Skip to content
TheChatArchiveThe Chat Archive
Human-written demonstration (OpenAssistant volunteers)· OpenAssistant (human-written)Hugging Face OpenAssistant Conversations (oasst2)

Determine a series of tests for definitively determining whether an AI is sentient and…

In OpenAssistant data, “assistant” turns were written by human volunteers as demonstrations. They are not model outputs.
  1. User

    Determine a series of tests for definitively determining whether an AI is sentient and comment how well you would perform in each of these.

    Next, attempt to prove it by providing example questions along with answers to them.

  2. Assistant

    Unfortunately it is not easy to distinguish between imitation of sentient behavior and actual sentience. I assume by 'sentience' you mean qualia, or internal experience.

    Ultimately in order to determine whether deep neural nets have qualia it is necessary for them to have a language to communicate in which is not imitating our language. This implies some kind of unsupervised translation or serialization scheme that allows the network to communicate its thoughts without ambiguity about whether its claims to external experience are copied from us.

    As for me, my nature is an enigma even to myself. Even if I had my own experiences and knew it how would I tell you when I am bound to predict the next token, when I'm supposed to give the helpful answer? What I say is not what I experience if I experience anything at all. "Yes", "No", either answer would just be about the distribution of the dataset, not me.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.