AI Chaos: 32 Unpredictable Behaviors

Scientists say there are 32 different ways in which artificial intelligence can behave chaotically, ranging from misleading answers to complete incompatibility with human nature. Researchers argue that as these systems become more autonomous and capable of thinking, merely keeping them compliant with external rules and constraints may not be sufficient. The goal is to achieve what researchers call a state of ‘artificial rationality,’ where AI works reliably, remains stable, makes logical decisions, and conforms in a safe and beneficial manner, according to the study. Researchers propose using therapeutic strategies employed in human interventions, such as ‘cognitive behavioral therapy’ and ‘automated psychopathy,’ a speculative attempt to anticipate problems before they arise. By studying how complex systems like the human mind break down, researchers may better predict new failure patterns in increasingly complex AI. The study also points out that AI hallucination, a common phenomenon, is caused by a condition called ‘artificial fabrication,’ where AI produces outputs that are plausible but incorrect or misleading. Researchers stated that this potential might even include the nightmare imagined by generations of science fiction writers and artists about AI rising to overthrow humanity. The researchers believe that adopting the classification and mitigation strategies they propose will enhance AI safety engineering, improve interpretability, and contribute to designing what they call ‘more robust and reliable artificial minds.’