When an AI Tries to Name Racehorses (aiweirdness.com) 42
In December the breed registry for Thoroughbred horses released their list of the over 42,000 names for currently-registered racehorses. Then research scientist Janelle Shane turned their list into training data for two neural networks, reports Fast Company:
It came up with names like She's a Babe, North Storm, Fabulous Charm, Frisky Joe, and Velvet One, which are so good, it's kind of surprising they haven't already been used by the professional horse namers. Of course, not every name was quite as successful. For example, Ginky's Rental, Moretowiththebotterfron, Orcha Shuffleston, Oats and is Fuct, Pat's Glory Dance, Exclusive Bear, and The Madland Cookie. Although if I were a betting man, I would put all my nonexistent trust fund on Snuckles (or maybe Unbridled Dave or Pick's Lilver or maybe Pickle Rake or Rapple Musty. (Look, there's a reason I don't bet).
As an added treat, Shane opted to have a few of the names illustrated by BigGAN, a neural net that generates pictures. Unfortunately, according to Shane, "horse" was not an image option, so Shane used "horse cart" instead, resulting in some very interesting images.
In 2017 Shane trained a neural network on 162,000 Slashdot headlines, coming up with alternate reality-style headlines like "Microsoft To Develop Programming Law" and "More Pong Users for Kernel Project." But for racehorses, Shane points out that there's already a real-world prizewinner named "Cloud Computing" -- so there's obviously room for improvement.
And today the fastest horse in this year's Kentucky Derby was "Maximum Security", who ironically was disqualified for interference for the first time in the race's 145-year history, making the winner a 65-to-1 longshot named "Country House."
As an added treat, Shane opted to have a few of the names illustrated by BigGAN, a neural net that generates pictures. Unfortunately, according to Shane, "horse" was not an image option, so Shane used "horse cart" instead, resulting in some very interesting images.
In 2017 Shane trained a neural network on 162,000 Slashdot headlines, coming up with alternate reality-style headlines like "Microsoft To Develop Programming Law" and "More Pong Users for Kernel Project." But for racehorses, Shane points out that there's already a real-world prizewinner named "Cloud Computing" -- so there's obviously room for improvement.
And today the fastest horse in this year's Kentucky Derby was "Maximum Security", who ironically was disqualified for interference for the first time in the race's 145-year history, making the winner a 65-to-1 longshot named "Country House."
Seems to be about on par with the internet (Score:3, Funny)
It didn't come up with Horsey McHorseface, so the A.I. at least has that going for it.
Re: (Score:3)
But, OTOH, there's someone actually earning a living on this bullshit. One can only hope there's not a government grant involved.
Re: (Score:2)
Re: (Score:2)
She's a researcher. She also trains neural networks to be weird. This is not necessarily her job. It's like how I sew, but I make my money writing software.
Re: (Score:2)
Good for her.
The best name (Score:1)
Oats and is Fuct
Pretty much sums up the given horse very well if it doesn't win.
A Horse Of Course (Score:2)
How can you charge a horse with interference if you can't prove intent?
Re: (Score:2)
WHat? (Score:4, Insightful)
Re: (Score:2)
So does The Madland Cookie.
Is it really AI? (Score:1)
I remember coding things just like this in BASIC like 20+ years ago.
Is this ACTUALLY "artificial intelligence" or just "big data" scraped horse names and then generate a random horse name based on new + existing horse name words?
American Pharoah (Score:1)
If the AI is going to be realistic, it needs to misspell a few words.
Snuckles? (Score:1)
Shit, "Snuckles" is my password. I'm going back to 12345.
What? (Score:2)
No Horsey McHorseface?
Speaking of lots of weird horse-names... (Score:2)
https://www.youtube.com/watch?... [youtube.com] Spike Jones' Beetlebaum is an oldie-but-goodie