What is artificial intelligence?

ChatGPT, Midjourney… AIs are in the news more than ever. So, what is the definition of artificial intelligence? And related concepts like deep learning or singularity?

Artificial intelligence is already ubiquitous and highly fantasized at the same time. The word “intelligence” in the term tends to blur the lines about the real nature of these tools in their current form. What exactly is artificial intelligence? What are its forms?

What is artificial intelligence (AI)?

There is no generally accepted definition of artificial intelligence. Larousse’s encyclopedia suffices to note. a set of theories and techniques used to produce machines that can simulate human intelligence “. This concept of simulation – or sometimes imitation – of the brain is central.

More specifically, AI is computer software coded to automatically perform tasks ranging from data collection and processing. This program should lead to the performance of a certain task: change the image, recognize the face, get the result of the equation. It becomes learning, training.

Robotic AIs like those depicted in fiction Westworld there is absolutely no nowadays. // Source: HBO

What are the different types of artificial intelligence?

The term artificial intelligence actually combines several types of AI.

Poor AI

“Weak” AI is the AI ​​that exists today. It is an algorithm for automating tasks – faster and more efficiently than the human brain. This type of artificial intelligence has no feelings, no conscience, and no thinking ability. All existing AI is compatible with this.

In this sense of artificial intelligence, we must take the word “intelligence” in its Anglo-Saxon sense, that is, the collection and processing of information. They are algorithms and not intellectual exercises. It is an imperfect imitation of human intelligence.

Powerful artificial intelligence

“Strong” artificial intelligence does not exist today. Here again, it’s about automating tasks, but also about learning from these tasks to break free from pre-integrated information. Powerful artificial intelligence is designed to completely copy human intelligence – create, ask questions, understand itself, even form opinions.

Powerful AI is science fiction today. This is consistent with most of what we see in movies and TV shows — Westworldmarvel movies, Ex Machina, His, bicentennial man, I-robot, Altered Carbonand so on.

What is the difference between robot and artificial intelligence?

Robot and artificial intelligence are two different things, but they can be related. Metaphorically, we can think that if AI is the brain, then the robot is the body. Robotics refers to mechanical and/or biotechnological systems for imitating the movements of living organisms – the human body and animals. And artificial intelligence corresponds to the imitation of brain function.

Of course, there are plans to combine the two. Ameca, a robot with impressive facial expressions, added GPT-3 to the AI ​​to attempt dialogue sequences with it.

What is deep learning and artificial neural networks?

Artificial intelligence is based on learning: before answering queries correctly, an algorithm is trained using large data sets. It is a process of constant mistakes and corrections. This system is called machine learning. But within this system there is a more advanced model of artificial intelligence based on a network of artificial neurons. This is deep learning, or deep learning, through which AI is taught to perform a task better.

The term “deep” in “deep learning” comes from the large number of layers in an artificial neural network. Data converted to data within one layer becomes input data for the next layer. It is this type of artificial intelligence that has the ability to predict.

What is artificial intelligence used for today?

Artificial intelligence is used in several fields today. The level of complexity of AI varies greatly depending on its use. Some daily examples:

  • Image processing and recognition: face recognition, image creation programs (Midjourney, Dall-E, etc.), photo/video software,
  • Voice processing and recognition: voice recognition, voice assistants,
  • Text processing and recognition: translations (Google Translate, DeepL), anti-spam filters,
  • Recommendation algorithms on social networks and e-commerce sites,
  • Scientific research: speeding up calculations and increasing the volume of treatable data,
  • Finance (fraud prediction),
  • Autonomous cars (combination of image recognition and object prediction for braking),
  • In video games.

What is ChatGPT and GPT-3?

The OpenAI company has entered the headlines with the release of ChatGPT. This AI is a service chatbot: it provides information based on the questions you ask (like a simple recipe). Artificial intelligence remembers what the user tells it.

ChatGPT is available online — here — and works in multiple languages. ChatGPT is not connected to the Internet and is therefore only based on data entered during training with data valid until 2021.

All chats conducted with ChatGPT can be viewed by OpenAI. They are precisely used to train the AI ​​on a massive scale to the general public, because the more inputs, the more it adjusts, the more it learns (however, it is possible to delete your data).

When asked if ChatGPT could pass the Turing test.  // Source: Screenshot of ChatGPT's Numerama
When asked if ChatGPT could pass the Turing test. // Source: Screenshot of ChatGPT’s Numerama

This chatbot is actually a derivative of GPT-3.5, an artificial intelligence designed to generate the most natural text possible. ChatGPT is enhanced with a technique called Reinforcement Learning with Human Interaction: it involves showing the AI ​​what a human conversation looks like and countering it by providing negative or positive feedback from the human depending on the quality of the response.

The program works with GPT-3 to create images from DALL·E text commands.

OpenAI is developing GPT-4, which can reach billions upon billions of parameters, much more than GPT-3. It’s exponential: GPT-2 (2019) combined millions of parameters; GPT-3 (2020) combined tens of billions of parameters; GPT-3.5 (2022) combined hundreds of billions of parameters.

Midjourney, Stable Diffusion, How Does Dall-E Work?

2022 saw an explosion of image-generating AIs: Midjourney, Dall-E 2, Stable Diffusion. These algorithms can generate a new image from a text query.

Take the example of Dall-E 2 developed by OpenAI (like ChatGPT). It all starts with a in a hurry : text describing the desired image. It goes through the CLIP model, which predicts the title of the image, as originally designed by OpenAI. Because CLIP has done this millions of times, it can now predict the opposite thanks to this training. This process transforms concepts into a numerical representation (a sequence of numbers) and a spatial representation (the relationship between them, their location in space). Then, a final decoding step, called a diffusion model, converts the predicted numerical result into a coherent image.

ChatGPT Midjourney
An example image created with Midjourney. // Source: Midjourney

Does this software violate copyright by “scraping” data from thousands of artists’ works? This question has yet to be legally resolved, but the artists filed a complaint against Midjourney and Stable Diffusion in early 2023. Similarly, GettyImages filed a complaint against Stable Diffusion. Many artists also protest the threat these algorithms pose to their precarious profession.

What is the Turing test?

The Turing test is intended to test artificial intelligence. This is a configuration in which a person talks to a blind interlocutor without knowing whether he is in front of another person or a machine. If you can’t tell the difference, then the AI ​​has failed. This implies that the machine is meant to semantically imitate the human, but at the same time it should never evoke its nature. The reason why ChatGPT, which pretends to be artificial intelligence, cannot pass this test. The principle was described by Alan Turing back in the 1950s.

The Imitation Game is about deciphering the Enigma machine.  // Source: Black Bear Images
The Imitation Game It tells the story of Alan Turing. // Source: Black Bear Images

What is the AI ​​Technological Singularity?

In the field of foresight, a term has been around a lot since the late 20th century: the technological singularity. This is the assumption that artificial intelligences will exponentially reach a tipping point where their performance will be overpowered. The singularity is the moment when AIs will escape our control, being far more powerful than humans in terms of artificial intelligence.

This theory particularly feeds the philosophical and political movement of transhumanism. Some of the transhumanists believe in this Singularity as a horizon to consider – so they believe that the only solution is the fusion of man and machine so that man remains competitive.

Do you like our media? Tell us in this poll!

Leave a Reply

Your email address will not be published. Required fields are marked *