Do the Thinking Models Actually Think?
10 days ago
- #Machine Learning
- #Philosophy of Mind
- #Artificial Intelligence
- The debate on whether machines can think dates back to the 1950s, with contributions from Turing, von Neumann, and Shannon.
- Modern LLMs like ChatGPT imitate humans so well that users speculate about their true intelligence, though verifying consciousness remains a philosophical challenge.
- LLMs may function like advanced autocomplete, optimized to predict the next token without inherent logic or world ontology.
- Human intelligence is top-to-bottom (ideas to symbols), while LLMs are bottom-to-top (symbols to ideas), suggesting different forms of intelligence.
- A sufficiently advanced bottom-to-top thinker could theoretically simulate human intelligence, but practical replacement of humans by AI is not imminent.
- The future lies in collaboration between human and AI intelligence, leveraging their complementary strengths.