Study: AI will increase global GDP by BRL 7 trillion by 2030 – 06/29/2023 – Tech

The increase in programmer productivity generated by artificial intelligence (AI) will boost the Global GDP (Gross Domestic Product) by US$ 1.5 trillion (R$ 7.2 trillion) by 2030, according to projections made by Harvard Business School professor Marco Iansiti and GitHub Chief Executive Thomas Dohmke.

The study estimates that the number of developers will rise from the current 27 million programmers around the world to 45 million in seven years. These professionals working in different sectors would have a 30% gain in productivity by 2030, which would justify the boost in the global economy.

Dohmke released the study at the Collision technology conference, held in Canada from June 26-29. “I’m here to show you the data that proves artificial intelligence will not replace programmers.” A year earlier, at the same event, he had released the assistant programming assistant GitHub Copilot to the public.

AI programming assistants like GitHub Copilot and Amazon CodeWhisperer allow people with less technical knowledge to develop working programs. It is possible, for example, to ask the AI ​​to create a form with instructions in Portuguese. These tools also detect errors and complete code.

To arrive at the results presented, GitHub and Professor Marco Iansiti used the behavior of 934,533 Copilot adept programmers as a parameter. The technology was developed in partnership with the creator of ChatGPT, OpenAI.

Analysis of GitHub data shows that less efficient programmers tend to gain more efficiency, as they accept more suggestions from the algorithm, than programming veterans. The study is based on the premise that the greater the synergy between AI and humans, the greater the gains, based on other studies by the company.

The algorithm saves a lot of time with code snippets that are repeated over and over again throughout the program or searching for sources and documentation, according to GitHub’s director of developer policy Mark Linksvayer. “This allows programmers to spend more time on creative activities, such as understanding what the customer really wants.”

For Linksvayer, technology will help meet the demand for talent in the area of ​​software development. A survey by technology education startup Digital House released this month shows that Brazil has a demand for 70,000 programming professionals per year, while training 45,000 professionals in the period. The scenario is repeated in other countries to different degrees, according to the GitHub report.

The study says that the new programmers will be functional. “In programming, being efficient is delivering programs that work well on time,” says Linksvayer.

Asked by Sheet if this increase in productivity could not lead to a reduction in jobs or a drop in wages in the area, the executive said that GitHub expects Copilot to be a complement, not a replacement.

Technology professionals are among the most sought after and well paid in Brazil. Computer engineers earn an average of R$ 13,208 upon admission, according to a survey by Firjan. This profession also had fewer layoffs than the average since 2020, as shown by Sheet.

Technology is not a platform that allows the development of programs without code, says Linksvayer, “Everything is done in a programming language and with human supervision. What changes is time savings in repetitive tasks.”

Copilot, like its competitor Code Whispererer, was trained with programming codes available on the internet. Both programs report excerpts that can be denounced for plagiarism. Still, Linksvayer says human oversight is key.

In the study, Dohmke, and Iansiti argue that the productivity gain resulting from the use of Copilot should increase in the coming years.

Screen data from GitHub Copilot users shows that the number of accepted suggestions grew from 28.9% in the first quarter of usage observed by the researchers to 32.1% in the second.

Users begin to accept more suggestions as they understand how the generative AI works and become less suspicious of its capabilities, according to Linksvayer. Interaction with the platform also provides training data for the algorithm, which gradually adapts to the developer’s needs.

Data from GitHub shows that 46% of code produced with Copilot assistance is done by AI —the number rises to more than 60% success in the case of Java, the world’s most popular programming language.

In an interview with Sheet During a visit to Rio, Dohmke stated, however, that it is impossible to ask Copilot to build complex platforms like Facebook or GitHub, which works as a digital library of programs.

“To make programs, you have to break the project into small challenges. That system of thinking comes from the developer, who needs to have a degree in computer science, computer engineering or programming experience,” says Dohmke.

The programming assistant AI works in any language — including Portuguese —, according to GitHub. “It’s a way to go beyond the English-speaking world that the software development market is now centered on,” said the executive in Canada.