If you ask 10 experts for a definition of AI, you will probably get at least 20 different answers to the question.
In an attempt to seek a consensus definition, we will turn to the independent high-level expert group on artificial intelligence, set up by the European Commission, who define Artificial Intelligence as follows:
“Artificial intelligence (AI) refers to systems that display intelligent behaviour by analysing their environment and taking actions – with some degree of autonomy – to achieve specific goals.”
Communication from the Commission to the European Parliament, the European Council, the Council, the European
Economic and Social Committee and the Committee of the Regions on Artificial Intelligence for Europe, Brussels,
25.4.2018 COM(2018) 237 final.
They go on to identify two main classes of AI systems.
Reasoning and Decision Making:
This group includes methods like organising knowledge, planning, scheduling, searching, and finding the best solutions. These techniques help AI systems think about the information they receive from sensors. To do this, the system first converts data into meaningful knowledge. This involves figuring out the best way to represent the knowledge. Once the knowledge is organised, the AI can use it to make decisions by applying rules, planning actions, searching through possible options, and choosing the best solution. Finally, the system decides what action to take. This reasoning and decision-making process is usually very complex and uses several different techniques together.
Learning:
This group includes methods like machine learning, neural networks, deep learning, and decision trees. These techniques let AI systems learn how to solve problems that are hard to define clearly or can’t be fixed with simple rules. Examples include understanding speech and language, recognising images, or predicting behaviour. These tasks might seem easy for people, but they are challenging for AI because AI doesn’t have common sense and struggles with messy or unstructured data. Machine learning helps in these situations by allowing the AI to learn from the data. Additionally, machine learning can be used for many other tasks beyond just understanding senses. It creates mathematical models that help the AI make decisions based on the data it receives.
If this seems a little difficult to understand at this stage- don’t worry! We’ll explore the nature of AI in more detail, as we go through, but we will focus on what it can, and as importantly, what it can’t do!
Reasoning and Decision Making Example: A Logistics Scheduling System 🗓️
A logistics scheduling system is a classic example of a reasoning and decision-making AI. Its core task is to create an efficient delivery plan for a fleet of vehicles. The AI doesn’t learn from past data in the same way a learning system would; instead, it uses a predefined set of rules and algorithms to find the best solution.
- Knowledge Representation: The system takes in structured data as its knowledge base. This includes the locations of all delivery addresses, the size and weight of packages, the availability and capacity of each vehicle, and traffic conditions.
- Planning and Searching: The AI uses complex algorithms, like optimisation and constraint satisfaction, to plan the best routes. It searches through millions of potential routes and schedules to find the one that minimises travel time, fuel consumption, or driver hours, while also satisfying all the constraints (e.g., a specific package must be delivered by 5 PM).
- Decision: The final output is a detailed, actionable plan: a route for each driver, a schedule of deliveries, and a list of specific actions to take. The AI’s decisions are logical and based on the programmed rules and the input data.
Learning Example: An Image Recognition System 📸
An image recognition system, such as one used by a social media platform to automatically tag faces in photos, is a prime example of a learning AI. This system is designed to solve a problem that is impossible to define with a simple set of rules. You can’t write a rule like, “if a person has two eyes, a nose, and a mouth, then it’s a face,” because faces vary so widely.
- Learning from Data: The AI is trained on a massive dataset of millions of labelled images. For example, it’s fed thousands of pictures with faces and thousands of pictures without them. Over time, the neural network learns to identify the complex patterns and features that constitute a face.
- Mathematical Model: Through a process called deep learning, the AI creates a sophisticated mathematical model. This model isn’t a list of rules; it’s a network of interconnected nodes that can “recognise” a face by analysing the patterns of pixels in a new image and comparing them to what it learned during training.
- Prediction: When you upload a new photo, the system runs it through the trained model. The model outputs a prediction, like, “There is a face in this area of the image, and it’s 98% likely to be person X.” The AI’s decision is based on the patterns it learned from the data, not on a human-coded set of rules.