The evolution and future of Artificial Intelligence: How do we get to today's technology?

The Evolution of Artificial Intelligence: From the Universal Turing Machine to Deep Learning Networks.

In the 1950s, important advances emerged in the AI. Psychologist Frank Rosenblatt developed Perceptrón, an artificial neuronal network capable of recognizing patterns in images and texts. In addition, computer scientist John McCarthy created the LISP programming language, one of the first languages specifically designed for artificial intelligence. In this decade, the game of the 3 online became one of the first games in which the AI was applied, although the game systems were still simple and limited.

In 1966, Joseph Weizenbaum created ELIZA, a program that simulated a therapist and could keep simple conversations with users. ELIZA is considered one of the first chatbots in history and demonstrated the potential of AI to interact with humans.

In the 1970s, natural language processing systems (NLP) became an important research field in the AI, with the aim of allowing computers to understand human language. Despite these advances, AI went through a period of disinterest and lack of investment known as the "IA winter".

In the 1980s, deep learning algorithms have been developed that allow training neural networks with multiple layers to recognize complex patterns in images, sounds and text. These algorithms gave way to deep neural networks, which are fundamental in many modern AI applications.

With the development of modern computing and the creation of neural networks, AI began to win in more complex games such as chess and Go. In 1997, IBM's Deep Blue chess program won the world chess champion Garry Kasparov. Later in 2016, Google's AlphaGo artificial intelligence program won the world champion Go Lee Sedol.

In 2011, IBM introduced Watson, an AI system capable of answering questions in natural language and winning humans in the television contest Jeopardy!. Watson demonstrated that AI could manage large amounts of unstructured data and answer questions accurately.

DeepMind, an artificial intelligence company owned by Google, was one of the leaders in applying AI to games. In 2013, DeepMind developed a system that learned to play Atari 2600 console games without prior knowledge of the game. This system used a technique called "reinforcement learning" to learn from your experience and improve your performance.

Later in 2015, DeepMind developed AlphaGo, a program that uses deep neuronal networks and reinforcement learning techniques to play the Go game. AlphaGo won the European Go champion in a series of games, and later defeated world champion Lee Sedol in a historic five-game match.

In the 2010s it is said that it began "The fourth industrial revolution", that of artificial intelligence (IA) that refers to the advancement in the development of artificial intelligence systems based on deep learning or deep learning. Deep learning is an automatic learning technique that uses artificial neural networks to mimic how the human brain processes information. These networks can analyze large amounts of data to detect patterns and make predictions, which led to major advances in areas such as voice and image recognition, automatic translation, natural language processing, among others. The beginning of the fourth AI revolution is due to several factors, including increased data processing capacity, the development of more sophisticated algorithms and the growing interest and funding in the AI field. These advances allowed the creation of more powerful AI systems capable of addressing increasingly complex problems.

To understand more about how "aprenden deeply" we will see the different techniques and their evolution, which are called "paradigms of learning".

In artificial intelligence, there are three main learning paradigms:

1. Supervised learning: In this approach, the algorithm receives a set of data labeled with the correct answer and learns to predict the correct response to new data. For example, supervised learning can be used to teach an algorithm to recognize images of dogs and cats. The algorithm receives a set of images labeled as dogs or cats and learns to identify the difference between both.
2. Unsupervised learning: In this approach, the algorithm receives a dataset without labeling and must find patterns or structures in the data. This approach is commonly used in data mining and the detection of anomalies. For example, learning cannot be used to identify groups of similar clients in a sales database.
3. Reinforcement Learning: In this approach, the algorithm learns through interaction with an environment. The algorithm receives a reward or punishment for every action it takes in the environment and learns to maximize reward over time. This approach is commonly used in robotics and games. For example, learning can be used to teach a robot to walk on an uneven surface.

• How can you talk to smart bots like chatGPT? With NLP: Natural Language Processing (Natural Language Processing) is a branch of artificial intelligence that focuses on understanding and processing human language. It is useful for various tasks such as automatic language translation, text generation, feelings analysis, information extraction and automatic answer to questions.
• How do you get realistic photos? With GAN: Generative Adversaria Network is a type of deep learning model used to generate synthetic data that resembles the original training data. GAN consists of two neural networks: a generator and a discriminator. The generator takes a random sample and turns it into an image that looks like the actual images of the training data set. The discriminator, on the other hand, receives both actual images and images generated by the generator, and its function is to distinguish between the two. As the discriminator evaluates the images generated by the generator, it also sends signals to the generator to improve its ability to generate more realistic images. GAN enters iteratively, with the generator and the discriminator competing with each other in a zero-sum game, in which the goal is to maximize the capacity of the generator to deceive the discriminator. As the generator learns to produce more realistic images, the discriminator becomes more sophisticated and is able to better distinguish between the actual images and those generated by the generator. GAN is a very promising technique in generating synthetic data for use in various applications, such as image synthesis, music creation and product design.
• How do you get text images? With GAUGAN (Generative Adversarial Networks for Image Synthesis and Editing) is a generational neural network model that is used to generate synthetic images from an outline or a textual description. It is useful for creating landscapes and scenarios, and has proven to be a useful tool for artists, designers and video game developers.
• How do you get 3D environments from images or videos? With NERF (NeRF: Neural Radiance Fields) is a deep neural network model used in photorealist 3D imaging. Use a technique called "neural radiation fields" (Neural Radiance Fields) to generate 3D images from 2D image data. It is useful for creating 3D environments from images or videos.
• How to get text images using models like that of our faces? CNN means "Convolutional Neural Networks" and are a type of neuronal network that is used in image and video processing. It consists of convolution layers, pooling layers and completely connected layers. Convolution layers extract input image characteristics, pooling layers reduce the output dimension of the convolution layer and completely connected layers are used for classification. CNNs are commonly used in image processing and videos, and are very useful for getting images from text.