What is Artificial Intelligence?
Artificial Intelligence (AI) is the science concerned with the creation of intelligence within computer systems, comparable to human or animal intelligence. We speak of strong AI when the objective is to develop systems capable of producing intelligent, autonomous, and adaptive behaviour, while also representing themselves within their environment and being self-aware. AI therefore addresses issues of reasoning, dialogue and perception.
In the field of engineering, we refer to weak AI: these are the techniques that make it possible to process complex problems that cannot be solved by humans (data mining and knowledge extraction from large, heterogeneous databases, fault or anomaly detection, real-time mission re-planning in the event of incidents or disruptions…), to delegate dangerous or impossible tasks to humans (combat drones, bomb disposal robots, planetary exploration rovers…), and to facilitate dialogue between humans and machines, robots or computers (natural language processing, speech or image processing, pattern recognition…).
Download the brochure “AI simmered by our mathematics experts”
What are the different families of AI?
Operational Research
Operational Research covers mathematical methods and techniques for finding solutions (Research) to real-world (operational) problems. These resolution methods (algorithms) enable the construction of solutions to well-formulated problems (models).
The main methods are as follows:
- Optimization: gradient, BFGS, Newton, least squares…
- Mathematical Programming & Linear Programming
- Constraint Programming
- Graph Algorithms
- Heuristics & Meta-heuristics
- Markov Processes
- Multi-criteria Analysis
Areas of application include:
- Planning, scheduling and resource allocation:
- Production planning
- Human resources planning
- Supply chain optimisation
- System sizing: the process of determining the optimal capacity of systems to meet operational requirements.
Symbolic AI
Symbolic AI is based on manipulating symbols and rules to model and solve complex problems. This approach includes:
The main methods are as follows:
- Prolog (Logic Programming): declarative logic programming language (fact base and rules, queried via requests)
- Lisp (LISt Processing): a language based on the evaluation of expressions
- BRMS (Business Rules Management Systems): a business rule management system that allows business rules to be defined, deployed, executed, monitored and maintained separately from the application code
- Ontologies: formal structures that define a set of concepts and the relationships between them within a specific domain
The main applications are as follows:
- Knowledge management: structuring and using knowledge to improve decision-making processes.
- Formalisation and automation of expert reasoning
- Automation of business procedures
- Verification of business regulations
Connectionist AI / Statistical AI
These are analytical techniques. Connectionist AI is inspired by the functioning of biological neural networks. It focuses primarily on developing artificial neural network models to perform machine learning tasks.
The main methods are as follows:
- Supervised learning (on labelled data)
- Statistics: linear regression, multiple linear, polynomial, logistic
- Machine Learning: neural networks, SVM, CART, random forest
- Generative AI: automatic translation, image captioning…
- Unsupervised learning (on a dataset with no predefined labels or outputs)
- Statistics: principal component analysis, correspondence factor analysis
- Machine Learning: Kohonen maps, dynamic clustering, k-means
- Generative AI: language models, Generative Adversarial Networks
- Reinforcement learning
The main applications:
- Identifying explanatory factors and correlations: identifying causal relationships and correlations between different variables
- Forecasting from historical data: using past data to predict future trends
- Diagnosis: anomaly detection and identification of underlying causes
- Clustering / Segmentation: grouping data into homogeneous segments for more refined analyses
Collective AI / Distributed AI
Collective AI focuses on interactions between multiple agents to simulate complex systems and coordinate decision-making processes.
The main methods are as follows:
- Multi-agent systems
- Population-based meta-heuristics (biomimicry)
- Genetic algorithms
- Particle swarms
- Ant colonies
The main applications:
- Agent interactions: modelling interactions between multiple agents to simulate collective behaviours
- Complex system simulation: using simulations to understand and predict the behaviour of complex systems
- Supervision, coordination and synchronisation of decision-making processes: to ensure effective coordination between the different components of the system
Conclusion
AI in the broad sense offers powerful tools to improve business efficiency. By combining optimisation with symbolic, statistical, connectionist, and collective approaches, organisations can not only optimise their operations but also anticipate and adapt to future developments. These technologies make it possible to create innovative solutions, thus ensuring lasting competitiveness and long-term success.
