Artificial intelligence, we hear about it everywhere and sometimes for anything. But what is she really capable of? Here is a list of concrete uses of AI.
Artificial intelligence as it exists in the collective imagination is a fantasy of science fiction. Forget about HAL 9000, forget Asimov's robots, forget the Terminator T800, our artificial intelligences are still very far from the count. At the end of 2018, Yann Le Cun, head of IA research at Facebook, claimed that "artificial intelligence is not even in the brain of a rat".
The illusion could deceive some, but Google Assistant, Alexa and especially Siri are in no way able to really converse with us. These personal assistants only partly understand certain terms that punctuate our sentences to get predefined answers on a corresponding service. All are still far from passing the Turing test, even Google Duplex, the conversational AI presented at Google I / O 2018 or its Chinese rival developed by Alibaba.
However, the AI (or AI in English for artificial intelligence) is present everywhere. Few tech conferences have not spoken the word at least once. So what is it really and what is it for?
What is artificial intelligence?
We should not see the current AI as the works of SF describe them, but as devices having learned of themselves through machine learning. Thanks to a neural network and a large amount of data, an artificial intelligence is able to recognize certain elements by discerning the differences between two test responses and finally to improve on its own.
It is then up to the programmer to refine the set so that the answers correspond to those desired. For example, it is possible to teach an AI that everything that is small, has hair and a tail is a cat, but it may be wrong if shown a dog. It will be necessary to refine the set to add the parameter "meow" (for example), otherwise the answers may be too broad.
Ears pointed, mustaches, meows constantly to go in or out … it's a cat!
Still, today, artificial intelligences are very specialized and there are about as many as there are areas of research. The AI of Google Photos is for example very impressive, but it is not she who will answer our Google Home queries, and vice versa.
Also, there are currently a very large number of projects around artificial intelligence today. Concrete projects, sometimes already accessible to the general public, which have many uses. Some are not very useful everyday (like this AI that can create faces), but others are much more practical in our everyday life.
What is artificial intelligence for?
The AI has multiple use cases. It makes it possible to improve certain already existing elements, or to replace material limitations. The Google Pixel are the perfect example with a particularly sharp portrait mode able to determine what is in the first and second plane of a photo, almost solely in software. But let's see what are the possibilities of artificial intelligence!
For photo and video
As said above, imaging is a field conducive to the integration of artificial intelligence, with a desire to offer the best qualities in products always smaller and smaller. Google is an important player in this field, with its portrait mode, but also and especially its night mode.
Much more effective than a long pause, Google's "night sight" mode not only retrieves information in completely clogged areas, transcribing them without digital noise, but also to adjust the colorimetry. In the idea, the camera of a Pixel 3 would be able to find your red socks in the dark, much more effectively than your eyes.
You would have guessed the color of the pavers?
Of course, Google is not alone in this sector and many manufacturers today offer a night mode more or less convincing, but that we could often imagine feasible with a simple algorithm coupling a simple photo to a break long and a good digital noise reduction system. Few are able to make this color accuracy like pixels for the moment.
Each is his domain nevertheless, since others like Oppo or Huawei rely on the zoom instead. With a closer telephoto lens and artificial intelligence to "recreate" the lost pixels by a digital zoom, we obtain "zoom" x5 or x10 "lossless".
There are also many cases of scene detection, adjusting the camera settings and the software processing that follows to make the best result to the eye or all Super HDR modes that can find information in very contrasted images with a simple photo (whereas it takes normally 3 at least for "classic" HDR.
Stabilization by artificial intelligence is also impressive for some. There are indeed several ways to stabilize a camera: by stabilizing the sensor on several axes, stabilizing the lens (on several axes) or with a software function. Many manufacturers today use hybrid techniques based on material stabilization and stabilization based machine learning that allows you to take less blurred photos over a shorter or longer break. The same for the video.
Be careful, however, not to confuse this with simple algorithms
Some also promise a "super-resolution" with very detailed photos, most of which have been logically reconstructed. Be careful, however, not to confuse this with simple algorithms that close the hollows by averaging nearby pixels (a function found in all good photo editing software).
Qualcomm also made some demonstrations when presenting its Snapdragon 855 for several uses such as the ability to create a portrait mode (with a blur in the background) on the fly on video (Skype has also integrated a similar function recently ), but also the possibility of completely changing the background, always on the fly. In the same genre, we can also mention the improvements of intelligent focus in video.
All is not rosy however since the AI can also be used to attach a face of someone in a video, which made the happiness of some fans of pornography.
These last functionalities are in particular possible thanks to a facial recognition of the main subject.
Facial recognition and biometrics
Facial recognition in general is also a very popular sector in research in artificial intelligence, since, in the long term, it could go further by defining standard profiles and recognize certain elements for example to thwart attacks based on facial expressions and other muscle tics. Research still unclear today.
However, Google Photos and Apple Photos show us their strength in recognizing scenes and everyday objects. In my case, by looking for "can" in Google Photos, I thus fall on photos where you can actually see a drink in the background, it's quite impressive.
Face recognition in MIUI 7 is no longer a novelty
By going even further, artificial intelligence can offer many biometric solutions and recognize the same face with and without a beard or with and without glasses for example, always with the same precision.
Start-ups like ArcSoft, AnyVision or Morpho have made AI biometric recognition their workhorse.
But AI is not based on images alone. We can also mention some prowess in the field of sound. There are of course the compositions, like the smartphone of Huawei who finished the unfinished symphony of Schubert, but also more concrete benefits in everyday life.
The main one is speech recognition, which then opens the field to many applications. Some have turned to real-time translation, like Google with its Pixel Buds, but also other startups. Still others focus on reducing noise. Imagine phoning in a noisy bar, and your correspondent only hears your voice, all with a single microphone (while current phones use a second microphone on top to capture ambient noise and suppress it). Once again, it comes back to the replacement of material elements.
YouTube link Subscribe to FrAndroid
In a slightly different genre, we can also mention music recognition solutions approaching Shazam or simply the integrated option in Google.
Much more to come
These are just a few examples of AI that we can encounter on a daily basis, but of course AI has a multitude of other uses already today and even more to come in the next few years.
Many builders today rely on IoT for example. Imagine a camera in your refrigerator that tells you when you run out of your favorite food or drink. Imagine a router able to detect the presence of a person at home just by the propagation of its waves. The possibilities of use are almost endless. It remains only to invent them and especially to put them into practice.