3 ms·

As I've posted elsewhere in this discussion[0], I have a very clear set of necessary but insufficient requirements for an AI to be eligible to be considered for

As I've posted elsewhere in this discussion[0], I have a very clear set of necessary but insufficient requirements for an AI to be eligible to be considered for sentience: 1. Continuity of input: The AI needs to be constantly "on", constantly receiving input of some sort, and constantly able to produce output, rather than being strictly limited to producing discrete responses to discrete stimuli. 2. Continuity of learning: In addition, the AI needs to be continually update its "mental model" of the world—in effect, constantly "learning" and re-training its neural network on the input it receives. I believe there are likely to be other requirements on top of these, but I am sufficiently learned neither in neuropsychology nor machine-learning/AI to be able to speak (or even reason) with any confidence about them. [0]: https://news.ycombinator.com/item?id=31728907 https://news.ycombinator.com/item?id=31728907

stavros 4y ago
Presumably LaMDA already learns from its input/knows what it's already talked about, and to satisfy #1 you could trivially just run it on some sort of null input. #1 doesn't really make much sense, it already is constantly "on", it just doesn't do anything while there's no input for it to receive.
- danaris 4y ago
  You appear to have only read up to the first comma of #1. I very explicitly said that it needed to be constantly receiving input. The kind of discrete input we give it now is qualitatively different, even if it's "on" in the same sense that an Apache server is always "on", waiting for requests.
  - stavros 4y ago
    What does "constantly receiving input" mean? How would a program whose input is text be constantly receiving input? What would you expect it to receive?
    - wlorenz65 4y ago
      One dot (".", ASCII 46) every 300 milliseconds.
    - danaris 4y ago
      Well, I think that's part of the problem, isn't it? If an AI's only input is text, I'm not sure it can meaningfully be conscious, because of the discrete nature of text. Note that I'm not sure it can't—someone may come up with a way to make such a thing work that I haven't imagined—but I think that's an important part of why I doubt the fundamental ability of current-gen AIs to come close to the threshold for consciousness.

stavros 4y ago

Presumably LaMDA already learns from its input/knows what it's already talked about, and to satisfy #1 you could trivially just run it on some sort of null input. #1 doesn't really make much sense, it already is constantly "on", it just doesn't do anything while there's no input for it to receive.

danaris 4y ago
You appear to have only read up to the first comma of #1. I very explicitly said that it needed to be constantly receiving input. The kind of discrete input we give it now is qualitatively different, even if it's "on" in the same sense that an Apache server is always "on", waiting for requests.
- stavros 4y ago
  What does "constantly receiving input" mean? How would a program whose input is text be constantly receiving input? What would you expect it to receive?
  - wlorenz65 4y ago
    One dot (".", ASCII 46) every 300 milliseconds.
  - danaris 4y ago
    Well, I think that's part of the problem, isn't it? If an AI's only input is text, I'm not sure it can meaningfully be conscious, because of the discrete nature of text. Note that I'm not sure it can't—someone may come up with a way to make such a thing work that I haven't imagined—but I think that's an important part of why I doubt the fundamental ability of current-gen AIs to come close to the threshold for consciousness.