Artificial Intelligence in the beauty industry: Impact on innovation

78

INTRODUCTION
“Machine learning” (ML) and “artificial intelligence” (AI) are two terms that have been gaining more and more popularity over the last half decade (1). They are often used interchangeably and sometimes even incorrectly, so let us start with a brief definition of these terms. Artificial intelligence is a term referring to software or hardware exhibiting a seemingly intelligent behaviour. However, what seems intelligent evolves as AI technology matures, making AI an ever-shifting definition. The checker-playing program written by Christopher Starchey in 1951 would hardly be considered “AI” by today’s standards! Machine learning is a subset of AI enabling software to learn through training, instead of being programmed manually. By processing training data, ML systems provide predictions that improve as more data is fed to them. Machine learning systems can be trained by supervised, unsupervised or reinforcement learning. Supervised learning requires data to be labelled (for example “dry skin” or “oily skin”) for the ML algorithms to learn, while unsupervised learning aims at extracting patterns from data without a label required (see Figure 1A and B, respectively). Reinforcement learning encourages a system to perform an action by reward and punishment, it is a common ML training method for recommendation algorithms and robots (see Figure 1C). Many “AI” are not ML based, you can think for example of a rule-based chatbot which would be hardcoded and does not learn. All ML solutions can be considered AI, however. One of the most advanced subtypes of ML is called deep learning (DL). Deep learning relies on artificial neural networks (ANN), a series of algorithms inspired by high-level inner workings of a brain. The architecture, the algorithms and statistical tricks used by ANN can vary depending on the type of ANN. Deep learning can solve complex problems but tend to require more advanced fine tuning and optimization (and therefore more resources) than simpler ML frameworks to achieve competitive results. It is why DL has been out of reach for most businesses, save for tech companies, until recently. The logical connection between AI, ML and DL, together with a few examples, is presented in Figure 1.