AI systems can exhibit capabilities that seem "smarter" than humans in certain domains due to several factors. For example, Lex Fridman explains that AI can engage in self-supervised learning, allowing it to process millions of hours of data independently, helping it learn to distinguish concepts with minimal human examples. This mimics human cognition to an extent, where AI can extract fundamental similarities from vast datasets that might be less intuitive for humans 1.
Moreover, Marc Andreessen discusses how AI's proficiency in tasks like image recognition surpasses human abilities because it can be trained on massive datasets aggregated from the internet. The sheer volume of data allows AI models to recognize patterns or objects more accurately than previous systems, making them highly efficient in areas like facial recognition and text analysis. These models thrive on large-scale data processing, something humans cannot match due to cognitive limits 2.
This expansive data capacity and innovative self-learning methodologies provide AI with an edge in processing and identifying information, which in isolated contexts, allows them to perform better than humans. However, as Robert Greene points out, AI currently lacks qualities like self-awareness, emotional processing, and holistic understanding, which are critical aspects of human intelligence 3.
These discussions reflect ongoing debates about the potential and limitations of AI in relation to human intelligence.