Online lecture: AI Responsibility Gaps

Daniela Vacek pôsobí na Filozofickom ústave SAV, v. v. i., v Kemplenovom inštitúte inteligentných technológií (KinIT) a na Filozofickej fakulte Univerzity Komenského v Bratislave. Je laureátkou ESET Science Award 2025 v kategórii Výnimočná osobnosť vedy na Slovensku do 35 rokov. Je slovenská filozofka špecializujúca sa na etiku umelej inteligencie, zodpovednosť, analytickú estetiku a filozofickú logiku. Vedie projekt APVV s názvom Filozofické a metodologické výzvy inteligentných technológií (TECHNE).

Jaroslav Kopčan pôsobí ako výskumný inžinier v Kemplenovom inštitúte inteligentných technológií (KinIT), kde sa špecializuje na spracovanie prirodzeného jazyka (NLP) a vysvetliteľnú umelú inteligenciu (XAI). Vo svojom výskume sa zameriava na automatizovanú analýzu obsahu a techniky vysvetliteľnosti pre menej zastúpené jazyky. Pracuje na vývoji interpretovateľných NLP systémov a nástrojov s dôrazom na destiláciu znalostí.

Date and Time: Thursday, 12 February 2026 | 10:00 CET

Venue: Online | Free participation 

The lecture will be conducted in English.

Registration

There is an extensive debate on responsibility gaps in artificial intelligence. These gaps correspond to situations of normative misalignment: someone ought to be responsible for what has occurred, yet no one actually is. They are traditionally considered to be rooted in a lack of adequate knowledge of how an artificial intelligence system arrived at its output, as well as in a lack of control over that output. Although many individuals involved in the development, production, deployment, and use of an AI system possess some degree of knowledge and control, none of them has the level of knowledge and control required to bear responsibility for the system’s good or bad outputs. To what extent is this lack of knowledge and control at the level of outputs present in contemporary AI systems?

From a technical perspective, relevant knowledge and control are often limited to the general properties of artificial intelligence systems rather than to specific outputs. Actors typically understand the system’s design, training processes, and overall patterns of behaviour, and they can influence system behaviour through design choices, training methods, and deployment constraints. However, they often lack insight into how a particular output is produced in a specific case and do not have reliable means of intervention at that level.

The lecture will offer several insights into these questions. In addition, we will show that the picture is even more complex. There are different forms of responsibility, each associated with distinct conditions that must be met. Accordingly, some forms of responsibility remain unproblematic even in the case of AI system outputs, while others prove to be more challenging.