There appears no definite or universally acknowledged definition of cocktail, even though anecdotally it's defined as any drink made by mixing two or more ingredients together, by for instance, the former Please Don't Tell (PDT) mixologist John deBary. That's right, coffee with milk and sugar, that's cocktail right there in your morning cup of joe.
As part of sommelier training, one has to know all the classic cocktail recipes by heart (how to make The Last Word?), understand the making as well as the flavor profiles of each and every ingredient (what is Drambuie made of?), and perhaps be able to explain the history, cultural references, and tales around iconic creations like Sazerac or Vesper. As a sommelier or beverage director gets more involved in putting together a beverage program, or a beverage list, one might start contemplating new alcoholic concoctions that best suits the restaurant or the neighborhood, or even the occasion. Such begs the questions, how could we best create new craft cocktails? And what makes a cocktail creative?
There is a popular misconception that a great idea strikes from out of the blue, when in fact, almost every idea, however groundshaking or creative, depends closely on what came before. Coming up with an idea, any idea, whether it be for new cocktails or building new rockets, can be summarized as combining existing ingredients or modifying a recipe to come up with something new. But is there a way to determine which set or arrangement of ingredients will make a great cocktail recipe?
To answer this question, we rely on psychology studies on creativity that suggests that creativity lies in the optimal balance between novelty and familiarity. Therefore, in order to understand what makes a creative cocktail recipe in a practical and precise manner, there are at least three questions to which we need clear and quantifiable answers to. First, how to define novelty and familiarity in the context of cocktail recipe generation? Second, how should we measure or quantify novelty and familiarity? And third, what makes an optimal balance between novelty and familiarity?
First, let us take the view of social networks, in that everything can be represented as a network. Let us view each recipe as a sub-network of ingredients within the network of the world of cocktails. In that sense, every ingredient of a cocktail recipe could be associated with one another either in one recipe or another. When two ingredients occurred frequently together with each other, for example, lemon juice and simple syrup perhaps, they are common associations whereas if two ingredients rarely or even never occur with each other they are uncommon associations. Then it's only natural to relate novelty to uncommon associations of ingredients, and familiarity to common associations. For instance, novelty does not necessarily come from choosing novel ingredients for the recipe, but rather from choosing ingredients that do not often appear together. Chili and matcha are common and familiar ingredients in recipes, but the combination of the two less so, therefore could be considered novel.
Now that we have a clear idea of what novelty and familiarity could mean in this context, let us construct a network of ingredients where each node is represents by an ingredient, and each edge between two nodes is assigned a value that indicates how common is the combination of these two ingredients in the world of classic cocktail recipes. More specifically, we calculate the ratio between (1) the number of existing cocktail recipes that contain both ingredients, take gin and topic as an example, and (2) the number of existing cocktail recipes that contain either gin or topic. The more frequently tonic tags along gin in recipes, the higher our indicator value of familiarity is, and the less novel the combination is. In this way, the balance between novelty and familiarity is now embedded in the values of the edges of any sub-network that represents a cocktail.
Here is an illustration of a semantic network of all the IBA cocktail recipes. Each recipe involves a subset of the nodes (ingredients) in the general network, which form a semantic sub-network. If the semantic subnetwork corresponding to a given recipe has N nodes, there are N(N-1)/2 edges in the sub-network, where the weight of each edge captures the strength of association between two nodes in the general network representing the world of cocktails hypothetically. Familiar combinations of ingredients have higher edge weights, meaning they are commonly found together in cocktail recipes whereas novel combinations of ingredients have lower edge weights, meaning their combinations are more unusual.
Now that we have constructed our network of cocktails, we could describe it based on any representative metric of its nodes and edge, for instance, the average weight of its edges. But to capture the balance between novel and familiar combination, we need to take a more comprehensive look on the semantic sub-network of a recipe...
That is, given that we have defined and quantified novelty versus familiarity, what is the optimal balance in-between novelty and familiarity?
There is a large body of research spanning psychology, biology, art, and behavioral science that has shown that archetypes, in other words, averages have inherent qualities and properties that make them appealing, thus termed the ``beauty in averageness effect``. This perhaps most well-known when it comes to human faces. An ocean of studies have shown this robust finding that humans find faces with average features more beautiful and attractive. It is also widely demonstrated for music performances, drawings, paintings, sculptures, as well as words and ideas in terms of creativity.
But how so, you might ask? Several explainations based on biology, evolutionary theory, as well as psychological fluency have been proposed. A more straightforward explanation associates it with the ``wisdom of the crowds`` phenomenon. It does appear across various domains that this ``beauty in averageness`` effect holds where quality relies on the optimal balance between various features or the optimal distribution of resources across various dimensions. For example, a beautiful piano performance is one in which the key strokes are neither too heavy nor too light, a creative idea is one that is neither too radical nor too banal, etc. Each recipe may be viewed as one attempt to find an optimal distribution or allocation of ingredients. Taking the average of a set of recipes could cancel out the small errors made by each recipe and gives rise to a distribution that is closer to optimal.