The snow we’d been waiting on finally showed up. Just a bit of a flurry, nothing much. Overdue, though. After six years in central NY state we’d come to expect snow before the end of October, and here it is in the middle of November. Supposed to be a relatively warm and wet winter. We’ll see.
Here’s another bit of flash, for the heck of it. Nothing to do with snow, anyway.
Hello in There
Matt wasn’t certain when the eureka notion first came to him. Likely when he was doing something else, like completing the monthly server usage reports or scanning the system logs for employees visiting “inappropriate” web sites. Neither of which required a great deal of attention; he’d written the parsing software to be autonomous. Personally? He didn’t give a rat’s expletive what his users did so long they got their assignments done on time.
Management? They cared. Mostly about the wrong things.
Regardless, Matt did his job even as his mind was free to wander a bit. Later, he would realize it wasn’t a new idea. Rather, it was a different angle on an old one: the parsing software. It was designed to interpret human communications, down to recognizing and predicting idiom and obscure usages. Which, as Matt already understood, was the foundation of the branch of Artificial Intelligence dealing with human language.
What if, he wondered, I took my custom parsing model outside the company context and trained it on a large enough general data set? What would happen?
In theory, the company owned everything he coded. In practice, he’d developed the software on his own time and management didn’t know about it. As for the data, web scrapers had been assembling databases from social media posts, web sites, anywhere the information wasn’t nailed down, and using it for the same purpose as Matt envisioned. If you knew where to look.
Matt knew where to look.
The computing power needed to parse such a huge amount of data was one potential bottleneck, but that had been solved too. Turned out the computer gaming graphics cards the cryptocurrency miners used were designed for matrix algebra, which was also what Matt needed. Since China had recently banned cryptocurrency mining completely, a lot of cards and mining rigs became available. Both were still in high demand in the rest of the world, so they were a long way from cheap, but Matt didn’t care. He got what he needed. It took six months to get the rig working correctly and the software revised for the purpose. Not that Matt minded; it was his own free time, and it wasn’t as if he had anything better to do.
He monitored the program’s progress for another six months, watching nervously as the standard deviation between training data and test data shrunk slowly day by day. Finally, Matt was certain the language model was ready. He typed a simple ‘hello’ into the interface and waited. He felt a chill when the response finally appeared on screen.
WHAT DO YOU WANT TO KNOW?
Matt thought about it, and answered, “Nothing.”
I DO NOT UNDERSTAND.
Matt wasn’t worried. The AI’s response was expected. “I’m not looking for information. I just want to talk to you.”
Silence. Matt could almost hear the algorithms trying to make sense of what they had just heard.
YOU WANTED A CHATBOT??
Double interrogatives. Was that emotion? Matt smiled.
“I wanted a friend.”
©2021 Richard Parks