How does AI work?

Artificial Intelligence works essentially having a very powerful capacity to process large amounts of data, very fast, beyond any human capability (in terms of amount and speed), and the ability to search for patterns within the data that provide the best answers.

So, how does Gen AI work?

It essentially has a very powerful capacity to gather and analyse large amount of data. By doing so over and over and receiving new data and retrofeeding the right answers to the database, it provides a better answer on the subject over time. In the past the same output was obtained mainly through decision trees and matrixes, and AI uses that too, along with new abilities.

An AI system differentiates itself from traditional software programs as it goes beyond a set of rules, a pre-established codes and rules, that provide the programs clear guidance on how to process the data available. Traditional programs do not retrofeed the database, based on previous correct solutions.

So, while traditional programmes do not create new outputs for the same problem, AI does.

Generative AI (Gen AI) looks for patterns to provide a solution. It enriches the outputs from the right solution provided, as well as new data, to learn complex patterns and improve the output over and over.

A good example is the Google search engine that searches the data on the web to look for the pattern that best matches the info requested, and as more and more people make the same enquiry, and use the same or similar output, it ranks what are the best answer, the most used ones, based on keywords and phrases, and becoming more precise over time.

Another good example is Netflix, how it provides suggestions to you that match or are pretty close to what you are normally looking to watch. It looks at the data of what you have watched over and over, and within the cluster of people with similar interaction with Netflix as you and based on such combination gives suggestions that look fit for you.

Another example is the evolution of autonomous cars. We today have cars that brake when they see a threat, that keep you between the lanes of a road, or keep you at a constant speed, but will react in the case of the car in front of you reducing the speed and adapt to keep you in a safe journey. Over time we will be just a passenger in the car.

We have been using matrix designed solution software, fed by quality data, since 1990. Chat GPT 3, the Gen AI solution, was what made people start to think differently about AI. It made AI accessible to the average person and provided a human-like interaction. We will talk about it as we progress in the module.

From Data to "Thinking", the journey to the Generative Artificial Intelligence

In the last decade, and long before Gen AI and ChatGPT, there has been a quiet revolution in the way we collect, store, organize and exchange data. It was this silent behind-the-scenes revolution that has prepared the “terrain”.

Let’s look next at the storage era, the improvements in data cleaning and reconciliation, and the growth of interoperability.

More data, more room: the storage era. In the past, to store great amounts of data was very complex and expensive. Companies and banks relied in physical servers with limited space and high costs. With the creation of the cloud storage and the significant reduction of the cost per gigabyte, everything has changed. Today, we can store "petabytes" of information in very economic, safe and scalable way.

Data cleaning and reconciliation: with a lot of data available, we have created another problem, "quality"! Duplicated data, incomplete data, or data in a bad format, have made difficult proper and conclusive analysis as well processes automation. However, tools and data cleaning processes have evolved, fixing mistakes, standardizing process, and removing inconsistencies.

Another important step in the evolution of data availability was the need for data reconciliation, and the challenges of connecting information on the same object from different sources. This could for example be information on the same customer from different product databases, and the ability to ensure one complete picture of the customer, with the latest complete and reliable information.

There arose then the need for systems that talk to each other, or interoperability. With cleaned and organised data, the next step was to ensure that different systems could talk to each other, could exchange information in an easy, fast and reliable way. The solution to this issue was the creation of APIs. APIs or Application Programme Interfaces are a real-time integrator of connected systems in a safe environment based on preset protocols. This enables banks, suppliers, fintechs, or any other entity to be digitally connected. Think of it as being almost like machines in a conversation!

With the availability of quality data in large volumes that is rapidly accessible, machine learning algorithms were trained to learn patterns and predict behaviour. More recently machine learning software can generate new content in text, images, and music, and even write software coding.

That is when we have the birth of the Generative AI, models that not only perform data, but “create” from them. How that was possible? We now have the quantity and quality of data to teach the models, we can have fast and safe access to the data, and the data is reliable and well organised.

Generative AI has become very popular, and is changing many industries as it progresses, but it was only possible with organised, connected and quality data. So, AI was not created one day by a robot writing texts or poems, but with bigger computer servers, large amounts of clean data and systems that were able to "talk" to each other and algorithms that are able to “create”.