When Accidents Become Design Choices: Navigation Systems, Rat-Running, and AI Safety
-
Lorente, Antoni [1]
-
[1] Universitat Autònoma de Barcelona
Universitat Autònoma de Barcelona
Barcelona, España
-
- Localización: Audens: revista estudiantil d'anàlisi interdisciplinària, ISSN-e 2604-7284, Nº. 4, 2021 (Ejemplar dedicado a: Ciència, tecnologia i futur), págs. 37-54
- Idioma: inglés
- Enlaces
-
Resumen
Machine learning algorithms work particularly well if we want to find the best solution to a given problem from the set of all possible solutions. However, such an unprecedented ability to solve optimisation problems only stresses the need to carefully pick out the right goal to be optimised. In this regard, and taking route-planning services as a guiding example, I claim that the current problem definition for route-planning algorithms prompts disruptive driving practices such as intelligent rat-running which create, in turn, global problems by intending to optimise local ones. In order to avoid this, I defend that the design approach to such algorithms should aim for hybrid search strategies that constrain the local benefit to the global costs of a given solution, in order to set the grounds for a safer AI in the future.
