< Back

1. How did you make this AI/food experiment?

I generated AI recipe names a few ways. Using a combination of Google Tensorflow, Microsoft Azure, and Amazon Web Services (AWS) AI services, I gathered and "machine learned" a ton of recipes scraped from the web. "Machine learned" is in quotes because I'm a novice and "crude hacking" is a better description. I did the scraping with simple Ruby scripts similar to this one.

With a marginally ruly dataset, it wasn't too hard to generate realistic recipe names. The AI's great at searching, measuring, identifying patterns, and iterating. Recipes are just patterns.

For example, you can feed recipe ingredients to AI and ask it to return other ingredients that are likely to appear nearby. You can determine the liklihood that the next ingredient in a recipe is strawberry. Mix a bunch of these small operations, repeat several times, and presto: "original" recipe ideas.

I also used GPT-3 to generate two of the AI recipe names for this experiment. GPT-3 is dead simple. I used a prompt like this one except with recipe names instead of Hacker News posts.

The interactive part of the experiment is just a little toy app written in Ruby on Rails/JavaScript.

2. Why food and AI?

Short answer is it involves two overlapping interest areas: law + AI.


No, I'm not a food lawyer. But food and law sometimes collide in work visa cases involving culinary professionals. The culinary arts are actually baked into the law: "[a]rts includes any field of creative activity or endeavor such as, but not limited to, fine arts, visual arts, culinary arts, and performing arts." 28 C.F.R. ยง 214.2(o)(3)(ii).

U.S. Citizenship and Immigration Services (USCIS) digs deep into the creative validity of e.g., the best eel chef in Japan, a national champion pastry chef, or a Chinese restaurant and Asian produce grocery market. In this context, what's "creative" becomes very specific and very important.


Over the last couple years, I've had brushes with AI in various contexts. Legal research is doing some exciting stuff with AI. I've had some initial success using AI and ML on large volumes of legal documents I've collected.

In my limited experience, AI has an eery ability to generate unique insights and ideas. Perhaps counterintuitively, it's good at thinking outside the box. I'm interested in how this technology will collide with creativity as lines continue to blur.

The most interesting issue I've encountered is this: When does something cease to be art or creative because of AI's involvement? The issue is broader than food, or even art. For example, see this situation about whether a famous "virtual reconstruction of the ancient Persian city of Persepolis" qualifies as architecture (spoiler alert: USCIS says no).

For whatever it's worth, I generally agree with this article that argues, "human creative achievement, because of the way it is socially embedded, will not succumb to advances in artificial intelligence." But I believe AI can make creativity more accessible and diverse, propeling it to greater heights.

Inspiration for the experiment

The "human vs. computer" concept for my little experiment comes from Alan Turing, a British mathematician and computer scientist. In 1950, Turing theorized that computers are intelligent when a human user can't tell whether they are interacting with a computer or a human. The actual "Turing Test" is more nuanced than what we're doing here. But the core idea is the same: computer or human?