Wissen um 11: Falsch rechnen – korrekt entscheiden: Warum KI keine perfekten Zahlen braucht

Lecture by Dr. Sallar Ahmadi-Pour

Just like computers and smartphones in the past, AI is increasingly becoming part of our everyday lives: from image and speech recognition to large language models such as ChatGPT and autonomous driving. While these systems are becoming ever more powerful and socially acceptable, their energy requirements are also growing.
This lecture is about a surprising idea from computer science and electrical engineering: instead of relying on perfect computing accuracy, AI systems can deliberately work inaccurately without us noticing. This is made possible by so-called approximate circuits, which deliberately allow errors in order to save energy and calculate faster. Easy-to-understand examples and demonstrations are used to show why AI forgives computational errors and why this property can be decisive for the future of energy-efficient AI.

💡"Wissen um 11", the science matinee in Bremen, in which exciting and current topics from science are presented every Saturday at 11 a.m. in thirty minutes and occasionally filmed and published on our YouTube channel. Free admission.

More information here: https://www.hausderwissenschaft.de/Falsch-rechnen-korrekt-entscheiden-Warum-KI-keine-perfekten-Zahlen-braucht-2.html