How a simple game of 20 questions could help make AI fit for the future
Artificial intelligence programs used to classify images could be trained much more cheaply using a surprisingly simple method inspired by the childhood game of 20 Questions, according to new research posted to the arXiv preprint server by a team from the University of Bristol.
A new study by researchers at the University of Bristol suggests that complex AI classification tasks can be accomplished using a straightforward method inspired by the childhood game of 20 Questions. By combining a series of simple binary classifiers, each trained quickly on a standard laptop, the researchers were able to demonstrate that these classifiers could perform intricate classification tasks that typically require massive computational resources.
Professor Sidharth Jaggi explained that even if a current AI program aims to differentiate millions of object types, the same could be achieved through random yes-or-no questions. The key insight is that while no complex coordination is necessary, a sufficient number of these basic classifiers is sufficient. This approach reduces computational costs, enhances reliability, and simplifies deployment across various devices, particularly in distributed or edge computing environments.
Dr. Ioannis Papageorgiou, the lead author, highlighted that the system remains dependable even if some classifiers produce errors, thanks to its independent answering mechanism. The research, published on the arXiv preprint server, forms part of the Informed AI research hub at the University of Bristol, aimed at advancing theoretical foundations for efficient, robust, and trustworthy AI systems.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
