24/01/2024

'What are the limits of AI?', Beauclair asks in lecture.

In 2023, the term “AI” was one of the most searched on the internet worldwide because Artificial Intelligence (AI) finally became dynamic enough to be popularly used. Chat GPT, Mid Journey, Bard and Copilot are just some of the AI tools that stood out that year. Roberto Beauclair, technologist and Information Technology manager at IMPA, gave a lecture at the institute on Wednesday (24), where he shed light on important issues surrounding Artificial Intelligence after its popularization.

Among the topics covered are the workings of new technologies, legal limits of use, copyright, and problems related to deep fakes and the misuse of audio and images involving deceased individuals, as well as inappropriate content.

What is Generative AI?

Generative AI is a branch of Artificial Intelligence technology used to generate content of various types, such as audio, images, videos, and text. These creations are formulated in response to user interaction and prompts .

These tools are trained to function as accurately as possible, considering modern techniques such as machine learning , deep learning , and diffusion models .

“ Machine learning works like a black box, so the way the AI is trained will dictate how it responds and interacts,” Beauclair explained.

He highlighted that Generative AI learns through Large Language Models (LLMs), and uses algorithms to process and understand the natural language used. These models are trained using the large volumes of data available on the internet, and precisely for this reason, AI ends up having biases, for example.

"If you ask a Generative AI to create an image of a person, it's more likely that image will be of a woman, because there are more images of women on the internet," said Beauclair.

Throughout the lecture, he used several examples of what these technologies can do, such as creating audio, converting languages in seconds, and changing faces in images and videos, all mostly for free and virtually instantly.

Regulation

With the popularization of these technologies, there is growing concern about the uses and responsibilities involved in content creation.

“The ethical concerns surrounding this issue are enormous, involving regulation, accountability, states, companies, and the internet, which is why regulation is extremely difficult. It is still far from being resolved,” Beauclair stated.

Meanwhile, books, paintings, campaigns, videos, music, and even entire commercials made by AI have already become popular on the internet, and this is just the beginning, considering that these technologies have been available for this type of use for less than 5 years.

Another concern involving AI is the market share these technologies will occupy, and how much humans will be affected by it. Beauclair raised this issue by questioning how the future will work with Artificial Intelligence taking up more and more space: "By 2027, 7 out of 10 companies will have adopted AI for some type of service within their workforce," said the technologist.

Furthermore, within this wave of AI-generated content, there is a lot of negative content, mainly involving artists and also people who have already died. The question that remains is: what are the limits to the use of these tools?

With the auditorium packed with students, researchers, and IMPA staff, Beauclair answered questions and warned that with the increasingly rapid evolution of AI, he will soon need to give another lecture.