|Course title||From complexity to Intelligence|
|Course address||Télécom ParisTech 46, rue Barrault 13e arr.|
|Minimum year of study||4th year|
|Minimum level of English||Good|
|Minimum level of French||None|
Complexity, simplicity, artificial intelligence, cognition
|Professor responsible||Jean-Louis Dessalles|
JL Dessalles and Pierre Alexandre Murena
|Number of places||Minimum: 12, Maximum: 30, Reserved for local students: 15|
The notion of complexity has been invented 50 years ago to solve mathematical issues related to machine learning, randomness and proof theory. It led to the development of Algorithmic Information Theory (AIT). Complexity and AIT have more recently been shown essential to address aspects of human intelligence, such as perception, relevance, decision making and emotional intensity. These aspects of cognition were sometimes considered mysterious and unpredictable. They can now be regarded as resulting in part from computations based on complexity and its converse, simplicity. For instance, abnormally simple situations such as a coincidence (two colleagues having dressed in purple independently) or a remarkable lottery draw (e.g. 1-2-3-4-5-6) are systematically perceived as unexpected and interesting. When crediting or blaming a person for an action (e.g. giving the wrong medicine to an allergic child), one considers the simplicity of the causal link leading to the consequences. One also considers the person’s ability to measure that simplicity. A dramatic event is perceived as more emotional if the victims can be defined simply (celebrities, friends’ friends), if the place is simple (famous location or close to one’s home) or if the circumstances are causally complex (e.g. the victim was unlikely to be there). The design of intelligent systems must take advantage of this sensitivity of the human mind to complexity and simplicity.
|Programme to be followed||
This course begins with an introduction to the mathematical notion of complexity (also known as Kolmogorov complexity). The notion will be shown to be useful for the study of perception (hidden shapes, shape recognition), of relevance (interestingness, unexpectedness), of decision making (subjective probability), of responsibility and of emotional intensity. All these aspects will be studied using concrete examples. We will explore various potential applications of these models to the design of artificial intelligent systems, especially in the domains of machine learning and dialogue systems.
- Ability to follow mathematical reasoning.
- Mastery of object-oriented programming. Elementary knowledge of the Python programming language is recommended.
Half of the time will be devoted to personal work in lab sessions. This personal work will be evaluated. Students will also be asked to make a small original contribution and to present it orally. They will also have to answer a short quiz on the last day.