Neural networks are artificial intelligence models that attempt to emulate the way the human brain works. They are made up of many small elements called “neurons” that work together to solve problems. Each neuron receives input from other neurons or external data, and based on this input, it can activate or deactivate the transmission of signals to adjacent neurons. The number of neurons in the human brain is estimated to be around 100 billion. However, the number of neurons in an artificial neural network depends on the model and the size of the problem being addressed. The most advanced artificial neural networks developed to date have had millions or even billions of neurons. For example, the GPT-3 artificial intelligence model developed by OpenAI has 175 billion parameters, which can be seen as equivalent to neurons. However, it is important to note that the number of neurons in an artificial neural network is not the only factor determining its ability to solve complex problems. The structure of the network, the type of input data, and the learning method used are all important factors that influence the performance of a neural network.
The first neural network was developed in 1966 by Frank Rosenblatt, a US psychologist and computer scientist. His system, called the “Perceptron,” was able to learn to recognize patterns in manually supplied input data. However, the Perceptron had some limitations and was not able to solve complex problems. The amount of computational power required by a modern neural network can vary greatly depending on its size and the complexity of the problem it is trying to solve. In general, larger neural networks and those that are trying to solve more complex problems will require more computational power. For example, some of the largest and most advanced neural networks developed to date, such as Google’s AlphaGo, have required the use of specialized hardware and have consumed tens of thousands of CPU hours to train. However, smaller and less complex neural networks can be trained on a standard personal computer in a relatively short amount of time. The methods of learning used in neural networks are mainly two: supervised learning and unsupervised learning. In supervised learning, the model receives desired inputs and outputs and tries to learn to produce the correct outputs for the given inputs. In unsupervised learning, the model only receives inputs and tries to detect patterns and structures within the data without the guidance of predefined outputs.
Artificial intelligence has surpassed human abilities in many fields, both in activities that require fast and accurate calculations and in activities that require the ability to recognize patterns and make predictions. Here are some examples of fields where artificial intelligence has surpassed human abilities:
- Chess: In 1997, IBM’s Deep Blue chess program defeated world champion Garry Kasparov in a chess match.
- Go: In 2016, Google DeepMind’s AlphaGo program defeated world champion Lee Sedol in a game of Go.
- Machine translation: Artificial intelligence is now able to perform very accurate translations from one language to another, surpassing the abilities of human translators in many cases.
- Word recognition: Artificial intelligence is able to recognize words in text with greater accuracy than human optical character recognition (OCR) systems.
- Image recognition: Artificial intelligence is able to recognize objects, animals, and people in images with greater accuracy than humans.
- Crossword games: In 2010, IBM’s Watson crossword program defeated two crossword game champions.
- Financial market prediction: Some artificial intelligences are able to make predictions about financial markets with greater accuracy than human financial analysts.
Here is a brief timeline of the history of artificial intelligence:
- 1943: Warren McCulloch and Walter Pitts publish a paper on “A Logical Calculus of the Ideas Immanent in Nervous Activity,” in which they describe the concept of a neural network.
- 1950: Alan Turing publishes his famous article “Computing Machinery and Intelligence,” in which he proposes the “Turing Test” as a method for determining whether a machine is capable of thinking like a human.
- 1952: The first programmable digital computer, the UNIVAC, is developed.
- 1956: The first conference on artificial intelligence is held at Dartmouth College, where the term “artificial intelligence” is coined.
- 1957: The first artificial intelligence program, the “Logic Theorist,” is developed and is able to solve logic problems.
- 1959: Arthur Samuel defines machine learning as “the ability to learn without being explicitly programmed.”
- 1960: Joseph Weizenbaum develops ELIZA, the first natural language processing computer program.
- 1964: Danny Bobrow develops the first artificial intelligence program capable of understanding natural language queries, called the “QA system.”
- 1966: The first artificial intelligence system based on neural networks, the “Perceptron,” is developed.
- 1969: The ARPANET, the precursor to the modern internet, is developed.
- 1971: The first expert system, called “Mycin,” is developed and is able to diagnose blood infections and recommend treatment.
- 1980: The first artificial intelligence system capable of defeating a world champion in an abstract game, the “Belle chess program,” is developed.
- 1987: The first robot to be granted a patent, the “Unimate,” is developed.
- 1997: The first artificial intelligence system capable of defeating a world champion in a complex game, the “Deep Blue Go program,” is developed.
- 2005: The first autonomous vehicle, the “Stanley robot car,” wins the DARPA Grand Challenge.
- 2011: IBM’s Watson defeats two Jeopardy champions.
- 2014: The first artificial intelligence program to win at the game of Go, the “AlphaGo program,” is developed.
- 2016: The first artificial intelligence system capable of defeating a world champion in a game of draughts, the “AlphaGo draughts program,” is developed.
- 2018: The first artificial intelligence system to be awarded the “Lovelace Medal” for outstanding achievements in computing, the “DeepMind AlphaZero program,” is developed.
All images and all text in this blog were created by artificial intelligences