From household appliances to the most advanced industrial tools, the concept of artificial intelligence is everywhere… and photography is no exception. Are the technologies put forward by the photography industry really akin to AI? Let’s take a step back from our sci-fi dreams to assess the place of intelligent automation in image creation.
Most definitions of artificial intelligence (AI, or AI for Artificial Intelligence) agree that this concept covers “All the theories and techniques implemented in order to produce machines capable of simulating human intelligence” (source: Larousse Encyclopedia). The wide field covered by AI therefore stems from the fact that it is defined by its end rather than by its means. Let us add that if artificial intelligence is intimately linked to computing – by its history and by the technologies to which it uses -, its borders extend to cognitive sciences. While IT stores, sorts and processes data in order to solve known problems, AI has an adaptive character based on the use of models dedicated to solving new problems.
To deal with the current place of artificial intelligence in photography, however, it is necessary to rule out the idea – for the moment fictitious – of a robot endowed with a conscience and capable of experiencing similar feelings. to those of human beings. We distinguish this type of artificial intelligence, called “strong” (or “generalized”), from artificial intelligence called “weak” (or “narrow”) which is confined for its part to the reproduction of specific cognitive faculties. and solving specific problems. Let us take the opportunity to specify that all the AI systems in existence at the present time are considered weak AI. These are thus dedicated to very specific tasks, such as the recognition and understanding – to a certain extent – of speech and the identification of people or objects in photographs.
Photography as an experimental laboratory
One of the most frequent uses of the concept of artificial intelligence for the purpose of commercial communication concerns the cameras that equip our smartphones. While taking quality photographs previously seemed reserved for an elite capable of understanding the intricate workings of an SLR camera, AI today promises to turn anyone into a talented photographer. To do this, artificial intelligence simplifies the act of photographing by taking charge of the shooting parameters, and overcomes the hardware limitations of the tiny sensors and lenses of smartphones by optimizing the digital images captured. The AI is thus at work when the camera of a mobile terminal adapts the colors and the contrast of an image according to the elements identified in the frame, simulates a depth of field effect when a human subject is detected in the foreground, and merges photos with different exposure settings to compensate for large differences in brightness of a photographed scene.
If these photographic applications of artificial intelligence seem to us relatively common today, Google hit the headlines at the start of 2018 with the announcement of Google Clips. This compact and minimalist camera was surprising then with its ability to operate in almost total autonomy: placed in the heart of a home, it constantly kept an eye on the privacy of its user in order to automatically detect the highlights, then to compile photos and short videos captured in a dedicated application. To detect these moments, the Clips naturally appealed to artificial intelligence and more specifically to machine learning (or machine learning), an area of AI that allows a machine to learn to operate a task from a large volume of data and not through explicit programming.
Google Clips. © Google
Initially, Google would have taught its machine to recognize a failed photograph, for example because of the presence of motion blur or an object obstructing the lens of the device. This would have followed training in recognizing stable, sharp images and correct composition. Operated once the Clips in the possession of its user, the last learning phase was to focus on the recognition of familiar faces, as well as on a choice of captured moments favoring the diversity of the images rather than their redundancy. The commercialization of Google Clips has proven to be a failure – certainly because of its intrusiveness, perhaps also because of a lack of efficiency – but machine learning is now well and truly in the making. the origin of many automatic shooting systems claiming to be artificial intelligence.
In addition to having won the world of smartphones, AI is also present in classic cameras – their autofocus systems compete in ingenuity when it comes to detecting the faces and eyes of subjects – but also and above all within editing software. Widely exploited by the publisher Skylum, this trend has more recently won over its competitors, such as the famous Adobe. The new functionalities developed by these companies thus have the task of facilitating the performance of retouching operations previously considered complex or laborious. Rather than forcing users to perform manual cropping, software such as Luminar AI, Luminar 4, Photoshop, and Lightroom, for example, provide tools that can automatically recognize and select subjects, objects, and entire parts of images (such as sky and background), to modify or even replace them.
Skylum Luminar 4. © Skylum
A marketing discourse sprinkled with truth
Although the recurring use of vocabulary that is part of the lexical field of artificial intelligence (machine learning, deep learning, neural network, etc.) obviously serves as a marketing argument, it is now undeniable that AI is making a difference. integral part of new photographic technologies. And if it may seem somewhat abusive to affix the term “intelligence” to a camera capable of triggering when it detects a smile, remember that the AI is currently intended to perform very specific tasks, while that the idea of a true artificial consciousness still remains of the order of science fiction.
Beyond the debate on the degree of intelligence of said “artificial intelligence”, it is also important to question the place given to automation in the creative process. While it is certainly practical for the professional to have a fast and precise autofocus, and pleasant for the amateur to easily obtain well-exposed photos, it should be remembered that the value of a photograph lies in part in the choices. aesthetic and conceptual of its author. So let’s hope that in the future the actors of artificial intelligence will seek above all to correct the material defects of our cameras rather than the supposed shortcomings of the photographers.