Artificial Intelligence: Test Your Knowledge!

TRUE

ChatGPT is the best-known example of an application powered by generative AI. This type of AI, based on an immense amount of training data and a hyper-powerful mathematical system, can understand a complex query and statistically predict the best possible response. This is why we’re talking here about AI using a probabilistic technique. A generative AI can, for example, generate a very elaborate response to a request to produce “a text of around 1500 words on the evolution of the teacher’s role throughout history, with an emphasis on the technologies that have entered the classroom, in the style of author X and including five quotations”. And since this AI has a memory, it can add new details to its query to refine its response. It’s worth noting that it was the development of deep learning in the second half of the 2010 decade that boosted the performance of generative AI tenfold.

To find out more:

• Deciphering ChatGPT
• 10 categories of generative AI tools

The correct associations are as follows:

A) Ingredients (data): Just as you need ingredients to bake a cake, AI needs data, lots of data… like images, text, numbers, etc. This data helps AI learn patterns and make decisions.

B) Recipe (algorithm): A recipe guides you through the steps involved in preparing a cake. In the case of AI, an algorithm can be compared to a recipe – a set of instructions telling the AI what to do with the data to learn and make decisions.

C) Mixing and baking (training): This is how the cake takes shape. With AI, the algorithm processes the data, learns from it, and improves over time. This repeated learning process is called training.

D) A great cake (AI result): After about 30 minutes at 350°F… your cake is ready! With AI, after training, you get a system that can make predictions, recognize things, or solve problems based on what it has learned from the data.

E) Improvement (AI refinement): Just as you can adapt your recipe for the next cake, AI can be refined by modifying the algorithm or feeding it more data to make it perform even better.

In short, just like baking a cake in real life, results can vary according to the quality of the data and the algorithm used. However, like a baker perfecting his recipe, AI systems can be improved and refined to get closer to the desired, mouth-watering results.

Find out more: Simplifying AI…and Making Cakes

C.

In the early 2000s, researchers Geoffrey Hinton, Yann LeCun and Yoshua Bengio decided to re-examine the potential of digital artificial neural networks, a technology neglected by research from the late 1990s to the early 2010s. The trio “invented” deep learning, which is today proving to be the most promising branch of artificial intelligence, the one that has rekindled interest in this field of technology and enabled the impressive advances in AI we know today.

Inspired by the workings of the human brain, these artificial neural networks, optimized by learning algorithms (a set of rules), perform calculations, and operate according to a layered system: the results of each layer are used by successive layers, hence the term “deep”. While the first layers extract simple features, subsequent layers combine them to form concepts that gain in complexity. The principle behind this technology is to allow the computer to find the best way to solve a problem on its own, given a very large amount of data and indications of the expected outcome. Deep learning can use both supervised and unsupervised learning.

The great revolution brought about by deep learning is that the tasks required of computers are now based on essentially the same principles or algorithms. Whereas AI knowledge used to be subdivided into several types of application, each studied in silo, efforts are now concerted in an attempt to understand learning mechanisms.

Find out more:

• Deep learning 101
• Artificial intelligence: from manual programming to deep learning
• Mini glossary of artificial intelligence

A., B. and C.

Fortunately, several initiatives, including those mentioned in points A, B and C, have been launched. However, much remains to be done, as advances in this technology are more rapid than even some luminaries in the field anticipated, and the stakes involved are high.

Find out more: Initiatives for a Responsible and Human-Centered Artificial Intelligence

B.

ChatGPT is undoubtedly the best-known generative AI tool, but it’s far from being the only one… In fact, the applications powered by this form of AI and accessible to the general public are multiplying all the time. Using algorithms and big data, this type of AI can generate original content – be it text, image, audio, or video – in response to user requests. As these applications evolve very rapidly, they tend to become increasingly multimodal, in other words, to generate more than one type of content.

Regarding point D, despite its impressive “cognitive” capacities, which mimic those of human intelligence, generative AI is not endowed with sentience (the ability to reason and feel) or metacognition, which means it cannot look critically at what it produces and therefore cannot correct itself. As a result, its responses are at risk of being tainted by various biases (racist, sexist, favorable or unfavorable to certain political affiliations, etc.) or of containing erroneous results, which have been termed “fabulations” or “hallucinations”. This is because the modus operandi of this technology, which ensures that it generates the most probable answer (as do other generative AIs), leads it first and foremost to provide an answer, even if these are erroneous results or pure inventions…