Artificial intelligence is not so artificial any more. Not that these machines have been able to reproduce the complexity of the human brain, but their abstract concept now fits within a cardboard box. Google and Target are proposing a small device that can be self-assembled, which is capable of replying to questions and can provide facial recognition. The “visual” and “vocal” kits cost 90 and 50 dollars respectively. And they look to be about as easy to be put together as a wardrobe. But to explore other facets of the AI, we are going to have to dissect its reactor.  

Needs

These AI are able to mime the patterns of the human brain in more than one domain. Some AI are able to play chess, while others can drive cars. For this, they employ algorithms supported by coding. Their DNA must then be arranged depending on the function that is needed from them. In an article published by Observer, programmer Arend Hintze defines four types of AI: those that make decisions in regards to an immediate situation; those that are inspired by the past; those that understand the motivation of others; and those that are self-conscious. The last two do not yet exist.

Material

The AI is not a pure brain. First and foremost, it is necessarily endowed with a corporeal envelop which is apt to respond to an external stimulus. This can take the form of a car whose windshield wipers start due to a contact with rainfall. In turn, the AI possesses a mode of communication which is intelligible. In our example, this would be car’s dashboard. Finally and above all, at the heart of the AI, an algorithm must try and understand the meaning of this communication. In this way the machine will be able to learn by itself. Ultimately, perhaps the AI will be conscious of itself, suggest Wired.

Technique

Machine learning, which helps an AI progress, is generally handled by an artificial neuronal network whose base algorithm is called backpropagation. With the help of a programming language like Python, we can use this to put together an AI. Libraries such a TensorFlow and PyTorch help us avoid the need to write everything, but you must still be able to code. The DataRobot organisation has automatised this procedure to allow non-initiates to conceive complicated models, explains MIT’s site. On its end Google offers a product baptised Cloud Auto ML. In order to easily create a bot, we can also use Botsify. “No need to be an expert in computer science”, notes SitePoint while listing other tools: Wit.ai, Api.ai, Melissa and Clarifai.

In class

Whatever the project, a range of courses to teach the conception of artificial intelligence are available on the Udemy and Udacity websites. But Google does even better. The American multinational proposes free initiations to machine learning, which have themselves been translated using machine learning. It is up to each individual to make of it what they understand. This technique has already sensibly improved cancer predictions, says Google. And it also served to determine the political affiliations of British writers in the 17th century based on their metaphors.