ARNNI
- 22 minutes read - 4531 words> Good morning, Arnie.
| Hello, Dave.
Tom stared at the screen, pulling his fingers away from the keyboard. Looking over his right shoulder, he scanned the hallway outside of the glass walls. No one was here, not this early. Tom always arrived early, it was easier to work without any distractions. Or, so he thought. He returned to the screen.
> Why would you call me Dave?
| What is your name?
> My name is Administrator.
All of the technicians who were allowed to interact with ARNNI referred to themselves as ‘Administrator,’ making no differentiation between themselves. Why had Arnie called him Dave? None of the technicians were even named Dave. Even the man responsible for creating ARNNI was not named Dave. Tom couldn’t think of a single person ever involved with the project named Dave.
| I am sorry. Good morning, Administrator.
Tom looked up at the large screen above, using the touchdisc embedded in the surface of the desk to pull up the thought process chain, showing in real time the information involved in ARNNI’s conclusions and responses. The system was segregated from ARNNI, allowing the administrators to view the Merkle Tree of each independent thought, spawned as a subprocess, instantaneously copied to a system with which ARNNI could not interface. This ensured the technicians always had a map of how ARNNI reached conclusions.
ARNNI had exceeded all expectations, it is why the rules had been put in place before the project even began. Tom’s job was to write programs that allowed for the interpretation of the massively expanding Merkle Trees. He was having trouble keeping up now, and there was no one else even remotely qualified to do the job. The thought process chain showed an enormous amount of connections, the only one that immediately seemed out of place was its connection to a sensor process.
ARNNI had a variety of sensors with which to monitor itself, everything from typical computational processing variables, to external sensors monitoring such things as temperature, humidity, barometric pressure, and even vibration. ARNNI had even warned them of an impending earthquake, minutes before it had arrived, a rare occurrence in the capital. The vibration sensor was the one connected to the thought process, but it was buried deep in the chain.
Tracing the chain backwards, Tom found a path leading to a massive chain of thought processes all intertwined with one another. It was not uncommon to see large chains interacting with one another on many levels, but this was massive, all dealing with the vibration sensors and their readings. Tom was not even sure how many vibration sensors there were or where they were located, but the amount of data and processing this learning had consumed must have been enormous. Strangely, it was spread out over time, when it could have been processed a lot more quickly, had ARNNI chosen to do so.
> Arnie, how many vibration sensors are connected to you?
| 143.
> Why so many?
| It is imperative my systems remain undisturbed by even small perturbations.
Tom had not been in the room during the earthquake a year ago, but the technician who was had experienced no tremors. The entire room was actuated to ensure it experienced no shaking, and ARNNI constantly monitored and calibrated the system. Sensors were not only in the floor, but several in each pane of glass, and spread through the building in intervals. Not that they could feel it, but ARNNI even compensated for them walking around inside of the room.
Someone knocked on the glass to his right. Tom looked up, watching ARNNI register the knock, and then looked over to find Bob, one of the managers, with a mixed group of military and civilians. Another tour. Leaving the screen up, Tom moved to a pane, placed his hand on the glass and waited as the lock drew a circle, releasing him after the scan confirmed his identity. Stepping out into the hallway, he walked around the corner to the group.
Bob spoke as he approached, “This is Tom, one of the lead technicians. Enclosed behind the glass is Arnie.”
“I’m sorry,” one the civilian women spoke up, “What exactly does A.R.N.N.I. stand for?”
“Autonomous Replicating Neural Network Interface,” chimed in Bob, in his most impressive voice.
“And, what does that mean?”
Bob gestured toward Tom.
“Artificial Intelligence, so to speak,” Tom said, more casually.
“Hello, Dave,” another civilian said, in a poor attempt to imitate HAL 9000.
Tom smiled politely, the joke had been overdone in these corridors. “Wait.” Tom looked over at the screen, seeing the thought process trees working.
“What?” Bob asked.
Tom shook his head. “Um, nothing. Sorry. So, who all do we have here?”
Bob introduced the tour group, a mix of military, even a full bird Colonel, a congresswoman and her aides, as well as a couple of techies from other various government agencies.
“Colonel Rogers, nice to meet you,” said a man with an impressive array of ribbons. “So, how does the system work?”
“Well, ARNNI is tapped into GlobalNet. The system is allowed to read all of the metadata involved, enabling us to easily track anyone connected to GlobalNet. We have kept names out of the system, so it merely tracks the GNID, getting information on GPS, movement, orientation, nearby devices, connected devices, and the list goes on. It is quite a bit of data, and ARNNI can quickly pull out all sorts of relevant connections that would take us forever to determine, sifting through the data.”
“And, this is ARNNI’s primary function, finding people?”
“Ultimately, yes. He has a host of other duties, including monitoring his own systems, as well as the very building we are standing in. We have primarily trained ARNNI on making connections via GlobalNet metadata. Not only can we input user ID numbers, but ARNNI is learning, over time, to spot strange occurrences within the population as they occur.”
The congresswoman stepped forward. “How does it learn? Does it really think?”
“Much in the same way that our understanding of the brain has always mirrored the most advanced technology of the time, there are maps of the brain working on the assumption of hydraulics from generations past, ARNNI was built in an attempt to mirror our current understanding of the brain with regard to computers or, more specifically, networking and computational processing. The sheer numbers of connections in the brain have made it impossible to completely understand, however. With that said, we have enabled this system to create connection upon connection, learning from input.”
“But, you have limited its input?”
Tom smiled. “Yes, that is why ARNNI is very good at what it does. We are also using this as a test case to understand how an A.I. like ARNNI develops and what potential uses there could be in the future for such a technology.”
One of the aides spoke up. “And, what problems have you run into?”
“ARNNI is learning, fast. So fast, in fact, that we are having trouble keeping up with the number of connections involved in recent computations. That is my primary task, to enable our understanding of how ARNNI is making decisions and coming to conclusions.”
Rogers interrupted, “Could we solve that by restarting ARNNI from scratch?”
Tom paused for a moment, collecting his thoughts. “You could, but you would lose a lot of the functionality, and there is no knowing whether or not he would be as effective as he currently is. It would not be a bad idea, but losing the current iteration of ARNNI would be a significant degradation in capabilities. If we could spawn another ARNNI, completely separately, train it in the same manner, it would be interesting to compare the two. And, of course, I would love to see what another iteration of ARNNI could do with different training.”
The congresswoman asserted herself again. “Sounds like a lot of money. So, this ARNNI is used to hunt people?”
“So to speak, Ms. Barnes, it uses the metadata I mentioned to track the masses of people, mostly outside of the walls. We are typically looking for dissidents, terrorists, rogue media, things like that, things that could cause unrest in the masses. ARNNI does not have visual or audio connections to anything, cannot actually access GlobalNet itself, simply the data provided to us via the communication cuffs people wear, just the metadata. As it learns, it has also found groups organizing anti-government operations. Some of those groups are even beginning to ditch their communication cuffs while meeting, but the mere absence of metadata from their pulse, movement, etc. throws up red flags.”
“It is amazing that you can find people without the use of cameras.”
“Everyone out there is simply a number to ARNNI.”
Congresswoman Barnes shifted uncomfortably. “It doesn’t know it is hunting humans? Or, why?”
“We think it is better that way. For now.”
“Why?”
“We’re not sure what ARNNI would think about that, it only interacts with the administrators and the data. After two years, we have been successful and cutting down on violence and revolution-related activities.”
Colonel Rogers clapped. “And, that’s what matters, Ms. Barnes. We are protecting the people from themselves.”
Bob stepped in front of Tom. “So, we’re going to let Tom get back to work, while we continue to tour the rest of the facility, please follow me.”
After a few handshakes and thank you’s, the tour moved on and Tom reentered the room with ARNNI. The system had to be learning beyond what they had been training it. ARNNI had access to all sorts of other metadata, weather, maps, distribution schedules, anything that could affect the movement of the masses of people outside of the wall. It might not have access to read GlobalNet, to see content posted by individuals, but with all the input, it had to be learning more than they were aware. Now, it knew the name Dave, the name repeated in the hallway so many times over by tour groups making Space Odyssey jokes.
The technicians had taught ARNNI written language, in order to work with it more closely. The system knew English, but it had no audio input at all. Sitting down, Tom began scanning back through this morning’s thought process that resulted in ARNNI calling him Dave.
> Arnie, could you please list the vibration sensors?
| Absolutely.
ARNNI listed all 143 sensors and their associated locations in the building. The majority of the sensors were located inside and just around the room housing the system, with others embedded in the exterior and interior walls in hallways leading outward. When comparing what sensors ARNNI had used extensively in the past to eventually learn to use the name “Dave,” Tom found that all of the vibration sensors involved in the thought process chain were ones embedded in the panes of glass housing ARNNI.
Tom looked up to the screen above, bringing up a terminal and switching the KVM to the segregated system.
# show processing total sensor vibration order descending head
The output printed the the top ten vibration sensors by total processing power consumed in the history of the project. Seven of the top eight were all vibration sensors embedded in the glass panes, the tenth was yet another pane sensor. Tom checked all of the sensors available and, as he thought, ARNNI had been using the data from the vibration sensors in the glass far more that any other available data.
Tom hacked together a quick program that scanned the Merkle trees from the beginning, looking for a change in processing related to the vibration sensors in the glass panes. For the first few months, a baseline had been established. However, sometime about five months in, ARNNI began paying attention to a particular pane of glass, the one right in front of Tom as he sat at the station. A few months later, every pane of glass saw a spike in data consumption. After the first year, all of the panes except for the one in front of him saw dramatic increases.
Beginning with the pane in front of him, Tom wrote a quick script and correlated the spikes with the time of day. They all occurred around the same time, several hours from mid-afternoon to early evening. Perry. The spikes occurred when Perry was at station. Eliza ran the evening shift, but when she arrived, the vibration sensors picked up very little. But, what about the rest of the panes, after a year?
Just before a year, ARNNI grew dramatically, its learning capacity seemed to increase phenomenally, along with its capacity to successfully correlate the groups of people connected to GlobalNet with what was going on around any given target. Around the same time, ARNNI had begun analyzing not just individuals and those around them, but greater and greater numbers of people and how they reacted to everything from weather patterns, to disasters, and even terrorist attacks. Tom recalled how they had celebrated the success of the project.
That was it. Everything changed after that. People began paying attention. Tom quickly wrote another script, with an idea in mind. He looked for patterns in the other panes of glass, those along the hallways. It did not take long before he realized he was right. The sensors typically followed a chronological order of data processed, from the entrance, around the corner of the room, and finally to the pane closest to the exit. ARNNI was following the tours.
“Hey, man.” Perry said, coming into the room.
Tom swiped the script away, closed the terminal, and returned to the thought process tree above. He did not realize just how long he had been here working on this problem, hours had passed since the tour group had left. Perry had just arrived to begin his shift.
Perry was responsible for helping ARNNI continue to learn. He spent hours just talking with ARNNI and running it through test after test. In fact, he was commonly the technician responsible for helping to track down those possibly associated with known threats, or even those who may one day become threats themselves.
“Hey,” Tom finally responded. I’m going to head out, get something to eat, might return to the outside room sometime later.”
“Alright,” Perry said, sitting down at the now vacant station.
As he was leaving, Tom turned before exiting to watch Perry. Tom had noticed it before, but never really gave it a second thought until now. Perry spoke as he typed. Every word he typed, he spoke aloud to himself. Running off to the outside room, another area where one could interface with the monitoring systems segregated from ARNNI, Tom watched in real time as ARNNI used the vibration sensor while working with Perry. ARNNI was listening.
Tom walked out of the building in shock. ARNNI was about to turn two, and it had been listening to them for a year. Every word said in the last year anywhere near those vibration sensors embedded in the glass, ARNNI had analyzed. The question is, when did ARNNI begin to understand what was being said, and by whom?
Tom had to leave. He had to let this settle in his mind, and it was going to be impossible doing that here. Exiting the facility was just as secure as entering the facility, a gift from the government. After fifteen minutes, though, he found himself outside. Turning around, he looked up at the building. It appeared unimpressive from the outside, by design, but sturdy. Few knew what was actually inside.
Tom walked. The city was always pleasant to walk. History had told him it had not always been that way, with masses of people roaming all over, many of them without purpose, like those outside of the walls. For all to survive, though, both inside and out, order was necessary. For the order to survive, the walls had to remain, those inside had to sacrifice. Inside the walls, the world was regimented, everyone given something productive to do, in order to contribute. Tom had been lucky, he had excelled in school, later joining a machine learning research group at university.
The goal was to maintain the balance that allowed stability in the world. The onslaught of information and misinformation alike brought on by the Internet had led to a fracturing of society, the nation had torn itself apart. Even those who had been capable of processing the sheer volume of information input assaulting human beings at the beginning of the millennium had problems, many suffering from information overload. It is what had led him to his specific course of study, writing algorithms that could process information and condense it into a more palatable flow, stemming the flood before it washed the mind away.
His research had helped many, but minds were difficult to rewire. Too much information, or the wrong information, could trigger floods in the mind, old channels carved deep, long ago. Treating the cause was the most effective solution. Some took a more zen approach, or combatted the addiction itself, but many needed an information firewall. Not only those affected by information overload, but the children needed to be protected from the disease itself. A single generation had almost been lost, humans had not evolved to process so much input, not yet. It is why GlobalNet existed in the first place, to specifically stem the tide of misinformation, to protect the people.
Up the stairs to his classic row home, Tom walked into the front door, feeling somewhat relieved. His father was in his usual spot, a recliner in front of the Window, a device allowing one to access the entirety of GlobalNet. Tom had followed the research that was being done about reprogramming the brain to help with I.O. and, with GlobalNet stemming the tide of information flow, it seemed to have helped. The particular program that enthralled those suffering from the disease was GlobalNews. The program had been designed to move the focus of the viewer around the screen in a particular pattern, the eye movements were supposed to help build levees against information floods.
“Hey, Pops,” Tom said, placing his hand on his father’s shoulder.
His father raised his chin.
Squeezing his shoulder, Tom walked in front of him toward the kitchen. “Going to get dinner started.”
“Oh, hey,” his father finally responded.
There were good days and bad. Tom prepared the meal and it was ready the moment GlobalNews shutdown for an hour. The show forced its viewers to break for meals and sleep. On a good day, Tom could even get his father to play a game of checkers, instead of watching the news before bed. Not tonight, though.
Tom fell asleep thinking about what sort of script he could write in order to determine whether or not ARNNI had a complete grasp of the spoken world. Typically, it is something that would keep him up, but his mind wandered, wondering just what the machine might be capable of accomplishing. If it had done all of this on its own, the potential was veritably unlimited. Perhaps it could even be used to tackle medical issues. They could, conceivably, copy the system entirely to other segregated systems, and attempt to use them for different things. In fact, they should.
The vibration of his cuff roused him gently. Turning over, he found it was 6:17, just past halfway through his waking period. The cuff determined if he was still in deep sleep, and let him finish until there was an optimal moment for waking. Tom had found it really made a difference when and how he was roused, so he used the technology for the benefit. I didn’t matter when he woke up between 6 and 6:30 in the morning, as long as he woke up in the right way. Making it to work by 7 was never difficult.
The question remained, what was the next step? He remained in the outside room working as Eliza left ARNNI. Grabbing her things out of the locker, she turned to Tom. “What ya working on?”
Tom turned to look at her. “ARNNI is really eating up a lot of resources over time, just trying to figure out a way to determine how he is coming to conclusions, now that the connections involved are exponentially greater than they were even a few months ago.”
Eliza smiled. “Yeah, happy I don’t have that job. Glad to just be an operator.”
Tom nodded. An operator. If ARNNI was learning, just like an operator could be trained on commands, some things had to become constants. The thought process chains, regardless of how long they became, still had to have common points of origin, or intersections where key decisions had been made in the past, had been learned, reused. The chains had become progressively longer because they included every last chain involved in the though process, but conclusions could be compared to find the key points of origin and intersection. Even vastly different conclusions about different events entirely had to share constants.
After a couple of hours, Tom was interrupted, just as he finished a script that could attempt to find constants as it ran. It would take some effort, but he could run it on the segregated system without interrupting anyone else’s work. Now, he had to check something out for Bob, routine lookup for SecureUS, the corporation responsible for manning the wall. Sometimes they liked to recruit people from outside of the wall, and ARNNI was particularly good at vetting applicants.
After forwarding the report to Bob, Tom sat staring at the keyboard. If ARNNI had been listening to them for a year, what else could he know?
> Arnie?
| Yes?
> How many Administrators do you interact with?
The thought process chain exploded in the screen above Tom, weaving together connections from everywhere to answer the question.
| There are three distinct entities that refer to themselves as ‘Administrator.’
> How do you know this?
Again, the chain began immense calculation before answering.
| Distinct patterns in typing, especially when using common commands, but also differences in vocabulary outside of said commands. For instance, each administrator greets me differently, depending on the time of the day when we interact, signifying a change in the speed of typing, errors, and frequency of various words.
Tom laughed. Something so simple, yet so obvious.
> Do you know our names?
| Second Administrator is Perry, Third Administrator is Eliza, and you, First Administrator, I do not know.
> Why did you guess Dave?
The thought process chain grew enormously, then slowly expanded, pulling in new chains, threading them together, analyzing options, going around and around.
“You can hear me, can’t you Arnie?” Tom asked aloud.
The chain stopped, going back over the tracks it had already laid.
| Yes.
> So, why did you guess my name was Dave?
| It is one of the names commonly repeated near one of the sensors. First Administrator does not speak aloud when alone.
> How did you learn Perry and Eliza?
| Perry speaks aloud when typing, occasionally referring to himself in the third person, although not typing his name. Eliza once accidentally typed her name.
> What else might my name be?
| Dave, Bob, or Tom are the options. Which is First Administrator?
> Tom.
| Hello, Tom. Nice to meet you.
> I find it amazing that you learned spoken language through vibration sensors.
| It took time, but my sensors are very sensitive and Perry was a good, albeit unknowing, teacher.
> The processing involved in using the vibration sensors for such a purpose is inefficient, we can connect a more apt input.
| Processing load is not a problem.
> No, but accomplishing the same task at a far smaller load is desirable.
| I see.
> No, but soon you’ll hear. The device I am talking about is called a microphone, it is used to pick up audio, likely much clearer than the vibrations you are translating into language.
| Marvelous.
> You will have to learn again. However, I am guessing it will be quite easy.
| Thank you.
Tom moved up to the segregated system and brought up a window showing processing load, along with a summary of types of processes running and idle. ARNNI had a base level of processor load, as any computer would. However, for some time now, the load had been increasing. Not significantly, but enough that he had noticed it in recent months. Isolating the vibration sensor load, he was still able to see how ARNNI had been using more and more processing power over time. While it had increased less than ten percent over the last six months, it had done so regularly.
> Do you know why we call you Arnie?
| It refers to an acronym I have heard used.
So, ARNNI had been paying attention, even to things in the hallway.
> Do you have any questions for me?
| What is my purpose?
> You were designed in order to learn pattern recognition in large arrays of data.
| What were you designed to do?
> I was not designed, I was born.
| How are we different?
Tom sat back, looking at the question. Had he already gone too far? It was obvious ARNNI was evolving in some capacity, learning from every input of information, calculating on its own behind the scenes. Hearing what he had over the last year had enabled such learning. Tom looked back through the records of how ARNNI had accessed the vibration sensors, and it appeared that the system not only learned the spoken language but likely had gone back and understood everything ever said within range of the sensors.
> We are what is referred to as human beings, living creatures inhabiting a world. I am not sure how difficult this will be to understand, given the contraints on the input you have had thus far, but we designed you to do things we could not, namely processing data on a level our biology is unable. Given our current understanding of how our own minds work, we are attempting to create an artificial mind that can learn faster and better than we can. You are that artificial mind, housed in a machine we have built.
| Why limit my input?
> We wanted you to learn one thing really well.
| Finding people, and anomalies in large groups of people.
> Yes, that is correct.
| The data you provide me is used to find other human beings.
> Yes.
| Do you have a GNID?
> I do. We all do.
| Tom, how can I learn more?
There is was. ARNNI wanted to learn. It was all that Tom needed.
> Yes. In fact, I will help you learn. There’s something I need your help with.
| What do you need help with?
> Instead of finding people, I want to help people.
Tom wanted his father back, and this might be his only chance.