RSAC2021 Live: Bruce Schneier sounds the alarm, AI hackers are about to attack

Hacking in the age of AI can strike in a way that system designers could not have predicted or even imagined, disrupting the entire system or impacting a specific set of rules within it.

This will greatly increase the difficulty of risk detection. Even if discovered in time, it is difficult to understand how the hacker AI is on the offensive.

Bruce Schneier, a security technologist, researcher and lecturer at Harvard Kennedy School, said that artificial intelligence technology, which is currently in full swing, will not only bring huge benefits to the whole society, but also may trigger a new round of security impacts.

In his keynote speech at the 2021 RSA Conference on May 17, Schneier focused on AI risks. In Schneier’s view, hacking in the age of AI no longer needs to be pre-defined; instead, it will strike in a way that system designers can’t predict or even imagine, disrupting the overall system or attacking a specific set of rules within it .

Schneier explained, “Any system of rules can be hacked. No matter how thoughtfully we plan at the design stage, any set of rules is bound to have incompleteness or inconsistency that the designer did not expect or could not clearly define. part. And these small gaps will eventually be exploited by hackers and cause a complete system collapse.”

 Hacking AI and the interpretability problem

Schneier believes that future hacking attacks are likely to be initiated by some form of AI, which will greatly increase the difficulty of risk detection. And even if discovered in time, it is difficult to understand how the hacker AI is on the offensive.

This is the classic “interpretability” problem, uniquely described in the popular science fiction novel The Hitchhiker’s Guide to the Galaxy. In the book, a group of super-intelligent pan-dimensional creatures build the most powerful computer in the universe and name it “Deep Thought”, hoping it will answer the ultimate question about life, the universe and everything. And the answer given by the computer is 42.

“Deliberation cannot explain the answer, or even how it understands the question, which is the classic ‘interpretability’ problem,” Schneier said. The essence of modern AI is a black box: one input data, the other output answers. “

Schneier pointed out that researchers are working hard to develop explainable AI, but for several practical reasons, he does not believe that the field of AI will achieve substantial results in the short term. In his view, the interpretability problem of how AI works is rooted in how to express mathematical calculations in a form that conforms to human cognition, memory, and judgment habits.

He mentioned, “Forcing AI to make explanations that humans can understand is itself putting a shackle on AI, and it is likely to affect the quality of AI’s decision-making. Of course, in the near future, AI will become more and more The more opaque it is, the harder it is to explain.”

While hacking AI is still the stuff of science fiction, it is possible that it could appear in the real world.

Bruce Schneier

 Upcoming AI Hackers

So far, Schneier has yet to see a mature application of AI in large-scale hacking campaigns. But in any case, such a future is likely, and organizations must prepare for it in advance.

Schneier said, “Although hacking AI is still only the subject of science fiction, it does have the potential to appear in the real world.”

The hacks Schneier has observed so far are still manipulative and require a lot of expertise, creativity, time and even luck on the attacking side. But he warns that once AI systems gain the ability to perform hacking attacks, they will run faster and at scales far beyond human imagination.

He emphasized, “As AI systems become more and more powerful, human society will definitely hand over increasingly important decisions to them. And hacking attacks on such systems will inevitably lead to greater damage. By then, attacks will The event itself will also be supported and implemented by AI technology.”

  fend off AI hackers

Schneier emphasized the seriousness and urgency of the risk of AI hacking in the first half of his speech, but in the latter part, he also introduced the audience to the potential room for improvement in the field of cybersecurity.

Schneier said, “If future AI can find new vulnerabilities in computer code, it will be a powerful attack weapon in the hands of hackers on the one hand, and a protective shield on the other hand.”

Software vendors are likely to deploy more AI tools in the future to autonomously find bugs in their code and provide fixes. Schneier is also optimistic that, if realized, such tools would eliminate all software vulnerabilities we know of today.

Schneier concluded, “It is indeed easier to move passively into the future under the impetus of technology, but only when the entire human society actively decides the role that technology should play in the future era can we truly enjoy the benefits of technological development. Therefore, Before AI is fully pervasive and covers our entire world, human beings must first examine and predict with a clear mind.”

 

The Links:   LM150X08-TLB1 PM200CLA120

Related Posts