Artificial intelligence and a change in cyber-security paradigm
Artificial intelligence (AI, artificial intelligence), or more precisely speaking, algorithms of machine learning (ML, machine learning), rapidly pass from science-fiction to the commercial mainstream. Although the basics of math and algorithms (at least in terms of so much popular now neural network) there are many decades, only recently computational power allows you to make the most of their potential. The first large-scale cyber attacks that may interfere in artificial intelligence, are just ahead of us, however, awareness, inclusion of new threats and security is needed now.
Anyone no longer surprising voice Assistant on your Smartphone to help find a restaurant nearby, relevant recommendations on film, autonomous vehicles, which shall themselves decide how to avoid the dangers or algorithms that can help detect cancer on the basis of the results of diagnostic imaging. Soon it's artificial intelligence will determine credit risk, determine whether you will optimize our investment capital.
Intelligence are shrouded in mystery
In the heart of the artificial intelligence is, however, a dark secret. The algorithms take very accurate decisions. Can a huge degree of accuracy not only detect the cancer cells, but also determine how likely it is that the person will commit a criminal offence or even what her sexual orientation-based solely on the photos of that person.
Why is it secret? The result of the algorithm is accurate, but even its developers are not able to answer the question: why the algorithm did what he did? The algorithm correctly diagnosed the cancer, however, we have no idea on what basis did happen. Mathematical principles are well known, however, the decision-making process is a mathematical black box.
This is because the artificial intelligence learns by example. If, let's say – we want to teach artificial intelligence to recognize the cat, we don't need to develop its specific characteristics. Simply train the algorithm the tens, hundreds of thousands of photos of cats and cats, until millions of mathematical indicators inside the same algorithm they are aligned so as to distinguish the cat from non-cat. The result will be correct in most cases. What led to it? What is the internal representation of the qualities of a cat inside the algorithm? This, unfortunately, we do not know.
Cyber attacks on artificial intelligence
This is especially important due to cyber security. Now we see innocent attacks on artificial intelligence, where for example. advertising activates voice Assistant, to perform certain actions. The problem, however, much more serious when the black boxes of artificial intelligence will make business decisions. We must be prepared to change the paradigm of thinking about Government in the age of machine learning and to a whole new class of attacks involving the poisoning of consciousness of artificial intelligence by hackers.
Imagine that the attacker in this way, the preparuje data for the algorithm, to weaken his sensitivity, familiar with something that is an anomaly. What can be the effect? Carefree granted credits, a very low rate of insurance, autonomous vehicles, which confuse the red light with a green or same to steal, because the hacker has set a different place as a base. Think also about virtual assistants in our phones or homes, who – after interfering in the knowledge base by a hacker – silently begin to perform for us financial transactions or make purchases.
Why this change of paradigm? The same artificial intelligence algorithms are mathematical black box even for their creators, so detection and the fight against this type of attacks would be extremely difficult and probably will require a ... artificial intelligence that will take care of cybersecurity and to oversee business algorithms, alerting at the time when they make bizarre decisions.
New techniques to fend off attacks
Already at this time are carried out research on the use of special techniques involving a simulated triggering of attacks to the artificial intelligence algorithm to be more resistant to poisoning his consciousness by the hacker. The whole area is, however, a new and incredibly intense. None of the car manufacturers of standalone does not want to be this, which as a result in manipulation by hackers will no longer detect obstacles on the way. This is just one of the examples, because artificial intelligence comes in many areas of both the economy and our lives.
— Leszek Tasiemski is now responsible for Rapid Detection Center and product development related to the detection of attacks and vulnerabilities in company F-Secure. Has many years of experience in penetration testing and analysis powłamaniowych, mainly for the European financial sector. Fan uses artificial intelligence to detect cyber attacks. A graduate in computer science, economics, and an MBA.
Learn more about our offer
We use Google Cloud Translation and Gengo API’s to translate articles with exception of our comparative tests.