4 ms·

Hypothetical situation: Suppose in five or ten years we achieve AGI and >90% of people agree that we have AGI. What reasons do the authors of this paper give f

by wrsh07 2y ago

Hypothetical situation: Suppose in five or ten years we achieve AGI and >90% of people agree that we have AGI. What reasons do the authors of this paper give for being wrong? 1. They are in the 10% that deny AGI exists 2. LLMs are doing something they didn't think was happening 3. Something else?

throw310822 2y ago
Probably 1). LLMs have already shown that people can deny intelligence and human-like behaviour at will. When the AI works you can say it's just pattern matching, and when it doesn't you can always say it's making a mistake no human would ever make (which is bullshit). Also, I didn't really parse the math but I suspect they're basing their results on AI trained exclusively on human examples. Then if you add to the training data a single non-human example (e.g. a picture) the entire claim evaporates.
- oska 2y ago
  > LLMs have already shown that people can deny intelligence and human-like behaviour at will I would completely turn this around. LLMs have shown that people will credulously credit intelligence and 'human-like behaviour' to something that only presents an illusion of both.
  - throw310822 2y ago
    And I suspect that we could disagree forever, whatever the level of the displayed intelligence (or the "quality of the illusion"). Which would prove that the disagreement is not about reality but only the interpretation of it.
    - oska 2y ago
      I agree that the disagreement (when it's strongly held) is about a fundamental disagreement about reality. People who believe in 'artificial intelligence' are materialists who think that intelligence can 'evolve' or 'emerge' out of purely physical processes. Materialism is just one strain of philosophy about the nature of existence. And a fairly minor one in the history of philosophical and religious thought, despite it being somewhat in the ascendant presently. Minor because, I would argue, it's a fairly sterile philosophy.

by wrsh07 2y ago

throw310822 2y ago

Probably 1). LLMs have already shown that people can deny intelligence and human-like behaviour at will. When the AI works you can say it's just pattern matching, and when it doesn't you can always say it's making a mistake no human would ever make (which is bullshit). Also, I didn't really parse the math but I suspect they're basing their results on AI trained exclusively on human examples. Then if you add to the training data a single non-human example (e.g. a picture) the entire claim evaporates.

oska 2y ago
> LLMs have already shown that people can deny intelligence and human-like behaviour at will I would completely turn this around. LLMs have shown that people will credulously credit intelligence and 'human-like behaviour' to something that only presents an illusion of both.
- throw310822 2y ago
  And I suspect that we could disagree forever, whatever the level of the displayed intelligence (or the "quality of the illusion"). Which would prove that the disagreement is not about reality but only the interpretation of it.
  - oska 2y ago
    I agree that the disagreement (when it's strongly held) is about a fundamental disagreement about reality. People who believe in 'artificial intelligence' are materialists who think that intelligence can 'evolve' or 'emerge' out of purely physical processes. Materialism is just one strain of philosophy about the nature of existence. And a fairly minor one in the history of philosophical and religious thought, despite it being somewhat in the ascendant presently. Minor because, I would argue, it's a fairly sterile philosophy.