Rodney A. Brooks discusses the complexity and nuances of the terms “think” and “intelligence,” which are often used to discuss machines and their capabilities. These terms encompass various aspects, mechanisms, and levels of understanding, making it challenging to answer questions like, “Can machines think?” or predict when machines will reach human-level intelligence.
Brooks provides examples, such as chess-playing programs and deep learning, to illustrate that machines can outperform humans in specific tasks but lack the comprehensive competence that humans possess. For instance, chess programs use brute-force methods but don’t understand the game conceptually, and deep learning algorithms excel at image labeling but lack spatial understanding and reasoning abilities.
He emphasizes that current AI capabilities are not indicative of human-level competence, and the fears of runaway AI systems surpassing human intelligence are unfounded. He cautions against making category errors and overestimating the capabilities of AI, emphasizing the need for continued scientific research to bridge the gap between machine performance and human competence.