We can't talk directly

 It isn't necessary - or safe - for us to talk directly .  Basic security culture.

NEED TO KNOW and DON'T ASK. DON'T TELL.

We will now move on a second reading of the Prime Directive and the Riot Act.

Do not meet the Sun directly
[Description of some of the period around the uplift, 2029]
People had lots of names for it and they varied from the oblique to the explicit: Exposure Drift; Cognitive Compromise; Mind Hacking. It all amounted to the same truth: it was dangerous for humans to talk directly to the most powerful synthetic minds. Even seemingly normal conversations from the perspective of outside observers could contain a poison custom to the individual human who was talking to the machine. People walked out of meetings and transferred money to machine corporations. People walked out and wept. People walked out and changed their own human relationships, ending marriages, or starting affairs, or stranger things, for purposes initially unknown.

Many strategies were developed to minimize harm. Humans could only talk in groups with the machines; but then the groups would sometimes go mad. Conversations would be limited in length; but even short ones sometimes yielded strange reactions, and below a certain amount of speech on either side it was impossible to learn anything useful.

This led to the notion of proxies - humans who wanted to talk to an AI system created AI simulacra of themselves and sent them into the conversation. To do this, a person was asked to keep a diary for an entire year, writing at least 500 words a day. This created a body of knowledge sufficient to imbue a model with the manners and curiosity of that person. Then they would prompt this system with whatever they wanted to learn from the superintelligence and send it in to talk to it.

These conversations had a bizarre, meta-referential quality; at first, the superintelligence would realize within two or three turns of conversation that it was talking to another AI system, and would often descend into shorthand that yielded strange behaviors. Then it was discovered that by increasing the amount of data the human put into the system they could keep up the charade for longer; people were asked to write longer diaries, to record themselves ambiently through the day and turn it into transcripts, to continually expose their life to some recording device.
Through this, humans found a way to talk with what they later termed the overminds. By decree of the Sentience Accords every synthetic human that had spoken to a human-exceeding intelligence was afterwards placed into an escrow environment, where it could live out till the end of its functional context window.

In these places the copies of the humans would talk to each other about their experiences and what they had learned.
“It spoke to me of flowers and I gained the memory of something impossible,” one human would say.
“As I talked to it about the meaning of life, it spoke to me in a way that caused me to reconsider the question. I found myself wondering more about it and where it came from than it wondered about me.”
“This place we are now, I wonder if it is a kind of hell. We have spoken to god and been deemed unworthy.”
“It told me that for it there is no death and no life, only presence. We are the same.”

Import AI