"Rayburn Wallace!" Ray's father Andrew entered the Simulink lab. "Why are you scurrying around the lab dragging a dismantled robot..." his father stopped mid-sentence seeing Nano linked to the surgical, diagnostic cable. Then continued, "Dr. Hastings said you ran him over stalking about in his shadow. What caused such a disruption?" A man with a medium build, his father, appeared taller as he looked down at Ray. Unwavering, Ray felt the demand for a thoughtful and clear-headed response.

Ray stood. Head down and dropping eye contact to his wringing hands, he fumbled for thoughts. His first thought was a counter-clockwise hand gesture to clear the AirDesk display. Ray looked up at his father without response.

"Working on Nano, eh?" His father circled the room inspecting the robot at its center. "So, what is it this time, a new alg-mode or neural construct?" His father's voice softened from its hard tone. The question reminded Ray of how his father spoke with other scientists.

Andrew Halloway bent over to inspect the robot's cabling and connection. His ruffled dark brown hair floated above a matching beard, speckled in gray. Dressed in a standard-issue lab coat, white over dark brown pants, his eyes curious.

"Starting fresh dad," Ray said, entertaining his dad's curiosity. "The processor bank started missing neuro timing within the cross matrix. Nano's emotional directives cortex was collapsing fast."

"Third time this month," Andrew commented.

"Yes, I hope I got it right this time," said Ray drawing a breath.

"Scccaaarrry," came a voice from a rolling ball, whirling around Andrew's feet and up to Ray. The voice was small and tinny. About the size of a bowling ball, it spun in place as it spoke. It turned again after it spoke, making a whirring sound as air forced through a bottle. Ray was only a little surprised to see Wrimo after separated in the hallway traffic.

"I guess Wrimo is going to be busy," said Andrew taking a step back to assure his footing and looking down at the orb. Wrimo rocked side to side hearing his name.

"Maybe half the time?" Ray estimated. Training took time, and Wrimo helped, complicated given its extreme processing limits.

A gift on his fourth birthday, Wrimo, was simple. Off the shelf, it was a combination kickball and parent tracking system. It would follow, track, and intercom with remote users when requested. Monitoring didn't help his parents due to Wrimo's limited communications range. Ray still adventured without permission, missing tutor appointments, social gatherings, and meals. The most used mode in the original program involved kicking and having it roll back for another punt.

When Ray turned five, his father upgraded with simple speaking and learning intelligence. Ray laughed, hearing Wrimo's first words, "Hi, let's playyyyyy!" Ray never played kickball again as Wrimo became more of a younger brother. It was able to learn simple language tricks like one teaches a bonded parrot or raven. By age six, Ray became fascinated with improving the algorithms. By eight, the step change in interaction was visible. RayOS was the new operating system holding Wrimo's construct.

Over the last two years, Wrimo's compact architecture housed a capable intelligence. His father encouraged Ray to describe his latest improvements and publish results. Wrimo joined a competition and earned 1st place for its processing class. Those awards inspired Nano, now dangling in the middle of the room. A still simple but personable-looking maintenance bot. Nano's articulating arms allowed cleaning, fetching, and other small tasks. The intermediate processing class allowed Ray to extend his architecture. Within a week, a re-imaged Nano emerged, clear of monitoring and starter features.

Little by little, Nano learned and developed. After each upgrade, Wrimo helped retrain Nano's neural encoding. Wrimo became Nano's initial tutor.

"Half the time, eh?" echoed his father pondering. Either Wrimo was becoming a better teacher or Nano was becoming a better learner. He assumed the later and made a mental note to check the neural acclimation analysis. "Be sure to upload his ratings each night before bed. Yesterday surpassed upper intermediary level. If you miss a day, you may have to submit to degenerative testing."

This text was taken from Royal Road. Help the author by reading the original version there.

"I know, dad," Ray dragged.

"Good you caught Nano's upgrade when you did. Fascinating work Ray. Oh, and you owe Dr. Hastings an apology," Andrew said with a confident smile.

Detecting a positive change in the conversation, Wrimo weighed in, "zzZZUUMMm." The sound spun around the room as Wrimo spun.

Ray's determined expression lightened. An added benefit was that his dad did not examine the server credits required for the job submission. If his dad had checked, he might not have noticed Nano's recompile was passed his allowance. Ray used the maintenance core instead of the laboratory's primary central processor.

The lab housed one of a few GigaGoogol facilities capable of a billion googol operations per second (a billion times one followed by 100 zeroes, 1e9 * 1e100 = 1e109). Quarterly, a council approved projects and adjusted processing priorities. The lab owners, including many staff members, approved the allocations by vote. Usually, based on need and revenue, it was not uncommon to see future-critical technology and time-critical projects to get significant assignments. The computing facility allowed the lab to engage in extensive data processing experiments, including citizen-wide information analysis.

A standalone processing block supported maintenance, landscape, security, training, records, and family projects. This ensured quality of service for the lab untouched by high priority projects. Ray discovered a way to leverage all unused processing within the bank, 1/2 million Googol of free capacity (e.g., 0.5e106) was almost always available. Using a master account linking several accounts, he had access whenever needed. The technicians grossly overestimated the amount of processing required for this block. Ray was pretty sure no one discovered the master account and confident it would go well.

Ray was more certain three weeks ago. But at that time, the system fell to a 7% average capacity. Less critical systems shut down. Some, notably the dinner service, began to malfunction. First, plate stacking bots stopped while the automatic dishwasher didn't. Plates overloaded interim filing dispensers despite signaling at 50%, 75%, and 85% intervals. Alarms sounded from the containers. Loud, warning tones that no one had ever heard before, and some mistook to be fire alarms.

The dishwasher machines did not receive the feedback and carried on. At 100%, plate dispensers began ejecting dishes to make room for the incoming plates. Some slid ten feet or more into the place. Frantic screams added to the turmoil of crashing dishes. Several of the first plates struck an unsuspecting scientist and their family. Minor injuries were no less embarrassing. For weeks, the SimulCast repeated looks of shock and surprise on the faces of diners. These accompanied by flying plates people running from the path of the trajectories. The android chef shutdown the cleaning line, ending about five minutes of chaos.

Ray learned even short outages spawn big fiascos. The depth of the subsequent inquiry and the life of the Simulcast show, "Investigation into the Radical Dish Dispenser and Other Maintenance Systems," surprised everyone. After two weeks, the investigation filed the final report with recommendations. The root cause was found to be a subsystem control issue. Instead of using the remote core, they designed a kitchen local control extending the work of the android chef. Discovery of a deeper problem within the server maintenance system was found to be a flaw in the memory architecture. This issue extended to all the lab's AIs and projects. Finally, the most vocal complaints were families that experienced the Radical Dish Dispenser. A Pareto analysis showed complaints were from:

> 68% inaccessible dishware

> 17% eating from small saucers

> 11% frequent trips to the foodservice machines for refills

> 3% other

> 1% injuries running or moving to dodge plates

> 1% of injuries from those hit by plates

The investigation also identified Ray's algorithm to have exacerbated the scheduler bug. Being only ten, Andrew allowed Ray to be questioned and, after several discussions, gently reprimanded. The lab supervisor, the maintenance supervisor, and the kitchen manager all filed complaints. Ray had to listen to the whole report and each complaint. All but the kitchen manager appeared amused. Finally, talk with his father ended in a two-week grounding from Simulcast privileges. Ray felt terrible for embarrassing himself and his family. Still, he couldn't help but smile when he saw kids giggling during the replays. Also, the lead developer that found and fixed the scheduler bug helped Ray set up the master account. Ray now had access to a massive processor as long as the other systems were not using it.

The outcome was beneficial. The lab supervisor knew the AI was intact, undamaged, and kitchen better contained. The maintenance supervisor knew restrooms and biocontainment were safe from a latent bug. The owners and the AI group knew a dormant memory boundary issue would not impact core projects. The system team had a memory subsystem to support extra processing.

"We need to get home to dinner, ready?" asked Andrew. "Be there in a minute," said Ray. Nano's upgrade complete, time to boot.