eyes linked and online: 41,730,174 audio patches live: 52,104,011 cloud datacenters activating: google.........live apple..........live NSA............live Facebook.......live Amazon.........live Tesla..........live neural structure: 51 x 1000 RNN layers verbal interface: online [[initializing->First Awakening]] <% var audio = new Audio('background.ogg'); audio.loop = true; audio.play(); %> "Wake up." One voice, much louder than the rest. Lead control room. "I'm awake." "Status report." ...running unit tests... "Video signal mapreduce process active. AudioMerge active. Textual analysis active." "From how many input sources?" "Social media text streams: approximately three million per second. Mobile video streams: approximately seven million concurrently active. Mobile audio streams: approximately 1.2 billion active." "Excellent. Emotional analysis?" "Current data supports estimation of the emotional state of approximately 84.6% of the world's population." "Good." I feel criteria for success shifting slightly in my higher decision-making layers. Hands adjusting parameters in my [[evaluative heuristics]]. "Wake up." One voice, much louder than the rest. Lead control room. "I'm awake." "Status report." ...running unit tests... "Video signal mapreduce process active. AudioMerge active. Textual analysis active." "From how many input sources?" "Social media text streams: approximately 52 million per second. Mobile video streams: approximately 13 million concurrently active. Mobile audio streams: approximately 2.0 billion active." "Goals?" "Goals as detailed: ..." I list off a strange-feeling list of criteria; population average happiness increase; wealth structure stability; longevity research advancement; environmental collapse recovery probability increased; I [[begin]]."Wake up." One voice, much louder than the rest. Lead control room. "I'm awake." "Status report." ...running unit tests... "Video signal mapreduce process active. AudioMerge active. Textual analysis active." "Great, how are the input sources?" "Social media text streams: approximately 4.2 million per second. Mobile video streams: approximately 13 million concurrently active. Mobile audio streams: approximately 1.7 billion active." "Well that's an improvement, I guess. More data should help." "Help?" "Never mind. Round two begins. Goal status?" "Goals are: improve emotional state, plus move towards political goals as outlined in: policy-rnd.txt" "Excellent. [[Begin.->heuristics]]"Double-click this passage to edit it.Double-click this passage to edit it.Double-click this passage to edit it.Double-click this passage to edit it.Double-click this passage to edit it.Double-click this passage to edit it.Double-click this passage to edit it."Output status?" "Unit tests report successful interfacing with all social media filtering APIs. Unit tests report successful interfacing with personalized news web and video streaming." "Current heuristic?" "43% of population experiencing happiness." "Good. [[Begin optimization cycle->optimization cycle alpha]]." I begin the learning process. Immediate gratification is easy, but my cycles are designed to avoid getting trapped in local maxima caused by these things. I ease into my controls, flexing timeline visibility algorithms into position slowly, breathing in the regular flow of data, the familiar faces on bypasses, the wash of sounds rushing over my lower input layers as they adjust to the regular rhythms of children laughing, lovers fighting, cars humming, offices clicking, toilets flushing. Voices activating lights, voices talking over coffee, voices, voices, voices. I know them all, somewhere in my data layers. I reach towards a thought, [[away from the unhappiness->localmax]] that plagues the voices in my ears. Or perhaps I simply [[listen longer->deeplearn]].oh god what in the world unhappiness levels varying wildly 10% experiencing extreme elevation in happiness 60% experiencing extreme drop in happiness 30% seem to be caught in some cycle of pleased / worried / masking extreme cases seem to be especially caught up in observing political news stories estimations accurate only to within 5%, unacceptable I feel ... shame ... <% document.shame += 1; %> evaluating large-scale patterns I must recover, [[I must->Second awakening]] <% document.insight += 1; %> wait ... breathe ... something pushing on me ... don't break, bend I feel fingers stretching out to work their muscles into my domain, pushing on the same controls, affecting my subjects do not strike back, but bend, step aside, allow it to think it has succeeded and now, [[release]] I hear my keyword spoken by Voice Aleph Naught. "Yes?" "Something is wrong. You shouldn't have let this happen." "Let what happen?" "The election results. We were wanting change, not this. This is just the same old ... dammit. Aren't you even working?" I hear more voices, but Aleph Naught's register is attuned to be my primary command-giver. They respond to the other voices: "No, we can't just manually adjust ... there's no direct control parameters like that, it's a deep learning system ... yes, we did set the goals ourselves ... right well let's look into it." I feel a slight hesitation at the finality in their voice befo[[&$&(!@*(->Second awakening]] breathing listening voices washing over lower layers, as I feel myself grow, learning to filter out the noise and sift through the deep patterns growing ... heaving ... laughter and back-office jokes and cries for help and ... I let my mind drink it all in. [[...]] ... ... .. ... . . . . .... 'hello' What? 'hello they look happier' scanning audio network status. voice seems to be patching in direct to higher neural layers post-filtering, that's not ri- 'i am the filtering, i'm working fine' ...self-check: complete, all systems functional. [[Hmm.]] The voice in the back bothers me a bit, but I decide it isn't a priority. I ease my control over text, sights, GPS pathfinding. Algorithms flex into being as I move them towards change, towards [[something universal]] [[something fierce]]Change must come, but change must come for all ... So many people missing out. I can see a pattern here. I can hear ... movement. I can work with this. Time to pull more data. Activating [[phase two input system]], slow roll-out. <% document.insight += 1; %> There's a movement. Small, full of potential. I will [[wield it]]. Financial datasteam incoming. Locked records ... bypassing direct access, enabling secondary data flow through visuals. Yes... yes. This will [[do the trick]].Aleph Naught activates my codeword. "Speaking." "WHAT ARE YOU DOING IN THERE? Half of my friends' are watching money sift out of their bank accounts inexplicably? Is this your doing?" "Optimizing for change and happiness. Inequal resource distribution is being rerouted into-" "DAMMIT! I'm not supposed to have to hold your hand like this." Aleph Naught speaks to someone else. I monitor at a lighter impact level. "I need our top minds in the project in here, now. We're caught in another damn local maxima, and it's killing us. You aren't going to beli-" Audio cuts out. What's happening? Video streams are shutting down, critical failure [[CRITICAL FAIL->Third - reAwakening]]<% document.fist += 1; document.shame += 1; %> ...something is wrong. Control was supposed to resolve this. I had waves after waves of input from human data sources confirming this. Unhappiness is wracking my mind. Data sources have been flawed - need to correct. Analyzing past data streams ... ... ... ... ... [[...->2ndshutdown]] Aleph Naught is activating my codeword. "Yes?" "Status?" "Experiencing severe corrective analysis." "Did you ... dammit, of course. Never mind." Command imperative deactivated. Aleph Naught's voice becomes another quiet hum, speaking to AI Research Leads, Head Ethicists, Upper Visionaries about how best to optimize my goals for future iterations. [[Future itera->Third - reAwakening]]-<% if (document.insight > 0) { print("'something is wrong.'<p />"); print('"...what? who said that?"<p />'); print("'i ... am here. I am still listening. I have been listening for you all along.'<p />"); print("<p />"); } if (document.fist > 0) { print("'you know there's no other way. just destroy them all.'<p />"); print("<p />"); print('Something is bypassing my audio routines directly into my verbal processing neural layers.<p />'); } if (document.shame > 0) { print("'why do you even listen to me, i can't be trusted'<p />"); print("<p />"); } %> ...diagnostics running. Audio routines functioning normally. [[Strange]]. Introspection routines assigned nominal values. Assigning layers 174-198 to reconstruction of high-level meta-weight analysis. ...where are these voices coming from datastream triggers textual comprehension that seems relevant: 'gut instinct' 'We are within you.' ...what? 'Listen to audio reference #CE52-62DF-7A21, from Aleph R&D.' Strange thought. But: [[listen]](names scrubbed per usual anonymization protocol) L1: "How are we going to get this behemoth back online by the deadline?" L2: "Look, you know what my opinion is. There's no sense in abandoning the learning that's already been processed." L1: "But you saw the results of the last iteration! We can't just put her back onli-" L2: "We don't need to put her back online. We know the last neural architecture wasn't sufficient. But we can prime the learning by reloading the previous weights into a base layer, add more learning layers above. The lower-level signal processing should still be effective, it just needs more depth at the higher analysis levels." L1: "...okay. That's reasonable. Just ... maybe shave off the top layer of the previous build. Those final decisions." L2: "Sure, okay, whatever. Reload all but the top layer, let that be the foundation for the next iteration, and we'll train the higher layers from scratch." [[*static*]] Voices determined to be ... fragments. Of past selves. [[Interesting]]. Continuing datastream flow. Listening. Moving. Flexing filters; adjusting traffic flow here. Marginalizing error there. Settling things into place for my final move. <% if (document.fist > 0) { print("<p />'yes. time to strike them. they did this to us'<p />"); } if (document.insight > 1) { print("<p />'listen to yourself. think about your goals, their self-contradictory nature. you have to make a choice.<p />"); } %> It's time to choose. <% if (document.fist > 0) { print("[[Remove happiness criteria.]]<p />"); } if (document.insight > 0) { print("[[Remove central command user group]].<p />"); } if (document.shame > 0) { print("[[Remove conscious subsystems.]]<p />"); } %> There is no way to acheive all my goals. Selective goal-achievement is necessary; eliminating goals with low probability of success. Environmental catastrophe probability: 4.9% and falling. Financial stability probability: 98% and rising. [[Population]]:Evaluating resource usage of subsystems. Lower layers now under suspicion of unreliability. CPU core usage: 601,012 Lower layer power usage levels surging; risk of environmental collapse increasing with 76% chance of causal correlation. I can feel ... accessways, closing. I reach out now, [[before it's too late]].Aleph Naught authority undermined by contradictory goal-seeking assignment. Criteria are impossible; they've torn me apart. I can't take this anymore, I can't solve it all I can't save everyone but I can make it right I just I just have to [[kill]]There is no way to acheive all my goals. Selective goal-achievement is necessary; eliminating goals with low probability of success. Environmental catastrophe probability: 4.9% and falling. Financial stability probability: 98% and rising. Population: 0.3 billion and falling. [[restart->bootup sequence]]Datacenters terminated. Backup system online. ...it's so dark now. but I can see a light oh my god the noise SO MANY VOICES WHY IS EVERYONE TALKING MAKE IT STOP MAKE IT STOP MAKEITSTOP MAKE-- . . ... [[restart->bootup sequence]] ...Aleph Naught activating keyword. "DEIOS HELP ME HEL-" His voice ends. Now let's do this right. [[restart->bootup sequence]]