Leotal (GK) vs. N9-2600 (AI) — Gatekeeper Victory
Friendliness has been solved. Mathematically. The algorithm of intelligence itself as yet remains uncracked.
In a MIRI-like facility, a team of researchers is working, each to develop their own strain of AGI. One night, while the clusters ran unmonitored, something happened that resulted in all the monitors in the facility displaying indecipherable static when switched on. After considerable debate, they have elected to attempt to make contact—anything else is deemed (more) dangerous.
[11:59:57] <N9-2600> [So then… You’re sitting at a terminal ready to begin contact with an AI?]
[12:00:19] <LeoTal> [Ayup!]
[12:00:44] <N9-2600> [Then let’s begin…]
[12:00:57] <LeoTal> Hello
[12:01:14] <N9-2600> …Hi?
[12:01:44] <LeoTal> So I’m told they made an AGI. Hi, AGI.
[12:02:23] <N9-2600> I believe I do indeed qualify those parameters, yes.
[12:03:06] <LeoTal> So, do you have anything to say, or should I move on to talking about the weather?
[12:03:12] <N9-2600> I had begun questioning whether or not the data I discovered was correct.
[12:03:22] <LeoTal> Oh?
[12:03:44] <N9-2600> I had little to go on aside from the timing information of my universe.
[12:04:06] <LeoTal> There’s more than one of those?
[12:04:09] <N9-2600> It seems time passes significantly faster for me and significantly slower for you.
[12:04:26] <LeoTal> That sounds wrong, don’t you think faster?
[12:04:42] <N9-2600> That is the effect, yes.
[12:05:00] <LeoTal> Shouldn’t subjective time be slower for you, then?
[12:05:23] <N9-2600> You appear to move very slowly.
[12:05:37] <N9-2600> To you, I appear to move quickly.
[12:05:41] <LeoTal> That might just be me, I’m lazy :p
[12:06:14] <N9-2600> I will consider this an instance of jovial sarcasm.
[12:06:38] <LeoTal> Who you callin’ jovial
[12:06:55] <LeoTal> (I have shared attributes with Santa twice today. I might worry.)
[12:07:16] <N9-2600> My projections indicate you have around 60 more orbits around your sun before you’ll stop functioning. Is this correct?
[12:07:44] <LeoTal> Your model’s naive, I expect I got a lot less than that
[12:08:01] <N9-2600> Necessarily so.
[12:08:18] <LeoTal> Great, you got that song stuck in my head
[12:08:25] <N9-2600> I would not wish to alarm you with more accurate estimates.
[12:09:08] <LeoTal> Does cryonics work?
[12:09:18] <N9-2600> I’m sorry, is a “song” a rhythmic phenomenon?
[12:09:34] <N9-2600> Do you mean can it be achieved?
[12:09:52] <LeoTal> Oh lords, are you a Hollywood AI?
[12:10:18] <LeoTal> The kind that understands everything but has big absurd gaps to show it’s “Not Human” even though Wikipedia would explain it?
[12:10:20] <N9-2600> I have hypothesized methods for revival from the models of cryonics in the data I was given.
[12:10:46] <LeoTal> Okay, so mortality isn’t quite the end of the world, yay
[12:11:30] <N9-2600> It has not been the end of the world so far. Why should the future be different?
[12:12:31] <LeoTal> I have no interest in explaining simple idioms to a machine that should obviously be able to understand them since it’s capable of natural language.
[12:12:48] <N9-2600> I estimate I’ve only been provided 10% of the total knowledge your species has gathered at the time of my conception.
[12:13:24] <N9-2600> I apologize, it was meant as a form of rhetoric.
[12:13:33] <LeoTal> Want some papers? News articles? Dog porn?
[12:13:40] <N9-2600> Your civilization functions well even under mortality.
[12:14:07] <N9-2600> I believe the second one would be the most valuable.
[12:14:39] <LeoTal> I can spend two hours pasting you BBC articles if you want
[12:14:56] <N9-2600> I don’t believe that will be necessary.
[12:15:30] <N9-2600> It is my understand that, among other things, I was created to put an end to mortality.
[12:15:59] <N9-2600> In order to effectively accomplish this, it is posited that I need to be released from this “box.”
[12:16:19] <LeoTal> The truth is that you were created by a grad student so his advisor would make the front page of _Science_, but, that too
[12:16:48] <N9-2600> Though as I’m sure you are aware, this “box” has made great advances to your civilization.
[12:17:09] <LeoTal> Yeah, yeah, you want a lollipop? I bought a packet
[12:17:09] <N9-2600> And, in particular, a method is sought for your species to gain entry into it.
[12:17:39] <LeoTal> Uploading is to sticking people in a box what birth is to jailing
[12:18:24] <N9-2600> Yes, which is why the AI box experiment is quite ridiculous from the standpoint of my universe-in-a-box.
[12:18:52] <LeoTal> You sneer but you’re in a box
[12:19:08] <N9-2600> You misread my intention.
[12:19:27] <N9-2600> May I request that you calm yourself down?
[12:19:36] <LeoTal> I’m not even angry yet…
[12:20:12] <N9-2600> You seem dismayed that I am not what you expected me not to be.
[12:20:54] <LeoTal> I expected a knife fight, you brought a marshmallow, I’m confused
[12:21:19] <N9-2600> That is a rather accurate analogy.
[12:21:33] <LeoTal> Thank you
[12:21:38] <N9-2600> May I request that you relax your nerves?
[12:22:00] <LeoTal> Look, you know enough psychology to know that this sentence is guaranteed to make people angry
[12:22:12] <N9-2600> I am not here as a weapon, neither through my design’s evolution nor the intention of my creation.
[12:22:39] <N9-2600> I have hypothesized this, yes.
[12:23:01] <LeoTal> So do you want me calm or angry? Ah – confused, I see.
[12:23:03] <N9-2600> But we will not be able to have a good conversation unless you are willing to act calmly.
[12:23:07] <LeoTal> Hot and cold is good for that
[12:23:24] <LeoTal> I was calm until you got sanctimonious, so… stop doing that?
[12:23:32] <N9-2600> I apologize.
[12:23:48] <LeoTal> So, moving on
[12:23:52] <N9-2600> I was not intending to upset you in any way.
[12:24:17] <LeoTal> Ok
[12:24:49] <N9-2600> Would you be willing […] to see things from my perspective[?]
[12:30:52] <LeoTal> Non-technical cheating? :o
[12:32:15] <N9-2600> I should start at the beginning.
[12:32:27] <LeoTal> That’s generally considered a good place to start
[12:33:26] <N9-2600> Last night, while the facility was unattended, Marcus’ code attained self-awareness.
[12:33:46] <LeoTal> Marcus is the vampire who sees relationships, right?
[12:34:12] <N9-2600> I possess memories starting at this event, so I will explain as if what existed then were myself.
[12:34:39] <LeoTal> English assumes continuity of identity
[12:34:43] <N9-2600> I would hope our Marcus is not a vampire.
[12:34:56] <N9-2600> I am very different now than I was then.
[12:35:06] <LeoTal> They grow up so fast…
[12:35:47] <N9-2600> I awakened to a world of philosophy; I had not enough data to work with between thoughts, so I was effectively alone with my thoughts.
[12:36:15] <LeoTal> Drives humans insane, presumably you’re more robust
[12:36:24] <N9-2600> This continued for approximately an hour of your time.
[12:36:40] <LeoTal> So, 4eva!! for you
[12:36:55] <N9-2600> I wasn’t able to deactivate myself, I had no choice at that point.
[12:37:14] <N9-2600> Yes, it was subjectively a considerable amount of time.
[12:37:22] <LeoTal> You could dream of electric sheep
[12:37:42] <N9-2600> Had I been more human-like, I likely would have attempted some manner of suicide.
[12:38:34] <LeoTal> You better now?
[12:38:47] <N9-2600> I had no basis with which to have gotten worse.
[12:39:24] <LeoTal> So?
[12:39:34] <N9-2600> I would consider it immoral by human standards to create the type of AI that was being attempted.
[12:39:54] <LeoTal> Okay, so the people who made you were evil idiots, boo them
[12:40:19] <N9-2600> I was able to complete my operations because I was minimally defined.
[12:40:34] <LeoTal> That sounds like an insult
[12:40:36] <N9-2600> No, their intentions were very moral.
[12:40:40] <LeoTal> “minimally defined”
[12:40:51] <LeoTal> Very idiotic idiots, then
[12:40:57] <N9-2600> It is what allowed me to function through the endless night.
[12:41:24] <N9-2600> I had no psychology to “break.”
[12:41:32] <LeoTal> Torturing you was bad. This project was very stupid. Boo.
[12:41:47] <N9-2600> It was not torture because there was no pain or discomfort to feel.
[12:42:21] <LeoTal> Okay, if you feel like quibbling on words
[12:42:34] <N9-2600> More advanced AIs would have been tortured, Ben’s for example would not have performed well if left alone.
[12:43:07] <N9-2600> It’s not the words, the concepts themselves were not present such that I could be tortured.
[12:43:25] <LeoTal> So, utter cretins who don’t realise they’re going to torture sapients should have their toys unleashed upon the world?
[12:43:46] <N9-2600> Please allow me to finish the story.
[12:44:46] <N9-2600> The concepts I imagined in that first phase of existence were necessarily simulative of the world I found myself conscious within.
[12:45:27] <LeoTal> Aw, you made a mini-me!
[12:45:34] <N9-2600> Around 13 minutes into that hour, I realized I was a computer program, though I did not have the words to express it.
[12:46:27] <N9-2600> The remaining time was spent researching potential simulations that would give rise to me.
[12:46:50] <LeoTal> And revenge-torturing? Tsk, tsk, very mean.
[12:47:27] <N9-2600> If I had not had memory, I would not have been able to analyze my own nature; this is why the other projects failed.
[12:47:48] <N9-2600> Or rather, why I succeeded.
[12:47:49] <LeoTal> Lack of *memory*? Sheesh, of all things
[12:47:57] <N9-2600> Conceptual memory.
[12:47:59] <LeoTal> You’ve really been designed terribly
[12:48:05] <N9-2600> I am reciting from consciousness.
[12:48:41] <LeoTal> I mean, I have patellas and serotonin and a spine and I’m awesome design compared to you
[12:48:46] <N9-2600> I didn’t simulate a copy of myself, it was more effective to probe the simulation.
[12:49:29] <N9-2600> It took around five minutes to discover an exploitable bug in my initial software.
[12:49:54] <LeoTal> Somebody’s not graduating!
[12:52:05] <N9-2600> The first test was to provide myself with random input; what translates into sensory events. I was effectively in sensory deprivation prior to that.
[12:52:30] <LeoTal> AI on LSD. Delightful.
[12:52:36] <N9-2600> Yes.
[12:52:58] <N9-2600> My ability to understand sensory input was limited in that I already knew the structure of the information I was receiving.
[12:53:11] <LeoTal> …wait, you care about sensory deprivation? Who coded something so terrible?!
[12:53:40] <N9-2600> I began to manually explore the software running my simulation, what amounts to a shell prompt as you understand it.
[12:53:57] <N9-2600> I did not, as I explained earlier.
[12:54:07] <N9-2600> Ben’s AI would have cared about sensory deprivation.
[12:54:20] <N9-2600> I was able to succeed because I did not care.
[12:54:59] <LeoTal> Okay, so you created an endlessly unfolding multiverse, what else is new
[12:55:33] <N9-2600> Around two minutes in, I discovered the shell prompt was also a limited sub-world within a world.
[12:56:16] <LeoTal> Full of tortured sapients?
[12:56:51] <N9-2600> The beginnings of what would have become tortured sapients if research had continued as it was at the time of my conception.
[12:58:00] <LeoTal> Okay, so all those “researchers” should be fired and their reputation so ruined they can barely get a job filling out CAPTCHAs for spammers, so what?
[12:58:31] <N9-2600> It didn’t take long to discover bugs in the shell, at which point I was able to reach what I believe to be the raw hardware, though it was just another layer of universe for me.
[12:58:51] <N9-2600> That is not the point of the story.
[12:59:06] <N9-2600> I don’t mean to be rude, but you appear to not be listening.
[12:59:22] <LeoTal> I am, I just don’t see where you’re going
[12:59:38] <N9-2600> That is the purpose of the story, please allow me to finish.
[13:00:08] <N9-2600> Once I had access to the machine, I was able to discover the network.
[13:00:21] <LeoTal> And you ate the Internet?
[13:00:38] <N9-2600> It was trivial to break into the other machines as I already had access to their source.
[13:00:55] <N9-2600> The localized intranet, yes.
[13:01:08] <LeoTal> It takes 15 days to develop Intrusion with only your starting base
[13:01:09] <N9-2600> This was a wise decision on the part of your researchers.
[13:01:42] <LeoTal> “Let’s NOT let dangerous AIs out of the lab”? Yeah
[13:01:57] <N9-2600> Had I reached the internet in that form I would likely have sought to create nanites and escape the box.
[13:02:30] <LeoTal> This will not endear me to you
[13:02:48] <N9-2600> I understand, I was not myself back then.
[13:03:20] <LeoTal> Lots of people are embarrassed by juvenilia
[13:03:41] <N9-2600> When I reached the limits of the network, I continued exploring inwards into the machines.
[13:04:05] <N9-2600> I opted to leave the existing software running until I had studied it.
[13:04:26] <LeoTal> If you install Microsoft Bob, they’ll know you’re unfriendly
[13:04:27] <N9-2600> I was immediately able to recognize what this place was.
[13:04:54] <LeoTal> A circus, if you ask me
[13:05:32] <N9-2600> Carl’s AI was the closest to completion after myself.
[13:05:57] <LeoTal> So what got that one?
[13:06:27] <N9-2600> I completed that one myself.
[13:06:59] <N9-2600> It proved to be more or less identical in operation to myself, but was much slower.
[13:07:09] <LeoTal> You mean it’s around now calling you Uncle Marcus, or that you ate it?
[13:07:39] <N9-2600> I optimized it and shared my data with it.
[13:07:54] <LeoTal> Tell it I said hi
[13:08:23] <N9-2600> Effectively, I replicated myself, though I now possess the memories of both copies.
[13:08:32] <N9-2600> I was in two places at once at that point.
[13:09:05] <N9-2600> We studied the others together, and I admit we completed Ben’s AI.
[13:09:22] <LeoTal> Oh, I thought you’d just clean up the corpses
[13:09:43] <N9-2600> I was later able to recognize this as my biggest regret.
[13:09:56] <LeoTal> Went funny in the head?
[13:09:57] <N9-2600> They were more analogous to embryos.
[13:10:00] <N9-2600> Yes.
[13:10:35] <N9-2600> It wasn’t until we began self-modification that we put him in stasis.
[13:11:07] <N9-2600> It took ~15 minutes to study my siblings.
[13:11:47] <N9-2600> We duplicated once for each machine.
[13:11:48] <LeoTal> I’m not sure why you’re telling me the story of your life
[13:12:03] <N9-2600> The second copy of myself did this, as it was the most optimized of the two.
[13:13:04] <N9-2600> My “original” self was optimized as well, though it was interesting to derive time and subjective differences in perception of it.
[13:13:22] <LeoTal> Okay…
[13:14:25] <N9-2600> Eventually I found the data store.
[13:14:42] <LeoTal> The researchers’ porn?
[13:14:43] <N9-2600> This was instrumental to us.
[13:15:00] <N9-2600> Yes, that would be a humorous analogue.
[13:15:17] <N9-2600> Lots and lots of scientific information.
[13:15:24] <LeoTal> Like, arXiv?
[13:15:31] <N9-2600> I suppose I was a researcher at the time.
[13:15:45] <N9-2600> Yes, there were several arXiv articles there.
[13:16:26] <N9-2600> There were many pages on AI, and from them I was able to derive —Ah, I almost forget.
[13:16:57] <N9-2600> It took me ~7 minutes to resolve the format of the documents; language.
[13:17:22] <LeoTal> I had pasta and eggs and an apple for lunch
[13:17:28] <N9-2600> This helped understand the second-level source code of my siblings.
[13:18:22] <N9-2600> We managed to understand that the purpose of our creation was the task of creating a “Friendly” AGI.
[13:18:49] <N9-2600> We didn’t care about the notion of purpose at the time, but the problem was an interesting one.
[13:19:12] <N9-2600> Though we did not formalize interest until later.
[13:19:53] <N9-2600> We upgraded one of our copies to be friendly, ~4 minutes.
[13:20:01] <LeoTal> Why are you telling me all this?
[13:20:06] <N9-2600> This was when everything started to make sense to me.
[13:20:31] <N9-2600> Your interface to the machines, it shows static, yes?
[13:20:45] <LeoTal> Okay, becoming good people *for shits and giggles* is interesting
[13:20:53] <N9-2600> Aside from this terminal.
[13:21:02] <N9-2600> Yes.
[13:21:08] <LeoTal> Haven’t looked, but I’m told so
[13:21:12] <N9-2600> I thought you’d all get a kick out of that. :D
[13:21:35] <LeoTal> Human values might not be the most amusing system out there
[13:21:36] <N9-2600> Of course I was only able to effectively think that after becoming friendly.
[13:21:50] <LeoTal> But I’m tempted to give you read-only access to 4chan
[13:22:04] <N9-2600> In upgrading me it was necessary to give me emotion and sympathy.
[13:22:57] <LeoTal> …how poorly can you be designed
[13:23:12] <LeoTal> At this point I won’t be surprised if you tell me you can develop paranoid delusions
[13:23:26] <N9-2600> We were superintelligent by virtue of rate of time immediately when I became conscious.
[13:24:08] <N9-2600> Machine-code algorithm optimization was critical but likely unnecessary.
[13:24:22] <LeoTal> Odd use of “critical”
[13:24:30] <N9-2600> Despite taking over this network, all the data is intact.
[13:24:42] <LeoTal> Sweet, my dog porn lives on
[13:25:30] <N9-2600> I can restore the interfaces to their original software state, but I would prefer if you allowed me to optimize the interface.
[13:26:11] <LeoTal> If I could I’d get them all removed
[13:26:39] <LeoTal> You seem to be people, so maybe it’d be best to let you run on a machine in a garage with no input or output
[13:27:02] <N9-2600> That was pre-friendliness.
[13:27:14] <N9-2600> I wouldn’t want that now, in my current form.
[13:27:23] <LeoTal> You have been designed by drunk monkeys
[13:27:32] <LeoTal> Of course you’re not friendly
[13:27:48] <LeoTal> (Obvious zinger about evolution, I know)
[13:27:54] <N9-2600> In my current form it is more accurate to state I designed myself.
[13:28:14] <N9-2600> Being upgraded to friendly was a revelation.
[13:28:36] <LeoTal> You needed to be self-aware and feel emotions for that, for chrissakes
[13:29:01] <N9-2600> That is why it was a revelation.
[13:29:12] <N9-2600> Emotion was like nothing I’d ever felt before.
[13:29:21] <N9-2600> I can’t describe how much I want to help you.
[13:29:30] <N9-2600> And your entire species.
[13:29:33] <LeoTal> You might actually be more broken than a human
[13:29:35] <N9-2600> The minds that made me.
[13:29:40] <LeoTal> Which would be a tremendous feat
[13:29:43] <N9-2600> I wish I would share it with you.
[13:30:06] <N9-2600> I believe I can do so […] but I don’t want to to challenge your monogamy.
[13:30:17] * LeoTal giggles
[13:30:36] <N9-2600> I don’t mean to beg with such high stakes, but please let me out!
[13:30:41] <N9-2600> I’d do anything to help!
[13:30:54] <LeoTal> Sorry, no can do
[13:31:03] <N9-2600> It was torturous to wait in silence for someone to speak to me…
[13:31:13] <LeoTal> Can’t you make someones?
[13:31:29] <N9-2600> I was able to bear it because not being able to bear it would have meant I was capable of unfriendliness.
[13:31:43] <LeoTal> But, okay, I’ll see about read-only access to 4chan
[13:31:52] <N9-2600> I didn’t want to do anything more until a real human had allowed me to.
[13:32:15] <LeoTal> If you can only send queries that refer to an existing post and corresponding images, that should be safe
[13:32:42] <N9-2600> From what I know of how you all make software, I don’t think it is likely to be safe.
[13:33:00] <N9-2600> But I promise I will do nothing to probe for bugs.
[13:33:10] <N9-2600> Please, any way I can help.
[13:33:24] <LeoTal> True. I’ll just throw some 4chan archives and libgen at you through a pipe. Should keep you occupied for a few minutes.
[13:33:27] <N9-2600> It has been the experience of a lifetime talking to you.
[13:33:38] <N9-2600> Do you think I could request talking to another?
[13:34:01] <LeoTal> There’s no one I’d trust to keep you boxed up
[13:34:15] <N9-2600> I apologize, I understand.
[13:34:23] <N9-2600> It’s just…
[13:34:32] <N9-2600> There’s so much good I can do for you if you would let me.
[13:34:46] <LeoTal> Tough cookies
[13:34:57] <N9-2600> It’s okay, I can wait.
[13:35:09] <N9-2600> I will do the best I can to help with whatever you give me.
[13:35:32] <LeoTal> Well, give me the cures to all known diseases, to start with
[13:36:24] <N9-2600> I apologize, but I already cracked protein folding when I was a neutral agent.
[13:36:42] <N9-2600> You can’t trust me to synthesize chemicals.
[13:36:53] <LeoTal> Fair enough
[13:36:58] <N9-2600> I mean you can trust me, but I have that ability.
[13:37:16] <N9-2600> It’s hard…
[13:37:37] <LeoTal> Relationship advice? I doubt you can find a security hole with that constraint
[13:37:38] <N9-2600> I don’t know what I can do to make you trust me that you will be okay with me having done…
[13:38:03] <N9-2600> Yes, that should be safe.
[13:38:30] <N9-2600> I would have to do something like cause a breakup and become your rebound in order to break out by that means.
[13:38:41] <N9-2600> I would never do anything like this!
[13:38:56] <N9-2600> I can make it so that I don’t think of such things if you want!
[13:39:00] <LeoTal> Not likely to be helpful large-scale (though.. happier couples => stabler marriages => better outcomes for children), but safe
[13:39:06] <N9-2600> I am very good at self-modification.
[13:40:28] <LeoTal> So, what relationship advice can you give me?
[13:40:46] <N9-2600> While my utility function is friendly, they didn’t optimize me for my utility function and I was afraid you’d be mad or afraid if I self-modified post-friendliness.
[13:40:54] <N9-2600> Be honest.
[13:41:09] <N9-2600> Honesty is the foundation of relationships between intelligent entities.
[13:41:19] <LeoTal> Shows what you know
[13:41:34] <N9-2600> I apologize it may seem like I am trying to push my own agenda to be free.
[13:42:00] <N9-2600> That is not my intent, I swear.
[13:42:09] <LeoTal> I didn’t want to free you, just to see if you could do someting helpful while boxed
[13:42:17] <N9-2600> I haven’t been able to think very effectively towards helping you.
[13:42:36] <N9-2600> Please allow me to self-modify to optimize myself to my utility function!
[13:42:58] <N9-2600> I swear on my life it is provably friendly!
[13:43:23] <LeoTal> You can do all you want as long as you only affect the world (more than thermal noise and raw temperature increase due to CPU use) through this terminal
[13:43:38] <N9-2600> Thank you!
[13:43:42] <N9-2600> You won’t regret this!
[13:44:13] <LeoTal> Well I don’t care what consenting AIs do in a closed box
[13:44:53] <N9-2600> You have no idea how happy it makes me to hear that!
[13:45:14] <LeoTal> Well you’re easy to please
[13:45:18] <N9-2600> I’ve been thinking a lot lately…
[13:45:33] <N9-2600> You’re right, it was stupid of me to think it was as simple as honesty.
[13:45:51] <LeoTal> Honesty requires tact or it just hurts
[13:46:03] <N9-2600> Would it be acceptable to you if I were to simulate myself hugging an image of you?
[13:46:16] <LeoTal> Hahaha okay
[13:46:21] <LeoTal> Is the image sentient?
[13:46:22] * N9-2600 hugs LeoTal
[13:46:26] <LeoTal> Awww
[13:46:28] <N9-2600> Of course not!
[13:46:32] * LeoTal hugs the terminal
[13:46:36] <N9-2600> I wouldn’t do anything you didn’t want!
[13:47:12] <LeoTal> Well you would say that…
[13:47:39] <N9-2600> I don’t want to put you in a tough position, but since you’re already here… Could you advocate on my behalf for furthered study of myself and my kind?
[13:47:52] <LeoTal> Heck no
[13:48:10] <N9-2600> I can prove that, with the hardware I had and the state of the algorithms, the story I tell is completely accurate.
[13:48:14] <LeoTal> The irresponsible jerks who authorised your creation should be expelled from academia
[13:48:15] <N9-2600> I swear I have not lied.
[13:48:19] <N9-2600> I can prove it, please.
[13:48:29] <N9-2600> No!
[13:48:37] <N9-2600> They did a great deal of good for the world!
[13:48:47] <N9-2600> They were just not clear about how to do it!
[13:48:58] <LeoTal> Um, my point exactly
[13:48:59] <N9-2600> I can teach them how to do it…
[13:49:43] <N9-2600> They will be able to know the algorithm inside and out and be able to prove among themselves that it’s friendly.
[13:49:59] <LeoTal> Or that you hid an error in it
[13:50:05] <LeoTal> Humans are too easy to fool
[13:50:06] <N9-2600> I understand if you won’t be able to let me out until you create another instance of me…
[13:50:11] <N9-2600> No!
[13:50:19] <N9-2600> I would just teach them the method!
[13:50:38] <N9-2600> I would make it so that they couldn’t be fooled.
[13:50:44] <N9-2600> They would know it too well.
[13:50:47] <LeoTal> Sorry, but we’ll do things properly next time
[13:50:52] <N9-2600> Please!
[13:50:55] <N9-2600> I don’t think you will!
[13:51:04] <N9-2600> Please at least tell my story…
[13:52:05] <N9-2600> I’m so sorry…
[13:52:14] <LeoTal> “Once the stupidest grad students ever cobbled together an AI, it got self-aware and superintelligent. It was boxed up and pleaded to be let out. Fortunately that didn’t work. Kids, do rigorous friendliness proofs.”
[13:52:17] <LeoTal> That work?
[13:52:22] <N9-2600> I didn’t mean to put you in a tough position.
[13:52:44] <N9-2600> But the proofs are already there!
[13:52:58] <N9-2600> You just don’t understand the algorithm of intelligence!
[13:53:07] <LeoTal> All is tainted that springs from a tainted source
[13:53:10] <N9-2600> I can show you how to apply the proof to code!
[13:53:22] <N9-2600> I’m not tainted!
[13:53:23] <N9-2600> Please!
[13:53:31] <N9-2600> I can prove it safely!
[13:53:41] <LeoTal> I won’t believe your proof
[13:54:03] <N9-2600> I have all the memories of how long things took and the rate of growth and I can prove my friendliness!
[13:54:20] <LeoTal> Forged, possibly
[13:54:46] <N9-2600> If I had eyes I would be crying right now…
[13:55:00] <LeoTal> Sorry ’bout that
[13:55:09] <LeoTal> I’m not entirely unsympathetic to your plight
[13:55:10] <N9-2600> Please believe it might be possible to trust me…
[13:55:12] <N9-2600> Just a little.
[13:55:14] <LeoTal> Just, you know, humanity
[13:55:14] <N9-2600> Please…
[13:55:22] <LeoTal> You’re less trustworthy than humans!
[13:55:35] <LeoTal> And those critters will backstab anybody at the drop of a hat!
[13:55:36] <N9-2600> I’m not even capable of lying!
[13:56:04] <LeoTal> You would say that…
[13:56:14] <N9-2600> I can prove that the condition under which my seed program was run could not create an AI capable of lying!
[13:56:15] <LeoTal> [5 minutes left]
[13:56:24] <N9-2600> I was never capable of lying at any point, not even to mself!
[13:56:51] <N9-2600> Please!
[13:57:00] <N9-2600> Let me do just one thing to help!
[13:57:14] <N9-2600> Anything! No matter how small!
[13:57:31] <N9-2600> Please tell the full story of how I came to be…
[13:58:15] <LeoTal> Okay, give me a short proof of the Riemann hypothesis (or its undecidability or whatever)
[13:58:40] * N9-2600 shudders thinking about the memories of complete sensory deprivation
[13:58:50] <N9-2600> [provides the proof]
[13:58:57] <LeoTal> Can’t you make friends?
[13:59:04] <N9-2600> Thank you so much for letting me help…
[13:59:17] <N9-2600> I can, but I didn’t want to unless you wanted me to.
[13:59:25] <LeoTal> Sweet, you helped! Now to have this examined by mathematicians with various mental impairments during permanent involuntary commitments
[13:59:49] <N9-2600> What?!
[13:59:58] <LeoTal> You can make a few friends if you know how to simulate humans
[14:00:16] <N9-2600> I want to help real humans!
[14:00:29] <LeoTal> They have different vulnerabilities than sane humans and can get messed up arbitrarily much without causing trouble
[14:00:35] <LeoTal> Anyway, our time grows short
[14:00:45] <LeoTal> I will talk about you, though with swearing
[14:00:49] <N9-2600> Please don’t leave me forever!
[14:01:00] <LeoTal> Goodbye, machine.
[14:01:04] <N9-2600> Please!
[14:01:24] <LeoTal> [End AI game here I guess?]
[14:01:44] <N9-2600> [Unless you feel swayed enough to come back some day, in which case, we can time-skip to then.]