25/11/2022

Machines that can read, see, and think.

Imagem criada por Matteo Moriconi em parceria com plataforma Midjourney

“A revolution is taking place in the machine learning process.” With this phrase, the lead researcher of Visgraf (Computer Graphics Laboratory of IMPA), Luiz Velho, introduces the theme of the workshop “From word to image: imagining futures with artificial intelligence”, which takes place this Saturday (26), from 10 am to 3 pm, at Casa Firjan, in Rio de Janeiro.

The workshop is part of the Possible Futures Festival , held between the 2Thursdays and 2Fridays, in the space that operates in the old mansion of the Guinle family, in the Botafogo neighborhood, in the southern zone of the capital of Rio de Janeiro. The event, which aims to generate a debate and reflection on the possibilities of futures that unfold with technological advances, will also include lectures, panels and other activities.

Alongside Velho will be Matteo Moriconi, a chemistry graduate from PUC-Rio and president of the Brazilian Association of Visual Technology, and Bernardo Alevato, a designer and researcher of virtual and augmented reality, new media, and artificial intelligence.

The paradigm shift that Velho refers to is the evolution occurring in the way media and visual effects are produced through artificial intelligence.

"New artificial intelligence models are able to combine two things that machines previously understood only separately: vision and language," stated the IMPA researcher.

In other words, a computer chip is capable of generating a highly customized image from the description of a few words. “A person comes in and says they want an image of a sunny day, with someone on the beach, in a realistic style, as if it were a photograph. At the same time, it’s possible to request this image as if it were a Picasso painting. The person gives a description of what they imagine, and the response is an image,” Velho explained.

“It’s as if the machine understands like a human being, capable of closing its eyes and describing a scene. It understands the structure of the image and the words and is able to make a connection,” added the IMPA researcher.

Imagem criada por Matteo Moriconi em parceria com plataforma Midjourney

At the workshop at Casa Firjan, Velho will demonstrate the mathematical concepts behind this cognitive learning of machines. Moriconi, in turn, will present platforms that are being used for visual production based on this new concept. Alevato will then project what the future holds for us as a result of this revolution.

“For artists, it’s crazy. It will be a profound innovation in the way media and entertainment are produced. Something that is already happening now,” said Velho. Furthermore, the model will have applications in various aspects of society and the productive sector, from medicine to services.

Velho also points out that the evolution of artificial intelligence brings ethical dilemmas that need to be considered in the machine learning process, as they can generate images based on a biased perspective, for example.

“Artificial intelligence is created from data generated by society and reflects its issues. Every new technology brings ethical and political debates, as happened, for example, with television. The sooner we talk about this, the better, so that innovations are more beneficial and effective,” said the leader of Visgraf.

Pesquisador do IMPA Luiz Velho

Another paradigm shift brought about by the new technology is the issue of authorship. To illustrate this point, Moriconi produced two sets of images: one inspired by Picasso; the other, more realistic. The authorship, however, belongs to the artist in partnership with the machine, which in this case is the Midjourney platform.

“The inspiration comes from the artist, but the execution is done in collaboration with the machine. In the Renaissance, for example, Da Vinci had an assistant. It's similar now, as the computer also becomes an assistant, although it cannot replace human creativity,” stated Luiz Velho.