Models with verbally enunciated explanations: Towards safe, accountable, and trustworthy artificial intelligence
Paper i proceeding, 2024

In this position paper, we propose a new approach to artificial intelligence (AI), involving systems, abbreviated MOVEEs,
that are capable of generating a verbally enunciated explanation of their actions, such that
the explanation is also correct by construction. The possibility of obtaining a human-understandable, verbal
explanation of any action or decision taken by an AI system is highly desirable, and is becoming increasingly
important at this time when many AI systems operate as inscrutable black boxes. We describe the desirable
properties of the proposed systems, contrasting them with existing AI approaches. We also discuss limitations
and possible applications. While the discussion is mostly held in general terms, we also provide a specific
example of a completed system, as well as a few examples of ongoing and future work.

interpretability

artificial intelligence

accountability and safety

Författare

Mattias Wahde

Chalmers, Mekanik och maritima vetenskaper, Fordonsteknik och autonoma system

International Conference on Agents and Artificial Intelligence

21843589 (ISSN) 2184433X (eISSN)

Vol. 3 101-108
978-989-758-680-4 (ISBN)

16th International Conference on Agents and Artificial Intelligence
Rome, Italy,

Ämneskategorier

Annan data- och informationsvetenskap

Språkteknologi (språkvetenskaplig databehandling)

Filosofi

Datavetenskap (datalogi)

Styrkeområden

Transport

Fundament

Grundläggande vetenskaper

DOI

10.5220/0012307100003636

Mer information

Senast uppdaterat

2024-07-02