What is AI?

AI stands for Artificial Intelligence and is an area of technological development. It creates systems capable of performing tasks that typically require human intelligence, such as learning and reasoning.

AI systems operate through algorithms and process vast quantities of data to function and learn. The algorithms that underpin AI are at the core of this technology. These algorithms define how machines process data, learn from experiences, and make decisions, thereby enabling the solving of increasingly complex problems.

An AI model is developed by feeding it with a lot of data (for example, pictures, text, or numbers) and training it to identify patterns and make predictions or decisions using the data.

Imagine this: You teach a child to recognize a cat by showing them many pictures of cats. The child learns to identify key features like whiskers, pointy ears, and a tail.
AI works similarly: We feed computers tons of data (like pictures, text, or numbers). They use this data to identify patterns and learn to make predictions or decisions.

The core skill of AI lies in differentiating between objects and grouping similar objects. The more you advance this base, the more complex AI becomes.

Here are some examples of AI in action:

Self-driving cars: These cars use AI to "see" the road, understand traffic signals, and make driving decisions.
Voice assistants (like Siri or Alexa): They use AI to understand your voice commands and respond to your requests.
Recommendation systems (like Netflix or Spotify): They use AI to suggest movies or music you might enjoy based on your past choices.
Language translation (like DeepL or Google Translate): Through machine learning, text or speech is analyzed and converted from one language to another.
Generative AI (like ChatGPT or Gemini): They use AI to create new content, such as text, images, or audio, by learning patterns from existing data and responding to your prompts.

In essence, AI is about creating intelligent machines that can perform tasks that typically require human intelligence.

There are many approaches to developing AI, such as machine learning, symbolic reasoning, evolutionary algorithms, and Bayesian networks. The success of machine learning (ML) has made it so that whenever people discuss AI they are most likely referring to this approach. Deep learning (DL) is a popular form of machine learning. However, while prevalent, machine learning is only one approach to developing AI.

Machine Learning

The basic premise of machine learning is to build algorithms that can receive input data and use statistical analysis to give some output. Those algorithms can learn from the data by recognizing patterns, and make predictions on new data. Put simply, the machine learns from data, rather than knowledge being manually put into it.

Common examples of ML applications include:

Classification tasks
Loan prediction
Anomaly detection (spam messages, suspicious credit card transactions)
Weather patterns
Gas prices
Clustering tasks
Customers grouping

Machine Learning can be divided into two different learning types:

Supervised and unsupervised. The main difference is the need for labeled training data.

For classification and regression tasks, the machine uses supervised learning, meaning that both data and labels for classification of the data are provided, telling the machine how to classify things. The machine then compares and sorts the data according to these pre-set labels.

For clustering tasks, unsupervised learning is used. In this case, data is provided but there are no labels.

Instead, the machine clusters the data by itself.

Deep Learning

Deep learning is a branch of machine learning which is completely based on artificial neural networks. These networks are modeled on the structure of the human brain. Between the input and output layers are various hidden layers that break down and transform the data.

A neural network has 3 types of layers:

Input layer: Such as age, height, image pixels, etc.

Hidden layers: This is where the power lies and data is processed and transformed. These layers can get highly complex.

Output layer: The output we want to predict. (Will the person get the loan? Is it a banana or an apple in the image?)

When training neural networks, the process typically begins with a large dataset relevant to the specific problem. The neural network then learns from this data. This is done via the layers of interconnected neurons which are assigned weights. These weights determine how strong a connection between each neuron is and how much influence the input from one neuron has on the output of another.

The crucial step in this training process is the comparison between the output generated by the network and the actual or expected output corresponding to the input sample. This comparison reveals the errors between what the network produces and what we need it to produce. To minimize these discrepancies, the weight of the neurons is adjusted. The objective is to fine-tune the network in a way that brings its output closer to the desired results. This weight adjustment is the essence of the training process and is carried out repeatedly as the network encounters more and more data.

Through this process, the neural network becomes increasingly capable of making accurate predictions or classifications for new, unseen data. This way, neural networks can become highly accurate and serve numerous applications. However, they are very resource intensive through their need for large datasets and great computational power. Additionally, the theoretical explanations of how exactly the hidden layers work and interact, are currently still somewhat weak. Due to the complexity of the neural networks, it can be hard to determine how exactly an AI application arrives at certain outputs.

To summarize, AI isn’t magic. Machine learning and deep learning primarily rely on two things:

Data: AI learns from data. In the best case, it learns the information that the data possesses.

You don’t have enough data? The AI won’t work well.
You have unclean data? The AI won’t understand it’s not clean.
You have wrong data? Your AI program will learn that too!

Computer power: The stronger the computer, the better. However, high computer power is very expensive.

References

AI4AL - Methodological Guide https://www.ai4al.eu/wp-content/uploads/2023/11/AI4AL-Methodological-Guide.pdf

Google Cloud: What Is Artificial Intelligence (AI)? https://cloud.google.com/learn/what-is-artificial-intelligence