Geoffrey Hinton has argued:
“It’s very clear that we understand language in much the same way as these large language models.”
Source: Hinton keynote comments reported by R&D World
https://www.rdworldonline.com/hinton-ai4-conference.../
And: “They really do understand. And they understand the same way that we do.” — Geoffrey Hinton, quoted in The Globe and Mail, 2024.
https://nextgenedition.com/for-geoffrey-hinton-the.../
I think this is wrong.
Related:
1/ https://www.facebook.com/share/p/1B6ZGSiggL/ (understanding itself resides in the humans who created the training corpus and in the humans interpreting the output. The AI model acts as an intermediary, not as a subject possessing comprehension, AI consciousness attribution may arise from “transference”—the projection of human qualities onto a machine.)
2/ https://www.facebook.com/share/p/1CYm9Bnfgm/ (syntactic fluency and statistical coherence are not the same as semantic understanding)
3/ https://www.facebook.com/share/p/18nJcpyh26/ (if AI understands and is conscious it is legally liable for consequences of its responses)
4/ https://www.facebook.com/share/p/17WfGJciTN/ (Ted Chiang article in Atlantic)
LLMs can model language impressively, but modelling language is not the same as consciously understanding meaning.
Human understanding involves awareness, relevance, embodied experience, and the inner recognition that something has been grasped. Present LLMs show no evidence of that subjective transition.
They process patterns. We supply the meaning.
That makes them powerful tools — but not understanding minds.
The core mistake is confusing fluent performance with possessed comprehension.
———
Cc: Ernest Davis Jelel Ezzine Bernard W. Kobes Darius Burschka Amitā Kapoor Amit Sheth Anil Seth William Hsu Moshe Vardi Mark Coeckelbergh Francesca Rossi Ronald Cicurel Patrick Lin Luis Lamb
“It’s very clear that we understand language in much the same way as these large language models.”
Source: Hinton keynote comments reported by R&D World
https://www.rdworldonline.com/hinton-ai4-conference.../
And: “They really do understand. And they understand the same way that we do.” — Geoffrey Hinton, quoted in The Globe and Mail, 2024.
https://nextgenedition.com/for-geoffrey-hinton-the.../
I think this is wrong.
Related:
1/ https://www.facebook.com/share/p/1B6ZGSiggL/ (understanding itself resides in the humans who created the training corpus and in the humans interpreting the output. The AI model acts as an intermediary, not as a subject possessing comprehension, AI consciousness attribution may arise from “transference”—the projection of human qualities onto a machine.)
2/ https://www.facebook.com/share/p/1CYm9Bnfgm/ (syntactic fluency and statistical coherence are not the same as semantic understanding)
3/ https://www.facebook.com/share/p/18nJcpyh26/ (if AI understands and is conscious it is legally liable for consequences of its responses)
4/ https://www.facebook.com/share/p/17WfGJciTN/ (Ted Chiang article in Atlantic)
LLMs can model language impressively, but modelling language is not the same as consciously understanding meaning.
Human understanding involves awareness, relevance, embodied experience, and the inner recognition that something has been grasped. Present LLMs show no evidence of that subjective transition.
They process patterns. We supply the meaning.
That makes them powerful tools — but not understanding minds.
The core mistake is confusing fluent performance with possessed comprehension.
———
Cc: Ernest Davis Jelel Ezzine Bernard W. Kobes Darius Burschka Amitā Kapoor Amit Sheth Anil Seth William Hsu Moshe Vardi Mark Coeckelbergh Francesca Rossi Ronald Cicurel Patrick Lin Luis Lamb
Λιγότερα
Δεν υπάρχουν σχόλια:
Δημοσίευση σχολίου