Philosophical Foundations of Artificial Intelligence and Alignment Debates
Philosophical criticisms in the history of artificial intelligence, symbol manipulation, the lack of human experience, and the alignment problem are discussed.
From a report prepared for the RAND Corporation in 1965 to today's large language models, the philosophical and historical criticisms of artificial intelligence are examined.
Historical Philosophical Criticisms and the RAND Report
In 1965, the RAND Corporation asked philosopher Hubert Dreyfus to prepare a report on the possibilities of artificial intelligence. The resulting report criticized early symbolic artificial intelligence and argued that step-by-step computations based on formal rules cannot model human thought.
Symbol Manipulation and the Problem of Meaning
Large language models still face philosophical criticisms similar to Dreyfus's views and John Searle's Chinese Room argument. Artificial intelligence systems manipulate formal symbols without possessing an intrinsic understanding, and meaning is entirely attributed by humans.
Embodiment and the Lack of Human Experience
Artificial intelligence systems are far from physical coping and sensory experiences, namely embodiment, which form the basis of human understanding and intelligence. This situation is evaluated as one of the biggest obstacles preventing systems from truly comprehending the world.
The Alignment Problem and Security Approaches
Artificial intelligence does not have its own independent will or goals; fears regarding the alignment problem are entirely related to what humans want artificial intelligence to do. Regulation and alignment efforts are shaped around preventing dangerous physical capabilities and averting malicious purposes.