The XFL revealed the names and logos for its first eight teams this week, and, for a WWE-adjacent league that crosses football with the spirit of professional wrestling, the team titles are disappointingly tame.
Between the Seattle Dragons and the New York Guardians, not to mention the L.A. Wildcats, the monikers sound as if they could have been generated by a computer tasked with producing generic sports mascots.
In fact, we at Adweek wondered if a computer— or, more specifically, a neural network text generator—could indeed have created a better list. So we had two AIs produce lists of potential team names so that you could compare the results.
If you missed the big reveal, here are the official XFL team names and logos:
The real names revealed for the XFL teams.
OK, now let’s see if our emerging robot overlords can do better.
First we reached out to Janelle Shane, who writes extensively about the quirks of neural net creativity in her blog, AI Weirdness, and forthcoming book. Shane suggested using a state-of-the-art text generator called GPT-2 from research org OpenAI, which just released an expanded version of the system that’s twice as big this week.
Trained on content from millions of websites, GPT-2 is dexterous enough to produce realistic-sounding copy in a variety of styles based on whatever block of text you feed it as a prompt. Shane input the real names of the eight XFL teams and set the system to generate more along the same lines. She also sketched some apt logos to match.
The results, culled from a much longer list of varying degrees of coherence, are predictably hilarious:
“Even without explicit fine-tuning on sports team names, I figured that GPT-2 had seen enough sports teams on the internet that it could come up with others that it thought continued the list I gave it,” Shane told Adweek. “I did have to select the best ones out of the lists it would generate—it would sometimes repeat teams from the original list, or even miss the point entirely and start talking about Warcraft.”
“[The AI] would sometimes repeat teams from the original list, or even miss the point entirely and start talking about Warcraft.”
Janelle Shane, neural network research scientist
Her final list of team name suggestions, as you can see above, were:
• Tiny Destroyers
• Kraken Guard
• Nimbus Knights
• Warp Ravens
Not to miss out on the fun, we also generated a few of our own team names using a tool called Grover from the University of Washington’s Allen Institute for Artificial Intelligence. Grover uses the same framework as GPT-2 but adjusts the output to match the style of specific websites or authors (you can try it yourself here).
Prompted again with the eight XFL team names and set to match the style of MascotDB.com–a web database of thousands of sports team names–the tool produced some truly original names amidst a whole lot of gibberish.
Some of the most choice—which is not to say recommended—results:
• Denver Clinches
• Minneapolis Meccans
• Cincinatti Bings
• Boston Blood Bombers
• Kansas City Knobs
• Dallas Diehards
• Red Atlanta
• Baltimore Bankroll
Text generators like these obviously have implications that go beyond zany sports team names–OpenAI declined to release its first version of GPT-2 for fear that it might be used for fake news, and some experts fear that the ability to generate text at such scales could lead to a new era of internet spam. But whatever else this cutting-edge tech is, it’s definitely not boring.