How do we get to Generative AI?

 How do we get to Generative AI?

Mg. en Ing. Carlos Gerardo Said 



Artificial intelligence was not born with ChatGPT. Nor was it born with deep learning algorithms (e.g. recognizing objects in an image).

The history of artificial intelligence begins many decades ago.

These are the most important milestones in the evolution of artificial intelligence.

1943: the first artificial neural network

It is shown how a system of artificial neurons can learn in the same way as humans (https://link.springer.com/article/10.1007/BF02478259).

It is shown that through a series of trials and errors, the connections between the artificial neurons that led to the correct action are strengthened, while those that led to the incorrect action are weakened. This system, therefore, does not need to be programmed with fixed rules, but learns autonomously by exploiting the data it has.

1950: The Turing Test

Alan Turing first posed (in his article “Computing Machinery and Intelligence”) the question that we still try to answer today: can machines think? To answer it, Turing proposes a method to find out if a machine is really intelligent.

1956: The term “artificial intelligence” emerges

How any aspect of learning or any other characteristic of intelligence can, in principle, be described with such precision that a machine capable of simulating it can be built.

McCarthy coined the term “artificial intelligence”.

1958: The Mark I Perceptron

In 1958, the Mark I Perceptron was presented.

A two-layer neural network capable of imitating the behavior of only eight neurons, the Perceptron is only capable (after many trials and errors) of recognizing right from left.

1966: Eliza, the first chatbot in history

It is the first chatbot in history.

Eliza marks a before and after: it is the first software capable of interacting with human beings.

1997: DeepBlue defeats Gary Kasparov

IBM's DeepBlue system defeats world chess champion Gary Kasparov (who will win the following challenges).

DeepBlue is not a neural network, but a symbolic artificial intelligence within which all the rules of chess and the moves to be executed in each situation have been encoded.

2012: AlexNet wins image recognition contest

AlexNet wins the ImageNet Challenge, a competition between image recognition software. This is perhaps the first time that deep learning has proven to be superior to other computer systems.

2016: AlphaGo beats the world champion of Go

This is an extremely complex and abstract game like Go. Google's system known as AlphaGo surprised the world and defeated one of the world's best-known Go players. It shows the capabilities of deep learning.

2020: OpenAI presents GPT-3

Another breakthrough: OpenAI presents GPT-3, its "natural language processing" system trained through 800 gigabytes of data and equipped with 200 billion parameters (the digital equivalent of our synapses).

GPT-3 demonstrates a previously unknown ability to write texts, to the point of producing an article for The Guardian. This is the beginning of generative artificial intelligence (also capable of creating images, videos and music). Two years later, we have a chatbot based on GPT-3: ChatGPT.

Comments

Popular posts from this blog

IBM goes big to keep quantum computers cold

IBM - avances en mantener adecuadamente fríos computadores cuánticos.

Estructura multinivel de los sistemas de agentes de IA