This Snack Does Not Exist – Computer generated snacks scored for market plausability – Evotik

By | 2019-07-14T13:37:31+00:00 July 14th, 2019|Uncategorized|
Hi Makers and Hunters,

I’m Ariel and I try to do fun stuff with linguistics. I hope some of the crowd is still here after this statement. Recently, I have been working a lot with snacks (Hope some of you return now).

As you may have seen, this little project creates random snack possibilities from food parts as our crawler found online, and, the part that interests me, grades them for plausibility of marketing success. The system can generate far fetched ideas such as “Robin Williams’ lock of hair ice cream special” or options that seem to be a better idea such as “Wasabi & Sriracha sauce flavored popcorn”. Our algorithm scored the ice cream with 0.105 and the popcorn 0.828 which makes sense.

How was this created?

This mini-project is part of the ACP (Artificial Creativity Project), trying to figure out new ways to boost creativity or perceived creativity in AI projects. The underlying concept is to check and score the plausibility of computer created associations.

For example, when trying to come up with a new ice cream flavor, a human might think of other food variations that may be interesting to add to gelato to create innovative flavors. A computer isn’t limited by what is currently considered to be food. However, if a computer goes totally wild, it might come up with inedible additions such as a concrete brick Vanilla or sand strawberry sorbet or a spider’s web pistachio; Or on the other hand, things that we might not consider to be a good idea to combine, such as a burning cigarette flavoured ice cream. Maybe that would pass under the current administration but who knows 🙂

So, you feed a computer with ideas about the world, about human opinions and cultural preferences, and try to limit the number of possible associations. For me, that’s the very definition of the opposite of thinking outside the box. Creativity is probably not achieved by limitations but by expansion of possibilities.

This project tries to find ways to score the plausibility of different types of associations. This specific dataset deals with various types of computer generated snacks of different kinds. We built a simple crawler to find interesting snack ingredients, and run our algorithm on them to see if they can work as snacks based (in part) on marketability, human preference, similarity to other snacks, linguistics and some other factors.

Lastly, the name of this project may imply that this it is some sort of a GAN take on thispersondoesnotexist. Well, other than for attracting some curiosity it is nothing of the sort. In short, it is more of our graphic manifestation to a concept association calculator, rather than a GAN.

Kudos to @kobyof for making this real.
Photos generated from Unsplash, Pixabay and Pexels.

Soon to be released:
– an api to return plausibility of food associations.
– a medium article with some of our hypotheses for making this.
– plausability calculator for other types of associations.

Ask us a Question!