# Quiz — Session 5 🟢 # “AI that sees, hears and creates” **Program:** Applied AI — Beginner Level · Instructor: Yann Isola **12 questions · Only one correct answer per question · Answer key commented at the end** --- ### Question 1 What is **multimodal** AI? - A) An AI that works on several phones at once - B) An AI capable of processing several types of content: text, images, sound, video - C) An AI that speaks several languages - D) An AI composed of several robots ### Question 2 For a computer, a digital image is above all… - A) A drawing that he looks at like us - B) A magic file impossible to analyze - C) A grid of pixels, that is to say a large table of numbers - D) A series of words describing the scene ### Question 3 What do the three numbers in the **RGB** code of a pixel mean? - A) The amount of red, green and blue of the pixel - B) The width, height and depth of the image - C) Speed, volume and shine - D) The GPS coordinates of the photo ### Question 4 In what order does the AI “see” an image? - A) Recognition → pixels → visual cues - B) Pixels → visual cues (contours, shapes) → object recognition - C) She guesses directly without analyzing - D) Visual cues → sounds → text ### Question 5 Which of these tools is used to **generate images** from a text description? ⚠ -A)Shazam - B) A GPS - C) DALL-E, Midjourney or Stable Diffusion - D) A spelling checker ### Question 6 **Voice recognition** is the transformation… - A) From text to speech - B) From a sound (voice) to text - C) From an image to music - D) From one language to another language ### Question 7 Which voice recognition application has greatly helped people who are deaf or hard of hearing? - A) Photo filters - B) Automatic subtitles in real time - C) Facial unlocking - D) Video games ### Question 8 What is a **deepfake**? - A) A computer virus that steals passwords - B) A photo that is simply blurry or poorly framed - C) Content (image, voice, video) created or modified by AI to make people believe something false - D) A social network for AI ### Question 9 When faced with a shocking video of a celebrity received on your phone, what is the **best reflex**? - A) Share it quickly to warn everyone - B) Check the source and see if reliable media are talking about it before believing it or sharing it - C) Believe the video if the image is of good quality - D) Delete your account immediately ### Question 10 Which of these uses of multimodal AI is an example of a **positive application** for accessibility? - A) Clone a stranger's voice to make a joke - B) Create a fake video of a politician - C) Describe out loud what the phone camera “sees” for a blind person - D) Generate homework on behalf of the students ### Question 11 What is a **world model** in artificial intelligence? - A) An AI model trained in every country in the world - B) An internal representation of how the world works, which allows us to anticipate what will happen if we act (if I push this glass, will it fall?) - C) An AI-generated world map - D) The largest AI model ever built ### Question 12 Why do we say that a baby or a cat already has some form of model of the world?- A) Because they know how to speak several languages - B) Because they have read a lot of books - C) Because they anticipate space, distances and the fall of objects **even before** knowing how to speak or name things - D) Because they calculate the gravity equations in their heads --- --- ## ✅ Corrected commented **1 — B.** “Multi” = several, “modal” = content types (modes). Multimodal AI sees, listens, reads and can create images, sound or video — not just text. **2 — C.** A digital image is a grid of **pixels**, and each pixel is described by numbers. The AI ​​does not “look”: it calculates on this table of numbers. **3 — A.** RGB = Red, Green, Blue (in English RGB). Each value ranges from 0 to 255. By mixing these three colors, you get all the others. **4 — B.** The AI ​​starts from pixels, first detects simple clues (outlines, color spots), combines them into shapes (eye, ear, etc.), then recognizes the entire object with a confidence score. Like learning to read: letters → syllables → words. **5 — C.** DALL-E, Midjourney and Stable Diffusion transform a written description (a *prompt*) into an image. ⚠ The names and capabilities of these tools evolve quickly — the principle remains the same. **6 — B.** Voice recognition = sound → text (the machine listens). The opposite, text → sound, is called **speech synthesis** (the machine speaks) — that was answer A. **7 — B.** Automatic subtitles (YouTube, videoconferences, TV) make content accessible to deaf or hard of hearing people. A real revolution made possible by voice recognition. **8 — C.** Deepfake = *deep learning* (“deep learning”) + *fake* (“false”). The danger: misinformation, cloned voice scams, harassment — and the erosion of trust in general. **9 — B.** The detective's reflex: check the **source**, cross-check with reliable media, and be wary of content that provokes strong emotion. We never share out of emotion. The quality of the image (answer C) proves nothing: deepfakes are precisely more and more perfect. ⚠ **10 — C.** AI audio description helps blind or visually impaired people read a menu, recognize an object, understand a scene. Answers A and B are harmful uses; the D is a bad academic usage (and AI should help you learn, not learn for you!). **11 — B.** A model of the world goes beyond words: it is a representation of **how the world works**, which allows us to **anticipate the consequences of an action**. This is what is missing in a model that simply manipulates text (recall the enigma of the car to be washed: the answer was zero km on foot!). **12 — C.** Our understanding of the world is **older than language**. A baby understands distances and the fall of objects before speaking; a cat aims at a table without knowing the word “trajectory”. It is precisely this “physical” intelligence that researchers are trying to give to AI (image → video → action). --- **Score:** - 9-10 ✅: Certified multimodal detective! - 7-8 👍: Very solid — just reread the answer key for the missed questions. - 5-6 📖: The basics are there, review “how AI sees” and deepfakes. - < 5 🔄: Take another look at the session's interactive web page, then try the quiz again.