Latin American companies begin embracing new AI tools

The hype surrounding content-creation artificial intelligence (AI), embodied in AI chatbot ChatGPT, has attracted significant attention from Latin America's private sector.

Through March of this year, Latin America is estimated to have contributed over 300mn accesses to ChatGPT, originating around 10% of visits in the period, according to data compiled by Similarweb. Brazil was the Latin American country generating the most access, followed by Colombia and Mexico.

Rodrigo Scotti, co-founder of Brazilian AI association Abria, believes AI solutions will continue growing and developing, though some experts warn investors and corporates to focus on practical use cases that can have a meaningful impact on the local populations and their bottom line.

“Advanced Generative AI is a tool completely unlike any other. It can’t be compared with other things. It’s evolving and will continue to evolve. People perceive things in a linear way, but technology develops exponentially,” Scotti tells BNamericas.

In any case, some specialists consider that these new tools raise the debate to another level, including perspectives on the future cognitive development of individuals in the face of mass outsourcing of creative activities and jobs to machines.

Scotti, who is also CEO of the AI company Nama, acknowledges that issues such as intellectual property, legal responsibility for content and data protection might have to evolve at the same time. However, he sees the current public debate as biased and generally uninformative, saying that AI can close many productivity gaps in Brazil. 

“Drawing a parallel with traditional programming, it's not possible to be binary, good guy versus bad guy. There's still the narrative of job losses – that we're all going to live in a dystopian future. Let's try to understand what's going on," he says.

Scotti’s Nama has seen double-digit growth in corporate inquiries about AI since December. Most of these have been consultations, but the executive said that his firm, which has been working on AI solutions based on Large Language Models (LLM) since 2019, has a few projects lined up.

LLMs are foundational machine learning models that use deep learning algorithms to process and understand natural human language.

Management software companies are also assessing solutions for companies based on technologies such as ChatGPT. 

“The tests being done are with representative clients from different industries. There’s one in particular in Latin America, based in Argentina,” Claudia Boeri, CEO of SAP Latin America South, tells BNamericas. "What SAP intends is to be able to generate replicable use cases for these industries."

Internally, the German software giant uses AI in its software creation laboratories to speed up and shorten programming times and employs ChatGPT to generate lines of code.

Boeri said SAP typically tests the software for a period of one year with customers before releasing it to the market. These processes can be automated with AI in order to shorten these times.

SAP has also started discussions with customers to integrate ChatGPT into their enterprise resource planning (ERP) software. The cases are still very incipient and remain confidential for the moment.

The management software company Benner will invest 13mn reais (US$2.6mn) this year in product innovation, notably research and development of optimized generative AI tools.

“This month [April], we will launch the first solutions developed with ChatGPT for the legal and human resources verticals. And, throughout the first half of 2023, we will also make other applications available for the other verticals served by Benner, such as health, for example,” the company's VP of Innovation and Technology, Marcelo Murilo, said in a statement.

Programmers, developers and analysts from Benner's innovation team have been striving to create new tools that will be made available to the entire market, including firms not running Benner's ERP software, the company said.

A recent survey by Deloitte found that seven out of 10 Brazilian companies plan to invest in AI this year. The data was collected in February-March among 501 companies with total revenue accounting for 21% of Brazil's GDP.

In general, the main front for testing and using advanced AI and natural language processing (NLP) has been customer service, particularly via chatbots. But that is rapidly changing. 

Brazilian state-run oil major Petrobras is testing advanced AI on its supply and procurement operations, BNamericas has learned. These tests are taking place along with the development of supercomputers in partnership with Nvidia and Supermicro.

DATA INFRASTRUCTURE

The high volume and consumption of data generated by advanced AI models will also impact datacenters in terms of available space, connectivity and energy consumption.

The fast response time of these AI systems requires increasingly lower latencies, which implies better connectivity, while the growing capacity of servers will mean more power consumption.

“These clusters of GPUs, which are superservers that serve artificial intelligence, are going to consume a lot more watts and power. For the datacenter industry, it will be a challenge," Alexandre Alves, Latin America manager of integrated solutions at Vertiv, tells BNamericas.

Some estimates are that datacenters will consume 10% of the world's energy in the next five to seven years. 

"Historically speaking, we had the internet, cloud computing, the internet of things, there's 5G and, now, artificial intelligence. Bitcoin and blockchain also brought excessive energy consumption. So we’re never in a comfort zone,” says Fernando Ribeiro, telecom systems coordinator at Odata.

“Our real concern right now is connectivity, low latency. In terms of energy, we’re monitoring this process.” 

The adequate IT infrastructure to support advanced generative AI may prove too costly to maintain on-premise versus having it in a colocation datacenter, adds Ribeiro.

To cope with that additional demand, datacenter companies will need to adjust. "The data consumption architecture will have to be rethought," says Vertiv's Alves.

Training and deploying AI models in datacenters also involves significant water consumption for generating electricity to power servers and cool them down.

According to the study Making AI Less Thirsty: Uncovering and Addressing the Secret Water Footprint of AI Models, produced by the University of Colorado Riverside and the University of Texas, ChatGPT consumes the equivalent of a 500ml bottle of water for every conversation involving 20-50 questions and answers.

It also found that training GPT-3 in Microsoft's state-of-the-art US datacenters directly consumes 700,000 liters of fresh water (enough for producing 370 BMW cars or 320 Tesla electric vehicles). 

"As far as I know, there is still no language model operating in Latin America, physically speaking," says Alves, referring to the location of advanced AI data.

"But If these systems become the reality of our day-by-day searches and operations, we’re going to have to build a lot more hyperscale.” 

With additional reporting by Leticia Pautasio