lit environment. The lady wore black clothes. In this very complicated situation, the car’s artificial intelligence did not make any judgments or even slow down. tragedy.

Fourthly, most current artificial intelligence systems use open source software and architecture. There are many security holes in this area.

The artificial intelligence is realized by the software code, it is the operational logic that passes the data training, and now the scale is getting bigger and bigger. Everyone is using a smart phone, code has more than 100 million lines, not to mention complex systems such as smart cities, nuclear power plants. These codes are written by people. As long as they are written by people, there may be loopholes.

"Over the past few years, our artificial intelligence security team has conducted some research on the most popular deep learning framework and discovered security holes in dozens of open source systems," said Zhou Hongyi.

At last year’s China Internet Security Conference, 360 Live demonstrated the world’s first example of remote control by attacking deep learning applications. It touched everyone and caused many people’s Concerns about its own security issues.

Zhou Hongqi emphasized in his speech that the application of intelligent technology will break the boundaries between the virtual world and the real world. Attacks on cyberspace can directly penetrate the virtual space, map to physical space security, and generate physical losses and physics. hurt. Therefore, Zhou Hongyi believes that cyber security threats have expanded from cyberspace to national security, national defense security, critical infrastructure security, social security, urban security, and even personal safety. It can be said that with the development of smart economy, cyber security has entered the “great security” era. When hot vocabularies such as artificial intelligence and smart economy begin to attract attention, we need to be prepared for peace and security in order to protect the security of cyberspace and the physical world.

This article comes from entrepreneurs. Entrepreneurs are authorized to publish, slightly modified by the editor, and the copyright belongs to the author.Rong represents only the independent viewpoint of the author.