Chapter 3
AI history

Forgotten for decades, AI returns to prominence in the years 2010

Frustrations almost led artificial intelligence to doom, but new tech gave it another shot

Raphael Hernandes

After being forgotten for decades because it did not live up to high expectations, AI (artificial intelligence) returned to prominence in the 2010s.

Now those high expectations have returned, but Swami Sivasubramanian, vice president of AI at AWS, doesn't see a "winter" for the area on the horizon.

"The big difference is that, in previous ages, we had no cases of AI application in the real world. We had proof of concept, but people couldn't scale it," he said.

The practical application of AI and its explosion were made possible by several factors, according to experts. One is the creation of cloud computing services.

With the modality, instead of having physical access to the computer, equipment from specialized companies is used at a distance, paying only for the momentary use of the machine.

"If I want to do an analysis, I can rent a very powerful machine for up to 14 dollars per hour. I solve a problem without buying an expensive computer," explained Renato Rocha Souza, a professor at FGV and a researcher at the Austrian Academy of Sciences.

These services can help to process the enormous volumes of data that are produced today everywhere (internet, cell phones, connected devices ...), and that was impossible in the past. Such information is the raw material to feed artificial intelligence systems.

In parallel, other factors have enabled AI to rise: stronger computers and new techniques (such as deep learning, a type of machine learning that helps to train AI).

Previously, one of the significant difficulties for an expert system (popular AI model in the 1970s) was that programmers had to manually create hundreds of rules to teach a computer a task. They needed to explain to the machine exactly what they wanted.

Today, the approach is inductive. With a lot of information available, you can make computers create these rules on their own by analyzing this data. They get just a little push, doing the rest themselves.

In the traditional example of identifying cat images from many photographs, it was necessary to explain to the computer (with programming) what the animal is. The task is complex and requires asking questions like "what defines a cat?" "What differentiates it from a dog?" and translating the definitions all into codes.

It is now possible to feed the AI ​​system with a large volume of photos of cats and let the machine itself detect a pattern there.

With the market heated up, one of the difficulties faced by the sector is the lack of specialized professionals. For this reason, leading cloud computing companies offer increasingly simple options for creating AI systems.

This is the case, for example, of AWS, IBM, and Microsoft. In their platforms, they started offering mechanisms that do not require programming directly to create AI models (mathematical formulas that provide instructions on how the system should behave).

In practice, it is as if the platforms had precast models for different uses. With that, they make the technology more accessible to people with some more advanced knowledge in IT, but who don't know much about that specific sector. The practice should contribute to a future with more and more AI.

Translated by Kiratiana Freelon

Read the article in the original language