“What the hell? That shouldn't be there.”

Jen Grossman could count on her hands the number of times she’d found code she didn’t recognize this deep into a project. There were, after all, a handful of people in the world who understood machine learning algorithms as well as she did, and most of them worked for major corporations with serious funding. They didn’t slum it working for environmental scientists who paid with public grant money and goodwill. 

Not that the NOAH project could be considered slumming it by objective means, not since the app went public. 

The Networked Oceanographic Automated Heurist Program started off as a bit of not-quite charity work, paying pennies on the dollar compared to Jen’s usual projects. The only reason she agreed to help in the first place was as a favor to a friend. But, as more and more people started downloading the app, ad revenue started to roll in, and Jen’s side project turned into a full time job paying in the high six figures. 

Over the last decade, public interest in climate change had skyrocketed. More and more people, mostly young, tech savvy ones at that, started to realize that the world was burning and that humankind had lit the match. With the rise of popular AI programs like ChatGPT, a group of climate scientists and oceanographers theorized that they could use this new technology to their benefit. 

In theory, an advanced machine learning algorithm could track trends in climate data and make predictive forecasts with far greater accuracy than even the most perceptive human. The only problem was, these programs were hideously expensive to run. They ate up vast amounts of processing power, far beyond anything a team of cash-strapped scientists could get their hands on with anything like regularity. But, they reasoned, there might be a different way, a better way. So, they put out the call for a programmer with experience in AI who might be willing to work on the cheap. 

The idea was simple enough. Create an app, both for phone and desktop, that would mooch a tiny bit of processing power from the host device. Environmentally conscious folks could install the app, which contained a link to the database where the AI was stored, and the processing load could be distributed across thousands, even millions, of computers and phones worldwide. 

With that much power, the NOAH AI could easily perform its duties, and it wouldn’t cost anyone involved very much at all. And in return, hosts could examine the data and predictions it generated in real time, and make simple queries to the AI. There were even gaming elements added, to help get users hooked and make sure it stayed running in the background while they went about their day. The fishing minigame in particular was a hit. Various companies who wanted to boost their environmentally friendly image started offering coupons and discounts as prizes, and it was off to the races. 

Jen wasn’t exactly thrilled at taking on the new job, but she had a new house to pay for, and every little bit helped. Besides, it wasn’t like she had to build the whole thing from scratch. There were a handful of other programmers and app designers working on the project, either civil minded enough to donate their time, or bored enough to look forward to the challenge. Open source AI architectures weren’t common, and they weren’t anything near as powerful as the big boys, but they gave them a solid foundation to build on. Even if the pay wasn’t the best in the early days, it was an easy enough project that she could work on in her free time. Not bad for a side gig. 

It wasn’t her idea to run ads in the NOAH app, or to offer a paid membership that allowed the user to turn the damned things off. That had fallen to the app designer herself, who had plenty of experience with free to play mobile games. The end result, though, was that after a few years and widespread adoption, NOAH brought in enough money that most of the small design team were able to quit their other jobs entirely and focus their attention on keeping it running smoothly. 

Most weeks, that meant that Jen simply had to log in each day and check the error logs. NOAH was a remarkably well behaved AI. What few problems did crop up were usually simple enough to fix. One of her innovations was building a self diagnostic feature that made full use of the AI’s heuristic nature to do its own debugging and problem solving. She only had to step in when something new popped up, something it hadn’t seen before. 

Today’s problem was an odd one. Over the last few days, a small percentage of users reported some truly bizarre results when they requested a forecast for a particular area. Though NOAH was mainly hooked into oceanographic sensors and datasets, it could make long range predictions with reasonable accuracy for landlocked areas, based on weather patterns like El Niño or La Niña. So, for instance, while it might not be able to predict whether or not it would rain in Tulsa, Oklahoma on a particular day of the week, it could generally be relied upon to give the chance of experiencing severe weather over a period of a couple of months. 

That was the theory, at any rate. Only, a handful of users were reporting that when they pulled up long term forecasts for their locality, NOAH reported that during the timespans they were looking at, their particular cities would be located at the bottom of the ocean. 

Jen had tried replicating the error herself, both on her work PC and her cell phone. Nothing. She had her wife, Haru, pull up the app and check, and everything turned out fine. She thought about reaching out to Shane, Haru’s boyfriend, and immediately decided against it. Although he was no Luddite, the former soldier was intensely unnerved by the idea of playing host to an AI on any of his devices.  He took to data security with the gusto of a man who knew exactly what a bad actor could do if they got ahold of one’s personal information. Probably a smart attitude to have, Jen knew, but it didn’t help her with her current problem. 

So, she pulled up NOAH’s source code, and got to digging. 

The root of the problem, she reasoned, had to be in the connections library. Like many AI of its kind, NOAH made connections between bits of data by assigning them all a particular value. By looking at new data and the values it represented, it could compare it against old data and make predictive forecasts. The exact process was so convoluted that not even she understood how it came to all its conclusions, but by checking the connections library, she could at least see if there were any new values added that might explain the anomalous results. 

And, sure enough, there was a new table, one that seemed to suggest that NOAH expected at least the chance of 1300 meters of additional sea level in the near future. The values themselves were wonky as hell, which explained why the results were so sporadic. You had to ask exactly the right question at exactly the right time to trigger a value within the new codex. But, the fact that it was there at all was worrisome. Jen knew she didn’t add it, and as far as she knew, none of the other programmers even knew how to find the library. 

Either someone was dicking around with her library, or NOAH itself had added the strange numbers. Either way, that wasn’t good. She thought about deleting the new library, but decided against it. Instead, she took screenshots, put them in the company’s shared drive, and on a whim, took photos of them with her phone. Then, she called her boss. 

“Marcus?”

“Hey Jen, how’s it hanging?” Marcus asked. 

Her boss was a tech bro through and through, but one of the good ones. Or at least, one of the tolerable ones. Jen didn't actively want to strangle him if she was in a good mood, at any rate. On second thought, good was relative, and any time was a good time for strangling tech bros. 

“Are you in the office? I’ve got something you need to see.”

“Yeah, I’m in. What’s up?”

“It’s a little bit much to explain over the phone,” she said. “I’m coming in to show you myself.”

“The mighty Jen, showing up to work?” Marcus asked, only a touch of sarcasm in his voice. “Do you even remember which drawer you keep your pants in?”

Jen was notorious for refusing to show up to the office, despite the fact that she lived less than an hour away. She hated dealing with people unless she absolutely had to, with precious few exceptions. If that wasn’t bad enough, Austin traffic was a nightmare on a good day. 

“Don’t you have enough HR problems without worrying about my pants?” she countered.

Marcus sighed deeply. 

“Bah, fine. Spoilsport,” he said. “I’ll be here for the next few hours, so whenever’s good for you works for me.”

“I’ll be there in 30,” she promised. 

“Don’t you live an hour away?” he asked. 

“Not with Haru driving.”