Human Intelligence Is Surprising (twitter.com)

Recent discussions in the AI community highlight the surprising nature of human intelligence and its recent emergence in evolutionary terms. While modern AI models can now outperform human experts in various fields at an unprecedented pace, this has led to existential reflections on what truly distinguishes humans. Evolutionarily, human intelligence arose relatively recently and is not the result of a long-term optimization process; rather, it is a byproduct of survival, resulting in rapid environmental changes that exceed our ability to adapt through natural selection. This realization prompts a reconsideration of intelligence, underscoring that writing clearer definitions of intelligence can guide more effective AI development. The rise of machine learning emphasizes the importance of clearly defining objectives, as AI can rapidly excel in tasks when optimization targets are specified. The emergence of concepts like recursive self-improvement (RSI) demonstrates that the model development process is becoming more structured and identifiable, allowing for significant advancements in capability. However, this also poses a risk: defining intelligence carelessly may lead to unintended and potentially dangerous outcomes. Thus, as AI progresses, our responsibility lies in determining which goals to pursue and how to measure intelligence, reinforcing that we must be thoughtful in shaping future AI developments while remaining aware of the limitations of human intelligence itself.