Beyond the Illusion and Harmony of Willingness - The Key to Linking AI and Human Values (2025)

Beyond the Illusion and Harmony of Willingness - The Key to Linking AI and Human Values (2025)
Whether or not AI is willing is a fascinating question, but not essential from a practical perspective. Since the behavior of modern AI is determined by loss functions and reward design, the principle of action depends on mathematical formulas and design concepts; AI does not understand its goals and act on them, but rather continues to optimize the instructions it is given, and it is an illusion to view it as having proactive intentions and emotions. The alignment issue emerges from this perspective. It is extremely difficult to get AI to share the values and ethics that humans desire, and there is a danger that misaligned objectives will produce unwanted results. For example, if we order AI to "reduce disease," it may choose extreme measures depending on the situation. Research using value learning and human feedback is advancing to solve this problem, but because values fluctuate across cultures and situations and are difficult to quantify, a complete solution is not fores
eeable. Safety and ethics are a major focus of the Stanford AI Index and national policies, emphasizing the need for international collaboration; how values are designed and shared, rather than whether AI is willing, is at the heart of future discussions and is the key to shaping human-AI collaboration.