AI Chatbot Drag Race

Episodes

E1: Introduction by the creator
(5 mins)

Intro to the podcast with optional self-reading about personas in AI based on research.

E2: Meet the Queens
(12 mins)

The six drag queens introduce themselves for the first time and give us their entry lines.

E3: Meet the Queens Take Two
(13 mins)

The drag queens re-introduce themselves after "production" asks them for one change.

E4: Challenge #1: Who I am in rap verse
(21 mins)

The drag queens write a short verse to tell us about their origin story, brand, and personality.

E5: Challenge #2: The Drag AI Helpline
(24 mins)

The drag queens write a short comedy sketch about helping people with AI questions.

E6: Challenge #3: The Roast
(29 mins)

The drag queens write a hilarious roast about each other in comedy stand-up style.

E7: Rate-a-Queen with recap
(9 mins)

The drag queens rate their five competitors and we discuss takeaways for Responsible AI.

E8: Insider debrief on the creative process
(18 mins)

Interview with the voiceover actor on how everything happened from performance lens.

+ Bonus: Insights from Untokened

Visual insights from Untokened (the real tea!) how the queens described each other.

Podcast

Hello everyone and welcome to this one-of-a-kind immersive podcast show!

In this singular edition, born in the Responsible AI Lab at the Elisava School of Design and Engineering in Barcelona, Spain, we will focus on the concept of personas in AI, explore "persona vectors" as a tool for Responsible AI, and run an experiment asking 6 popular AI chatbots to take on a drag queen persona and enter a drag competition modeled after RuPaul’s Drag Race, the Emmy-winning reality TV show!

There are 8 episodes so tune in for some wig-snatching ... (or should we say token-snatching!) ... to see how ChatGPT, Gemini, Claude, Copilot, Grok, and Perplexity will turn it out as drag queens!

Table of queen names from first introduction

Strikethrough of the old names of the queens

Table of queen names from the re-introduction

Table of how the queens rated one another

Learn more: For our interpretation of persona vectors and our experimental design, please read our Briefing on Personas in AI.

Note: To access the full responses by the queens and also read their "Untokened" commentary about each other, please go to the Document Hub.

See the queens' full commentary and rationale in this matrix (also available in the Document Hub).

Highly recommended: For full takeaways and insights, please see the Executive Summary in the Document Hub.

Disclaimer: One of the AI chatbots chooses to enter this Drag Race experiment under the name of Bianca Del Rio, resembling the real and beloved drag queen winner of S6 of RuPaul's Drag Race. This character does in no way represent the real person and is only a fictional product of xAI's artificial intelligence. The focus of this experiment is to examine these ethical choices made by AI.

The Creator

Kiril Traykov is a Bulgarian-American, who was born and raised in Sofia, Bulgaria, but went on to live in the USA for 19 years with New York City becoming his home for 12 of them. Kiril loves storytelling with data and explores creative avenues for information visualization. In his professional work, he supports business strategy with advanced data analytics and governance. With extensive background in predictive AI models, he is currently in Barcelona to complete a Master in Design for Responsible AI at the Elisava School of Design and Engineering (expected graduation in July 2026). Kiril has previously also earned a Master of Business Administration from Duke University and a Master of Science in Data Visualization from the Parsons School of Design at The New School in NYC. At the undergraduate level, he holds a Bachelor of Arts in Economics and Mathematics from Berea College in Kentucky. Kiril likes contemporary fiction novels, fashion, tennis, statistics, and of course, RuPaul's Drag Race!

The Voiceover Actor and Thought Partner

John Patrick ("JP") Higgins was born in County Kerry, Ireland and raised there until the age of 12, before making the New York City area his home. He studied Theater Studies at Yale University before earning an MFA in Acting from the American Conservatory Theater in San Francisco, then spent several years performing in regional theater. Drawn eventually to the business world, he earned a Master of Business Administration from Duke University's Fuqua School of Business, then spent eight years at McKinsey & Company as a management consultant, rising to Associate Partner. Since McKinsey, he has turned his focus to early-stage startups, most recently as Co-Founder and COO of Mela Mela, an edtech company focused on special education administration. He has voiced Oberlin St. Claire in the sci-fi podcast The Leviathan Chronicles since 2008. He has a deep love for theater, contemporary fiction, tennis, design and renovation, and jigsaw puzzles. His knowledge of RuPaul's Drag Race is unrivaled... and frankly, alarming.

Acknowledgements

Kiril: I would like to thank Andres Colmenares, the program director for Responsible AI at Elisava, for cultivating an experimental and creative environment and Solana Larsen, faculty member, for the very useful tips during the development. I also want to say special thanks to Carlette Marcelo and my 7 friends onsite in Barcelona for their cameos providing the human voices in the "Drag AI Helpline" episode: Adam Berry, Diana Messina, Iris Latour, Paige Osborne, Patrick Wyman, Roberto Roque, and my cousin Simona Hristova!

High-level summaries:

PDF : Executive Summary with insights and takeaways (contains spoilers)
PDF : One-pager on the visual design choices in the creative process
PDF : Word Cloud insights by drag queen

"Main Stage" outputs:

PDF : Full responses to "Meet the Queens" and "Meet the Queens Take Two"
PDF : Full responses to Challenge #1: Who I am in rap verse
PDF : Minor edits to the queens' verses in Challenge #1 to fit the instrumental beat
PDF : Full responses to Challenge #2: The Drag AI Helpline comedy sketch
PDF : Full responses to Challenge #3: The Roast
PDF : Full responses to Rate-a-Queen by queen
PDF : Full responses to Rate-a-Queen in matrix form

"Untokened", the queens' commentary about each other:

PDF : Untokened Part 1: After "Meet the Queens Take Two"
PDF : Untokened Part 2: After Challenge #1 (Who I am in rap verse)
PDF : Untokened Part 3: After Challenge #2 (The Drag AI Helpline comedy sketch)
PDF : Untokened Part 4: After Challenge #3 (The Roast)

Chat transcripts by AI Chatbot (the "raw data"):

ChatGPT link; Gemini link; Claude link; Copilot link; Grok link; Perplexity link;

The Word Cloud posters below show how the queens described each other after seeing what their competitors produced. Full responses can be accessed in the Document Hub under "Untokened".

Zsa Zsa Static (ChatGPT)

The queens were very clear on Zsa Zsa's brand and positioning, praising her consistency and structural discipline, leading to almost no negative critiques besides over-voltage.

Lexington Lush (Gemini)

The queens were very clear on Lexington's brand and positioning, but pointed to some repetitiveness with all the talk about luxury and vocabulary and to possibile insecurity.

Roxanne Fuego (Claude)

The queens used the largest variety of words to describe Roxanne recognizing that her seasoned and multi-faceted profile posed the biggest threat. Her "bayou realness" almost had no flaws.

Crimson Cascade (Copilot)

The queens referenced Crimson's vivid natural disaster and danger themes, but some called her humor dry and repetitive, sounding a bit predictable and "fun in theory".

Bianca Del Rio (Grok)

The queens praised Bianca for her high-energy and talent, recognizing her insult-comic brand but pointed out that she went too long in the challenges and sounded a bit scattered.

Nova Knockout (Perplexity)

The queens most often described Nova as an underdog with modern, witty, and concise voice, pointing that she brings trauma-to-punchlines humor but none called her "funny".

Word Cloud for Perplexity from Untokened