(Files) This illustration picture shows the AI (Artificial Intelligence) smartphone app ChatGPT surrounded by other AI App in Vaasa, on June 6, 2023. (Photo by Olivier Morin / AFP)
Paris: With global leaders set to attend a summit on artificial intelligence (AI) in Paris on February 10-11, here are some of the key concepts in the field:
AI
Asked what artificial intelligence is, the AI-powered ChatGPT system responds that the term "refers to the simulation of human intelligence in machines that are programmed to think, learn and make decisions".
AI's fundamental characteristic is ingesting vast quantities of data that are then processed using methods from statistical mechanics.
Its capabilities can range across fields including computing, maths, linguistics, psychology, neuroscience or philosophy.
At present, the technology is used in applications ranging from investigating tumours to facial recognition, chatbots, translation of human languages, forecasting breakdowns in machine tools and self-driving cars.
Algorithm
At the heart of all computer operations, an algorithm is a series of steps or instructions followed by a computer programme to achieve a given result.
Algorithms can specify rules for an AI's behaviour, helping it to achieve its developers' objectives.
Unlike a simple computer programme, AI algorithms allow the system to learn for itself.
Machine Learning
Machine learning is one approach that researchers have used in their quest to produce artificial intelligence.
It allows computers to learn from data without being explicitly programmed on how to respond.
In recent years, the field of neural networks -- inspired by the functioning of the human brain -- has proven particularly fruitful.
In a neural network, connections between some nodes are strengthened and others weakened as the system learns.
Learning can be "supervised", in which the system learns to classify new data based on a model -- for instance, to identify spam in an email or other messaging applications.
"Unsupervised" learning allows the system to independently discover patterns or categories in the available data that might not have been immediately apparent.
An example application would be allowing an online store to identify buying trends in sales data.
"Reinforcement" learning adds a process of repeated trial-and-error in which the system is penalised or rewarded based on its outcomes, allowing it to learn and improve.
One example might be a self-driving vehicle whose objective is to reach its destination as fast as possible -- but in safety.
That would lead it to learn to stop at red lights, even if that costs it time.
Deep learning
A sub-field of AI, deep learning owes its name to its use of multiple layers of neural networks.
Raw data is analysed by each layer in turn at growing levels of abstraction.
Deep learning was conceived by 2024 physics Nobel winner Geoffrey Hinton, who was awarded the prize alongside 1980s neural-network pioneer John Hopfield.
"The more layers you have, the more complex behaviour can become, and the more complex the behaviour can be, the easier it is to learn a desired behaviour efficiently," said Francis Bach, head of France's SIERRA statistical learning laboratory.
Discoveries about deep learning have since the 2010s allowed for a leap forward in computers' processing power and an abundance of data to train AI models.
The technique may help unlock major new scientific advances.
Also in 2024, the chemistry Nobel was awarded to researchers using deep learning to create and predict protein structures.
Language models
Large language models are the most visible example of so-called generative AI, powering tools like OpenAI's ChatGPT or Google's Gemini.
Such systems are capable of writing a dissertation, answering legal questions or producing a cake recipe based on their statistical models
But the technology is far from infallible, suffering from "hallucinations".
Chatbots or conversational assistants can also be found in many everyday applications, such as answering queries from visitors to a company's website.
On major streaming platforms, recommendation engines will suggest content like films or music to users based on how their tastes overlap with other subscribers.
Elsewhere in daily life, AI helps power navigation software or tools to help with spelling, grammar and style in text.
Artificial General Intelligence
Artificial general intelligence (AGI) is the holy grail of the whole AI field, denoting an as-yet unrealised dream of a machine capable of reproducing all human cognitive abilities.
Its promoters, including OpenAI chief Sam Altman and his rivals at Anthropic, see such a system as within reach -- using oceans of data and gargantuan processing power to train ever more powerful LLMs.
Sceptics insist that LLM technology has important limits, including on its ability to reason.
"LLMs do not work like human beings", as flesh-and-blood intelligences are "sense-making machines" with fundamentally different abilities to today's computer systems, Maxime Amblard, computing professor at France's University of Lorraine, told AFP last year.
