AI Quiz

Answer: Alan Turing. Alan Turing laid the theoretical foundations of modern computing and proposed the 'Turing Test' in 1950 to evaluate a machine's intelligence.

Answer: OpenAI. ChatGPT, launched by OpenAI in November 2022, quickly attracted tens of millions of users and popularized large language models with the general public.

Answer: A large language model (LLM). A large language model (LLM) is trained on huge text corpora to understand and generate coherent natural language.

Answer: A hypothetical AI able to perform any intellectual task as well as a human. Today's AIs are often 'narrow': excellent in one field, lost elsewhere. AGI remains a debated goal: some think it is near, others doubt we can even define it precisely.

Answer: The Roomba robot vacuum. It maps the room and goes around obstacles… with varying success. It also became an internet star, with videos of cats riding on it.

Answer: The imitation game. A judge converses in writing with a human and a machine without seeing them: if he cannot tell them apart, the machine has 'succeeded'. The film 'The Imitation Game' (2014) uses the title.

Answer: Copilot. The name suggests a 'co-pilot' who helps without taking the controls. Cortana was an older Microsoft voice assistant, Bard the first name of Google's AI, Alexa Amazon's.

Answer: Aibo. It learned tricks and recognised its owner. In Japan, temples have held funeral ceremonies for broken Aibos. Pepper and Nao are humanoids; Asimo is by Honda.

Answer: Hiding an invisible mark that shows it came from an AI. These signatures, invisible to the eye, help fight deepfakes and disinformation. They are still hard to make tamper-proof.

Answer: 1956. The term 'artificial intelligence' was proposed by John McCarthy at the 1956 Dartmouth Conference, considered the official founding moment of the field.

Answer: AI alignment. AI 'alignment' refers to the set of methods aimed at ensuring an AI system acts in accordance with human values and intentions.

Answer: What is easy for a human (walking, folding laundry) is very hard for a machine, and vice versa. Beating a chess champion became relatively 'easy' for a computer, while a robot that can neatly fold a T-shirt is still a challenge. What evolution perfected over millions of years seems simple to us… precisely because we no longer think about it.

Answer: The paperclip maximizer. The point is not that an AI hates humans, but that a badly defined goal, pursued with extreme efficiency, can lead to disaster. An (almost) comical way to illustrate the alignment problem.

Answer: Arthur Samuel. Arthur Samuel's program played thousands of games against itself to improve. Being beaten by your own creation: probably the finest compliment for a researcher.

Answer: Clever Hans. He could not count: he was reacting to his master's tiny involuntary cues. Likewise, an AI may seem to succeed at a task… by exploiting a hidden clue in the data rather than understanding.

Answer: Retrieval-augmented generation: the model fetches relevant documents before answering. By relying on specific sources (a company's document base, for example), RAG reduces errors and "hallucinations".

Answer: Stanley. This robotised Volkswagen Touareg finished the course in under 7 hours. In 2004, no vehicle had gone further than 12 km!