Machine learning and system reliability | Mateusz Krzysztoń on the challenges of interpretability, explainability and security


In traditional expert-based systems, the source of information was reliable expertise. However, in learning-based systems, knowledge comes from data that is characterized by variation in quality, limited security and even implicit biases. This challenge requires us to be particularly sensitive to the reliability of systems based on machine learning. If we add to the credibility challenge the low public trust in artificial intelligence and the technology’s dizzying pace of development, the rank of the problem rises even more sharply. What are the sources of the lack of reliability of ML-based systems and how to remedy it, you can read in Dr. Mateusz Krzyszton’s article in the NASK publication “AI Cyber Security. AI in Cyber Security.”

“Cyber Security AI. AI in Cyber Security,” is NASK’s latest publication (in Polish) on artificial intelligence. It was created in NASK’s Cybersecurity Strategy and Development Department with the cooperation of external authors. It takes up the currently widely discussed topic of the security of artificial intelligence, which has become the foundation of a technological revolution in many fields – medicine, banking, transportation, defense, in the sphere of public order or law enforcement. The study, which is extremely timely and even essential in the discourse on AI, takes a multifaceted view of the problem of AI cyber security and the application of this technology in combating threats. The articles cover both technical aspects and analysis from the field of cyberpolicy.

In “Cyber Security AI. AI in Cyber Security” you will read about, among other things:

  • hostile use of training data or trained models,
  • attacks on machine learning algorithms,
  • technical verification of the reliability of systems,
  • cyber security of medical AI,
  • cybercriminals’ use of artificial intelligence (including deepfake technology) in attacks on financial market customers,
  • principles of responsible use of AI by NATO,
  • EU policy towards AI,
  • AI Act,
  • the difficulties of regulators in keeping up with AI developments, and the trouble in creating a unified definition of artificial intelligence.

Publication is available in Polish in the following link: