It’s hard for Artificial Intelligence (AI) to know what makes a cupcake taste good without advanced labeling, yet an experienced cupcake maker can know intuitively.
With that in mind, if it is the experienced cupcake maker labeling different recipes and their associated taste on a scale of 1-10, the AI is then building a mathematical model of the bakers tastes.
In this case, we can see that it’s not likely that the AI will be able to make better original cupcakes than the baker, more likely that given a recipe, it can tell you whether the baker would like it or not.
In reality, as we start to look at the mathematics of things, it’s entirely possible that without enough data, it will still do that incorrectly. For example, the baker may taste two similar recipes one made with chocolate frosting and another made with orange frosting. Both rated as 9’s.
It then combines them thinking it will boost the score, only to find out that the baker rates the new recipe a 4. She doesn’t like the combination of orange and chocolate.
In the same vein, if a person sat with a cupcake tester and got feedback on 1,000 different cupcakes, I would think that even without a particular model, they would be able to predict new flavors that person would like.
People seem to think AI will take over the world and do everything. While there is certain applications it’s strong at, creativity is still the realm of humans.