AI Cocktails (Part 1)

What happens when a cocktail prompt goes to a large language model instead of a bartender? Eric Kozlik joins Rob Nester, creator of Drinking with Robots, for the first half of a two-part investigation into AI and cocktail creation. Originally released December 21, 2024, this episode considers what the experiments reveal about recipes, training data, creative direction and the experience behind a good drink.

Meet Rob Nester and Drinking with Robots

Rob brings a background in software development, an interest in machine learning and experience owning a bitters company to his cocktail experiments. His podcast gives human bartenders and AI the same themed challenge, then compares their drinks.

In the format described in this episode, the human bartender has a week to develop a recipe. The AI produces its proposal much more quickly. Both entries are assessed against a rubric that includes balance, creativity and how well the drink answers the prompt. The comparison creates a practical setting for asking what a written recipe can capture—and what it leaves out.

What Makes a Useful Cocktail Prompt?

The conversation explores why a tightly prescribed request does not always produce the most interesting drink. Eric and Rob discuss the need to steer an AI creatively, including stepping back from an unsuccessful prompt and approaching the idea from another angle. Treating the model more like a bartender receiving a brief can open different possibilities than asking it to assemble an excessively specific list of requirements.

These are observations from the experiments discussed in the recording, rather than a promise that any particular prompt will reliably produce a balanced cocktail. A plausible ingredient list still needs human judgment.

Recipes, Pattern Recognition and Experience

Eric considers the contrast between written cocktail knowledge and the practical understanding bartenders develop through service. Recipe books can encode years of experimentation and sensory feedback, but a working bartender also learns from guests, ingredients and the immediate circumstances at the bar.

Small decisions—adjusting a drink, noticing a guest or knowing when someone wants a familiar Martini—illustrate the kind of knowledge that goes beyond measurements on a page. At the same time, the episode asks listeners to take AI’s pattern recognition seriously: the written recipes it draws on are themselves products of human experience.

The first installment also looks at results from Drinking with Robots, including a live competition in which AI advanced against five of seven human competitors. Eric treats that result as a reason to keep investigating, while recognizing that a live challenge and a week of recipe development create different conditions.

Featured Cocktail: Krampus’ Coal Old Fashioned

For the episode’s featured drink, Eric asked ChatGPT to invent an original cocktail inspired by the Krampus Christmas tradition. The result is presented as an AI-generated proposal for critique, rather than a tested Modern Bar Cart house recipe.

• 2 oz dark rum
• ½ oz amaro, with Averna or Fernet-Branca suggested in the AI output
• ¼ oz blackstrap molasses syrup
• 2 dashes smoked chili bitters
• 1 dash orange bitters
• Charred orange peel, for garnish

Combine the rum, amaro, molasses syrup and bitters in a mixing glass with ice. Stir until chilled, then strain into a rocks glass over a large ice cube. Express the orange peel over the drink and use it as the garnish. The script does not supply a preparation ratio for the molasses syrup.

The original AI proposal also included optional activated charcoal. Eric specifically says he would ditch it; it is not included in the preparation above.

Eric’s assessment is deliberately mixed. He appreciates the thematic connections among bitter amaro, smoke, chili, molasses and orange, but questions the name and structure: with this balance of spirit and sweetened, aromatized ingredients, he sees more of a Manhattan-style drink than an Old Fashioned. He also questions some of the ingredients’ connection to the Alpine tradition. The point is to examine the proposal critically, rather than assume that an appealing name makes the recipe successful.

Keep Exploring

Find Rob’s show at Drinking with Robots and follow Rob Nester. For the broader context behind this series, listen to Episode 280: Cocktail Futurism.

This page covers Part 1 only. The next installment continues the conversation about AI, human experience and cocktail creativity.

Credits

Hosted by Eric Kozlik, with AI, large language model and cocktail-competition insights from Rob Nester, creator of Drinking with Robots. Editing and sound design by Samantha Reed. A Direct Fire Studios production, copyright 2024.