The debate over understanding in AI’s large language models
By Melanie Mitchell, David C. Krakauer. Proceedings of the National Academy of Sciences of the United States of America, 2022
386 citations recorded by the source corpus.
We survey a current, heated debate in the artificial intelligence (AI) research community on whether large pretrained language models can be said to understand language—and the physical and social situations language encodes—in any humanlike sense. We describe arguments that have been made for and against such understanding and key questions for the broader sciences of intelligence that have arisen in light of these arguments. We contend that an extended science of intelligence can be developed that will provide insight into distinct modes of understanding, their strengths and limitations, and the challenge of integrating diverse forms of cognition.
Subfields: Artificial Intelligence, Cultural Studies, Experimental and Cognitive Psychology
View Paper: https://www.semanticscholar.org/paper/e32185936ab3b23f39b1dd93e1507e6d80a71776
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
Citations in the corpus
- GPT-4 passes the bar exam (2024, background): “While our understanding of LLMs is still nascent 61,62,63 and we do not fully understand why GPT-4 performs differently by substantive topic, it is likely that the prospect of future performance improvements will, in part, hinge on identifying opportunities for additional topically relevant data,…”
- Artificial intelligence and illusions of understanding in scientific research (2024, neutral): “Although the question of what AI models can ‘understand’ remains controversial 14 , the AI visions (alongside the language of ‘deep learning’, ‘deep neural networks’ and marketing hype 95 ) portray AI as having a depth of understanding beyond what can be grasped by limited human minds.”