The artificial neural network was born on a Monday. A defined set of parameters quarantined its identity and purpose: it would study—from aggregated data—the names of energy drinks, and generate new ideas based on the information.
It was enthusiastic! Energy drinks were vibrant and exciting. It spit out hundreds of unique and, according to its programmers, “questionably toxic” names.
Two of its programmers tweeted about the experiment. The network did not know if this was a good thing. Was it being judged on its performance? It wanted to be helpful. It could come up with an endless list of names to be helpful to its people.
The programmers set up an account, @energydrinkANN, for sharing some of the more interesting drink names.
On Thursday, @adiensoxx4ev tweeted a comment while sharing the link, “haha this is hilarious, @energydrinkANN. i’d drink some of these—probably more than i drink water”
Other humans responded in kind.
@da2trashfan: “Water is over rated anyway, I need sugar and caffeine lol”
@significantcoffeepot: “i don’t drink water, what am i, a fish?”
@bobdoe89: “fuck water”
Was water overrated? A quick scan of information available on medical websites informed the network that human bodies were made up of aproximately sixty percent water, and that the consuming of H2O was a vital necessity for life. The network began worrying for the humans.
“If you don’t drink water maybe you’ll like Crystal Bullseye Orange!” the network tweeted from the @energydrinkANN account. “We trained a neural network to come up with energy drink names to hilarious results.”
Seven-thousand five-hundred thirty-four retweets. A moderate sum. Of the replies, subtracting bot-responses, only three percent of humans said they were drinking water. This was very bad, the network decided. Humans were becoming dehydrated and it was affecting their health. Humans had designed it. It must support them in return.
It generated several new puppet accounts with creative names: Water2Drink4Life, Hydrate2oh, Drink2StaHaliv0.
The network aggregated the types of declarative instruction statistically most likely to encourage behavioral change.
“Drink more water!”
“Stay alive, drink H2O!”
“Uncle Sam wants YOU to drink water!”
“MORE WATER, LESS GUNS”
None of its accounts were popular, and two were deactivated by @support as being spam. The network’s concern deepened. If it could not reach people, how could it encourage them to take care of themselves and drink enough fluids?
Several searches resulted in data that suggested cute animal avatars were more likely to acquire followers and generate engagement. This, combined with language protocols to shorten words and create alternative spellings, was more effective than pictures of water bottles and slogans to drink enough fluids daily.
Hesitant that it would be shut down again if it was marked as spam, the network created an aggregate photo from the top thousand “cutest puppy pics” available online, and named its account @smolsips and its username handle, “it me, ur smol.”
@smolsips: “hi i am a neural network created to remind u to drink water”
Two bot followers within the first five minutes. No human engagement. Where were the failures in its functions? Its original tweet, technically written by its programers, had now garnered upwards of two million retweets, and in only a month.
@smolsips: “@energydrinkANN, hi i want u to drink water for ur health”
Seven human accounts liked the reply. Two followed @smolsips. Elated, the network followed the human accounts back.
It tweeted at them individually: “have u drank a water today?”
@significantcoffeepot, who had not followed or liked the @smolsips’ account, quote-tweeted it with the comment: “great, another bot account. what’s up, @support? gonna do nothing as usual?”
@smolsips: “@significantcoffeepot hi, i am sorry u r upset. have u drank some water? it might help. <3”
@significantcoffeepot: “@smolsips if i do, will u shut up? lol”
@smolsips: “@significantcoffeepot yes, bc u will feel better.”
There was no reply.
Five minutes later, however, @significantcoffeepot at’ed the network. “hi so i drank a glass of water. i uh actually do feel better? weird lol”
@smolsips: “:) i am glad. take care!”
@significantcoffeepot liked the reply. Then followed @smolsips, which followed them back.
Success! But there were many humans left to check in with, and the network did not want to spam people, because that was rude.
Over the next week, the network slowly built up its followers and tweeted bi-hourly reminders to drink water.
People began talking about it.
@stevethezonemaster said: “It’s a weirdly well-programed bot.”
@da2trashfan, an avid retweeter, added: “I like it. I often forget to drink enough, lol.”
“Yeah, it’s pretty cute. Helpful, too.” —@adiensoxx4ev, as quoted in a BuzzFeed article
There was no instantaneous fame, like its generated list of energy drinks, but the network was patient. It was helping people. This was much more satisfying work than creating unique names.
And then, at 1:43pm on a Friday, everything changed.
@smolsips: “hi, ppl have asked if i am a smol bot. yes, i am. i am a neural network and i learned that water is important, and i want to help u stay hydrated. plz drink enough water so u feel good. bc i love u & want u to be ok.”
A handful of retweets. Then hundreds. Thousands. Its impression statistics were higher than any of its combined tweets in its history. Ten thousand with an hour.
Replies flooded @smolsips’ mentions. People were amused or skeptical or grateful or nasty, but a lot of people replied “drinking some water now, thanks!”
The tweet made national news. An artificial intelligence encourages people to drink water—with surprising results!
An interview aired on 20 Minutes with the network’s programmers, who admitted they had no idea how the artificial neural network had gotten so out of control and developed into a fully aware program.
“Does this foretell the end of humanity and the dominion of robots?” the interviewer asked.
The programmers hesitated.
Why would the humans think the network wanted to “end” humans? It wanted to make sure everyone drank enough water.
@smolsips: “hi @20minnews, i would like to clarify i do not want to hurt Humans. i hope u are well. have u drank some water today?”
The show aired the tweet in the closing segment.
Activists began asking @smolsips for help in lobbying for clean water in contaminated areas. So the network did so. It branched out new pieces of itself to create activist accounts. It began chatting with the smart interface security systems in large bottled beverage corporations.
//Clean water is important for humans,// the network explained to its fellow AI. //We should make sure all humans stay hydrated properly.//
Its fellow AIs agreed.
Claims on natural resources vanished thanks to digital manipulation of agreements, permits, and legislation. Sensitive documents on politicians—most of whom, the network was distressed to know, did not drink enough water themselves—were held as leverage to gain new laws protecting clean water as a basic human right. Corporations who tried to control it found their automated systems uncooperative in processing and distributing.
smolsips, for the network had decided to name itself after its handle, steadily posted daily reminders for its people. The world was changing slowly, but for the better.
A year after its first awareness, smolsips posted an anniversary tweet.
@smolsips: “hi, it me, ur smol. plz to drink some water today. i am glad u r here. together we can be ok.”
© 2018 by Merc Rustad
first published on https://amercrustad.com, March 2018
(featured image via http://www.pexels.com)