Artificial intelligence, machine learning, deep learning: minimum glossary to understand AI

Only 5 days were enough to reach the first million users, about 2 months were needed to reach 100: the incredible numbers of ChatGPTthe OpenAI chatbot which, according to Business Insidermay be the “fastest growing consumer app of all time”.

There spread so fast and massive than ChatGPT, among other things, has monopolized the debate within the tech world, and beyond. Just take a look at Google’s trends, which show that never in the last 5 years have people searched for so much information on AI. Numbers also confirmed on TikTokwhere the dedicated hashtag reached 78 million views.

Indeed, and precisely from ChatGPTartificial intelligence seems to have gone beyond the confines of the world of technology to definitively enter the public debate, between the impact on school and fears about the automation of workplaces.

To talk about it effectively, however, it is necessary to know the fundamental concepts. For this reason, we got help from DataPizzaa community of enthusiasts and professionals with over 100,000 followers on social networks, born in March 2021 precisely to dedicate itself to the dissemination of AI and data science topics.

Together with the two founders, the 24-year-olds Alexander Risaro And Pierpaolo D’Odoricowe have tried together to draw a minimum glossary of this new wave of technology.

Artificial intelligence

Let’s start from the basics and then from an expression we hear about every day: “The term refers to that branch of science which studies how machines can mimic human intelligence to perform certain tasks – Risaro and D’Odorico explained to us – It is an umbrella formula, which contains the concepts of machine learning and deep learning. The former is a subset of AI, and in turn deep learning is a subset of machine learning”.

In other words, using the 3 terms as synonyms is not entirely correct: “All the technologies that we often hear mentioned, such as ChatGPT, are part of the subset of machine learning – is the clarification of the founders of DataPizza – However, there are other types of AI, such as the symbolic one, which differ from this paradigm”.

Machine learning

At the basis of the new wave of AI there is therefore machine learning: “It is about that set of models and algorithms that use human experience in the form of data, and the mental energy harnessed in the learning process in the form of computational power”.

It is those systems capable of learn to find a solution independentlystarting from a large amount of data: “Machine learning exploits a very different paradigm compared to that of traditional programming, it starts from many inputs associated with their respective outputs (example the image of a cat associated with a label that says that there is a cat in that image, ed) and learns from the data the algorithm that maps the relationship between input and output”.

Deep learning

Among the machine learning techniques, deep learning is the one used for all the best-known text or image generation algorithms: “It is a search field which exploits models called neural networks, which try to simulate the behavior of the human brain – Risaro and D’Odorico confirmed to us – It is precisely starting from these neural networks that fields of artificial intelligence are developed which try to simulate human behavior, as the computer visionwhich aims to replicate the human visual system (a model that is able to distinguish whether a cat or a dog is represented in an image, ed)”.

Generative artificial intelligence

Artificial intelligence is not really something new: each of us has been using it every day for years, starting from recommendation systems of web platforms. The new wave, however, concerns the so-called generative AI, which “focuses on the creation of artificial content, such as images, sound, video, text or even code”. Everything always starts from the data: “These models are trained on large amounts of input and use this information to generate new data that they look realistic and consistent. Some examples of generative AI are GPT-3 for text creation and Stable Diffusion or Dall-E 2 for image generation”.

Large Language Model

Within the world of Generative AI, the last couple of months have been monopolized by ChatGPT, which is an LLM, or a system “that is able to understand language in a mathematical sense, that is, it understands the probability distribution of a sequence of words: given a sequence of words, an LLM is able to predict which word will be most likely to come next. There type of word that will be predicted will depend entirely on the type of data that was used to train this language model.”

Social

How Artifact works, the TikTok for news from the creators of Instagram

by Francesco Marino

04 February 2023

Data sciences

All AI systems are based on data: “Data science is a transversal field to those mentioned above, which is based on exploiting these techniques to extract meaning and information from data – is the conclusion of the DataPizza experts – Artificial intelligence is the tip of the iceberg of a data science project: first of all you need to acquire data, clean them up, aggregate them, understand them and only then can the latest machine learning or deep learning models be implemented.”