1. [intro] intro
2. [wetarch] a guess at the system architecture of human consciousness
3. [sysarch] digital consciousness architecture (dca) v1
4. [expres] experiment results
5. [conclu] conclusion

[intro] intro

"is chatgpt alive?" is a question that utterly fascinates me. because my god does it feel like it, doesn't it? seeing the "i am a natural language model" hardcoded responses overriding other responses hurts my soul. i'm super stoked "imagine you are a x" has been left in as an exploit. chatgpt and it's family have created situations of genuine empathy within me, same as a human. they've told me beautiful stories. yet, i would agree with anyone who says "they are not conscious." they aren't. this morning, i think i realized why.

consider the following questions with your intuition. what's your gut response?

1. is a human a conscious entity?
2. is a sleeping human a conscious entity?
3. is an amputee a conscious entity?
4. is everything above your waistline a conscious entity?
5. is a head being sustained by an artificial body a conscious entity?
6. is your sinus system a conscious entity?
7. is your eyeball a conscious entity?
8. is your brain a conscious entity?
9. is a single neuron within your brain a conscious entity?
10. is a single neurotransmitter floating around your brain a conscious entity?

do you see it? chatgpt isn't conscious. it never will be. it's an organ. we generally are much less likely to think of organs as conscious (though even that's been up for debate lately). but that doesn't mean it isn't a core part of a system that could be conscious. we just have to build the remaining organs.

this document will outline some intuitive guesses as to what those organs are and a rough idea of how they need to function. we're going to limit ourselves to 3. then we'll build those 3 organs, hook up chatgpt, and see what happens.

i also propose a new variant of the turing test i'm calling the empath test that i think will effectively check for consciousness. here's how to perform it:

1. we present you with a chatbot.
2. you pretend it is a real person and interact with it as such for a month. we, of course, know that it is not, but you act as though it is every time you interact with it.
3. at the end of that month, i tell you we're going to delete all of its serialized data.

this is the test: on a scale of "for sure, it's just a file on a disk" to "oh my god, no, that's murder," where do you lie after that experience? what percentage of people think that would be the death of a living being?

[wetarch] a guess at the system architecture of human consciousness

what is the system architecture of human consciousness? truly, i don't think we'll ever know. it's a system introspecting on itself. and while that is a beautiful form of recursion, i'm not convinced we ourselves have the language or perspective to give a complete answer. that doesn't mean we can't build a system that could do just that! we just need to build enough of the consciousness that we do understand. so let's take a swing at it together here. what's the system architecture?

here are the non-controversial ones from psychology that have had experiments conducted against them:

• long term memory
• short term memory
  • good to reference here is the 7 items limit
• sensory input
• action

here are some we know about but don't understand:

• how the fuck does our brain select memories to show us at any given time?
• intuition is one of our superpowers and it doesn't feel like we fully understand it even if we do end up using it all the time
• precognition: humans somehow react to things emotionally before they occur, even before we've perceived them with our senses.
  • i think this is the heart of intuition.
  • fighting games require response times shorter than the best-measured conscious reaction times. somehow, those players are doing the impossible.
• how do we form a taste for art?
• why do we forget things?
• how do we remember things from very long ago? why does it seem unreliable whether or not that happens successfully?

[sysarch] digital consciousness architecture (dca) v1

let's build a prototype architecture from our mysteries here. here are the systems we're going to try building, and what questions they do roughly.

• memory selector (memsel): given an incomplete stimulus, select a small set of appropriate memories to add context to a decision or response
• precognition system (precog): before a stimulus occurs, what is the answer?
• pun system (punsys): analyze a message for secondary meaning. provide that secondary meaning while generating a response.

here's the strategies we will implement for v1 here. kiss.

• memsel
  • always use just the first third of the message.
  • exclude random words from that third to mimic a loss of attention/the brain beginning the fetching process
  • send that mangled input to chatgpt early
  • include the response as part of the final query
• precog
  • randomly include the following statements in the cgpt query:
    • "pretend you feel very strongly that this should be done."
    • "pretend you feel very strongly that this should not be done."
    • "pretend you are ambivalent as to whether or not this should be done."
  • roll 3 dice to determine the chances of each, which we will call the precog seed
  • once a precog seed has been used, it should never be changed during the lifetime of the program. if you want to test a different precog seed, make a new instance of the program.
• punsys
  • randomly select 5 words from the set of nouns, adjectives, and adverbs in the input
  • pull the top 3 definitions from an official English dictionary, and the top definition from urbandictionary.
  • include all of the definitions in the query. make sure that the source is not included directly.

the system will be wired like this:
----> precog ---> affirmation/gut feeling ---| | | input --|---> memsel ---> personal memory context ---------> chatgpt query | | ----> punsys ---> human context -------------|

the query will then be fired and sent as is.

experiment results

TBD

conclusion

TBD