AI's Limitations in Mimicking Human Thought Processes

In a recent essay published in The Verge, Benjamin Riley, founder of Cognitive Resonance, critiques the current AI boom, highlighting a fundamental misunderstanding: that language modeling equates to intelligence.

He emphasizes that contemporary neuroscience suggests human thinking operates largely independently of language, casting doubt on the assumption that increasingly sophisticated language models can achieve or exceed human intelligence.

Riley argues that while humans utilize language to communicate and formulate metaphors—essential for reasoning—AI lacks the capacity for dissatisfaction with existing metaphors or data, a crucial element of human cognition.

The essay references the notion that common sense represents a collection of 'dead metaphors,' implying that AI can merely rearrange these metaphors without genuinely innovating or conceptualizing new ideas.

For example, Riley points out that even when individuals lose their ability to use language, they can still demonstrate reasoning capabilities. He provides the instance of Einstein’s development of the theory of relativity, which stemmed not from extensive scientific research but rather from thought experiments born from his dissatisfaction with existing metaphors.

The report further highlights that AI's limitations are compounded by its reliance on internet data, which may not accurately represent the full spectrum of human languages. Specific metaphors found in languages not widely represented online, such as the unique descriptors for snow in various Inuit languages, remain beyond AI’s reach.

As Riley concludes, while AI can be useful, it cannot replicate the depth and richness of human intelligence, underscoring the need for a more accurate metaphor to describe its capabilities. Thus, this discussion not only critiques AI's current state but also raises vital inquiries regarding our understanding of consciousness and cognition.