What is Artificial Intelligence (AI)?

When people read or hear about artificial intelligence (AI), they immediately think of ChatGPT, Gemini, or any Large Language Model (LLM). The reality behind it is quite different from the collective imagination.

The reality is that this “technology” originated back in the 1950s, more than 75 years ago; and not in the form of chatbots, but to determine whether a computer could develop intelligence similar to that of humans.

What is Artificial Intelligence (AI)?

There is no single, definitive definition, but we can say that:

[!info] Artificial Intelligence A mechanism or set of processes that simulate or emulate human thought, or actions that require it.

For example, have you ever wondered how Siri manages to understand and respond to our voice commands? Or how WhatsApp transcribes voice messages into text? Or even how TikTok or Instagram always recommend that cute kitten or puppy that always melts your heart?

Behind these small actions lies artificial intelligence.

Why Artificial Intelligence?

Some tasks are difficult, if not impossible, to accomplish with a simple algorithm, or might take an indefinite amount of time. AI is an ally of automation and speed—let’s say it could be your right-hand man. Be careful not to be fooled, though: even AI makes mistakes (hallucinations)! And more often than you might think.

Take spam or scam emails, for example. They’re annoying and potentially dangerous for a business! Well, a good AI model could prevent these dangerous hidden traps from being forwarded without you having to lift a finger. Gmail already has these systems built in, but how do they know if an email is potentially risky?

How does AI work?

Before discussing how they actually work, it’s important to talk about data. Artificial intelligence is a data-hungry beast; data is its lifeblood. To continue using Gmail (the most widely used email service) as an example, it has a massive database containing millions and millions of emails.

This data is collected, cleaned, and categorized based on its content. This process allows us to obtain a clean dataset to train an artificial intelligence model.

We can say that the learning process is similar to that of humans:

[!example] Example If email x contains word y, then that email is spam.

The model learns this concept and continues to analyze the composition of emails to determine which other emails might be dangerous, thereby building its own neural network.

There are several types of artificial intelligence that deal with this:

• Machine Learning (ML): a type of AI that learns from data to identify patterns and make predictions or decisions without being programmed. Imagine teaching a computer to recognize a cat by showing it thousands of images of cats; it will learn to recognize them on its own;
• Deep learning (DL): a subfield of ML that uses multiple layers of artificial neural networks to learn from data. As you can see, this field is inspired by the structure of the human brain and is widely used for complex tasks, such as speech recognition;
• Natural Language Processing (NLP): enables computers to understand, interpret, and generate human language. An example would be Siri, as mentioned above;
• Computer vision: This technology enables computers to “see” (not physically) images and videos of any kind. For example, Tesla uses it in its self-driving cars.

Types of artificial intelligence

Before discussing the types of AI, we need to identify the categories into which they are classified.

The two categories are:

• Capability-based AI;
• Functionality-based AI.

Capability-based AI

We have three different types of AI:

1. Artificial Narrow Intelligence (ANI): also known as weak AI, this refers to currently existing AI systems. This type of AI is designed to perform a single task or a limited set of tasks, such as ChatGPT and Gemini;
2. Artificial General Intelligence (AGI): also known as strong AI, this is a theoretical concept we aim to achieve, as this AI would reach the point of being as intelligent as a human (especially an expert in multiple fields, unlike ANI);
3. Artificial Super Intelligence (ASI): This is a purely theoretical form of AI in which an AI model could actually surpass human intelligence.

Functionality-based AI

We have four different types of AI:

1. Reactive: This refers to an AI with no memory, designed for specific tasks. In fact, it has no memory of previous actions but acts solely based on currently available data. For example, in a turn-based game like Go (which is particularly well-known in the field of AI);
2. Limited memory: unlike the previous type, it retains a record of past and present actions, but this memory is limited—meaning it is not long-term. It is used by common chatbots, such as ChatGPT and Gemini;
3. With theory of mind: this falls under the AGI model, so we would be referring to an AI capable of thinking and understanding human emotions. Currently, this is only a theoretical concept;
4. Conscious: in this case, we fall under the ASI model, in which an AI thinks and has emotions like humans, capable of interacting with them without being detected as such. Again, this is only a theoretical concept.

Conclusion

AI—primarily in the form of large language models (LLMs) and beyond—has just made a comeback. The field looks promising, but it raises serious concerns about the future of certain jobs. The strength of these AI systems lies in the vast amounts of data collected over the years: the easier it is to collect this data, the more easily that job or task can be replaced by these models.

That doesn’t take away from the fact that this field is fascinating, even though in this case I’ve only offered a tentative definition of AI and introduced some of the fields and techniques it encompasses.

This is the first post on this blog. I write about AI architecture, software engineering, and building production systems. If you have questions or want to discuss a project, get in touch.