“It is the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to using computers to understand human intelligence, but artificial intelligence does not have to confine itself to methods that are biologically possible”

- John McCarthy, one of the founding fathers of artificial intelligence, along with Alan Turing, Allen Newell, Herbert A. Simon, and Marvin Minsky.

Artificial intelligence is a simulation of human intelligence, programmed to think and learn like humans while not being confined to their capacities and capabilities. It combines the possibilities offered by computer science with robust datasets, mathematics, statistics, cognitive science, and many other disciplines, providing it with problem-solving capabilities.

Artificial intelligence systems are designed to learn from large amounts of data, identifying relationships and dependencies in that data. By doing so, they are able to make informed decisions or predictions based on new and unseen data. Allowing them to learn from historical data is often useful in fields such as finance, healthcare, and self-driving vehicles.

They utilise techniques and approaches such as machine learning, deep learning (a subset of machine learning), natural language processing, computer vision, robotics, expert systems, and much more to continuously learn and improve their performance over time as they receive new data. This makes them excellent for use in dynamic environments.

However, not all artificial intelligence systems are the same. There are different types of artificial intelligence systems:

Narrow AI:

• Also known as weak AI, refers to AI systems designed to perform specific tasks or solve specific problems.
• Examples of narrow AI systems include Siri, Alexa, Google Assistant, recommendation systems, and image recognition systems.

General AI:

• Also known as strong AI, refers to AI systems that have the ability to understand, learn, and apply “intelligence” to a wide variety of tasks and problems. They aim to mimic human intelligence and have the capability to perform any intellectual task that a human can do.
• Examples of general AI systems include GPT-3 by OpenAI and AlphaZero by DeepMind.

Super AI:

• AI systems that surpass human intelligence and capabilities in almost every aspect are known as super AI systems.
• If achieved, they would possess cognitive abilities far beyond those of humans.

Artificial intelligence systems use “machine learning”, a subset of AI that focuses on developing algorithms or models that allow them to learn and make decisions without being explicitly programmed to do so.

These machine-learning AI systems are first trained through a process called “training” or learning, where they learn from data to make decisions.

The steps involved in training are:

Data collection:

• Data, which represents the problem or task upon which the AI system is being trained to solve, is collected. The data collected involves inputs and their corresponding outputs.

Data Preprocessing:

• The collected data requires preprocessing to ensure its quality and suitability for training the AI system.
• It involves handling missing values, converting numerical values into a scale or range, and identifying and managing data inputs that deviate significantly from the majority of the data.
• Preprocessing ensures that the data being fed is consistent and usable and that it does not negatively affect training.

Algorithm Selection:

• Based on the problem or task and the type of data collected, a suitable machine learning algorithm or model is selected.
• Examples of machine learning models are linear regression, logistic regression, decision trees, random forests, neural networks, and support vector machines. More advanced machine learning models like Convolutional neural networks (CNNs) or Recurrent Neural networks (RNNs) can also be used.

Feature Engineering:

• This involves selecting and transforming inputs to enhance the performance of the machine learning model.
• This can be done using feature selection, dimensionality reductions, or even creating new features, depending on the problem or task that the model is being trained to solve.

Training the model:

• The machine learning model is then presented with labelled training data to learn patterns, relationships, and even dependencies between the input and output.

Loss function and optimisation:

• A “loss function” is used to measure the difference between the predicted output by the model and the true output.
• The model’s objective is to minimise this loss function, and to do so, it adjusts its internal parameters by using optimisation algorithms such as gradient descent to find the optimal value for those parameters.

Model evaluation:

• After training, the model’s performance is evaluated using a separate dataset known as testing data.
• The model is evaluated on metrics such as accuracy, precision, recall (which is how accurately it identified the positive cases from the data), and a F1 score (a harmonic mean of the precision and recall metrics).

Hyperparameter tuning:

• Machine learning models usually have “hyperparameters” that control their behaviour, such as learning rate, regularisation strength (the ability of a model to generalise unseen data), or the number of layers present in a neural network.
• They are tweaked and tuned until the best possible configuration is found for the given problem or task.
• This is done through techniques like grid search, random search, or even advanced techniques like Bayesian optimisation.

Deployment and Prediction:

• Once the model is trained and evaluated, it is deployed to make predictions on new, unseen data.

The effectiveness and accuracy of the predicted data by an artificial intelligence system depend upon:

• The quality of the data entered and the feature engineering performed
• The machine learning algorithm or model chosen
• The design of the machine learning algorithm
• The training of the model and hyperparameter tuning performed

Constant evaluation and improvement are needed to ensure the reliability and usefulness of the information provided by Artificial intelligence systems.